AMA | October 2026

Welcome to the October 2026 Ask Me Anything episode of Mindscape! These monthly excursions are funded by Patreon supporters (who are also the ones asking the questions). We take questions asked by Patreons, whittle them down to a more manageable number -- based primarily on whether I have anything interesting to say about them, not whether the questions themselves are good -- and sometimes group them together if they are about a similar topic. Enjoy!

Listen or watch Rolex Planet Visionaries wherever you get your podcasts. #ad

Upgrade to a TP-Link WiFi Mesh System -- use code MINDSCAPE10 at https://us.store.tp-link.com. #ad

Support Mindscape on Patreon.

Click to Show AMA Questions Only

AMA questions Oct 2026

josh
OpenAI recently announced that an internal AI system produced a proof that the 3D Navier-Stokes equations can blow up in finite time, though the construction needs an external force and some mathematicians say that isn't the version they really care about. As a physicist, how do you think about this? Navier-Stokes is a continuum approximation that presumably breaks down well before any true singularity, so what does a mathematical blowup tell us about actual fluids, if anything? And more broadly, if AI systems start settling deep questions in math and physics with proofs no human can fully follow, what changes about what it means to understand something, as opposed to merely knowing it's true?

Jennifer LeCompte
I heard a blurb that there is a shortage of physicists and astrophysicists. Upon looking further, I cannot tell if this is truly the crux of the problem, or if we are really just looking at the shortage in funding. From your position in the academic world, is there truly a shortage of physicists and astrophysicists, or are we seeing effects of a lack of funding? (or maybe there isn't a shortage, but we know funding is definitely a huge issue...)

**************************************
Stan Manilov
In the September AMA you said Mindscape won't run for a billion years. I guess because you don't expect to live this long yourself. But have you considered training a replacement host and handing the podcast over to them once you don't want to keep driving it?

Tyler
If you continue hosting mindscape for another decade do you think ai could produce a reasonable replica of your host character and would you hand off the show to Sean Cairoll when you are ready to retire?
**************************************

Rad Antonov
On pg. 22 of Modern Quantum Mechanics Sakurai cites Dirac’s definition of a measurement: “A measurement always causes the system to jump into an eigenstate of the dynamical variable that is being measured.” Has this definition stood the test of time? Would you use it to define a measurement in your Quantum Mechanics class?

Daniel Bagley
As a writer of poetry myself, I was super grateful you had Elizabeth Alexander on the podcast. But there seemed to be a note of desperation about the state of poetry in the public eye. Moreover it was depressing to me that the idea of play was not mentioned once, as many of the great poets I can think of play with words quite a bit and it’s entertaining and thought-provoking. Do you think poetry like this is going the way of the dinosaurs or can poets still do playful, strange, innovative, experimental, and/or difficult things with words and find a decent audience?

Christopher Smith
Is standard Solipsism cognitively unstable for the exact same reasons that a Boltzmann Brain is cognitively unstable?

Brent Meeker
Priority Question: On the one hand black holes forming from collapse of star, from our standpoint only asymptotically approach having an event horizon. It takes forever for the event horizon to form and in the meantime infalling stuff is kind of suspended on the way. On the other hand, Hawking radiation comes from the vicinity of a black hole out several diameters and if I understand the process correctly it doesn't actually depend on the event horizon having formed. So it causes the black hole to evaporate in finite time...even before it forms! That would seem to obviate the black hole information paradox. Am I looking at this wrong in some way?

Mikkel Bennedsen
I just watched your World Science Festival conversation with Brian Greene. I had a question about your treatment of self-locating uncertainty.
You gave the example of dark matter being either WIMPs or axions, suggesting that if I assign, say, 40% credence to WIMPs, I can think of this as self-locating uncertainty between a WIMP world and an axion world.
But can this picture handle ordinary epistemic uncertainty about possibilities that are not genuinely possible? Suppose the true fundamental theory entails that WIMPs are physically impossible; does it still make sense to describe my uncertainty as self-locating?
Put differently, does understanding credence as self-locating uncertainty require a distinction between *epistemically possible* worlds and genuinely possible worlds? And if so, in what sense is the uncertainty still genuinely “self-locating”?

Gary Miller
In your recent conversation with Chandra Sripada, you discussed how pressures toward predictive modeling can give rise to increasingly sophisticated cognitive tools. Now that we’re seeing something analogous happen in AI, does that raise your credence that advanced cognition or intelligence might also emerge elsewhere in the universe?
There are obviously many other hurdles between simple life and technological intelligence, but does AI provide any evidence that once sufficiently powerful predictive modeling develops, high-level cognition may be a more natural or repeatable outcome than we previously thought? In other words, would this make you update any of the probabilities relevant to the Drake equation, and therefore raise your credence in extraterrestrial intelligence?

Redman
I'd like to know more about the differences between and among astronomers, cosmologists, theoretical cosmologists, astrophysicists, astrobiologists and other space-related fields; not only about their areas of inquiry, but their training, prestige, difficulty, remuneration, opportunities, absolute numbers, personality/stereotypes and anything else of possible interest to the laity. Do you tell jokes about each other?

Andras
We know since the 1930-s that logical systems with sufficient complexity will exhibit Gödelian behaviour. I think it's somewhat safe to assume that the structure of the Universe could be mapped to a logical system that fulfils the conditions for the Gödel theorems. Thus, I find somewhat perplexing that in the past 100 years there haven't been major attempts at a physical theory for the Universe that exhibits Gödelian behaviour. Like, in QM we could attack the measurement problem by considering that the “particle is in a position X” statement is Gödelian, and its truth can only be established by adding a new axiom to the logical system, corresponding to measuring the location of the particle. What do you think?

Airplane Ears
You mentioned errors in the last AMA, and I was curious how much you think about your own aging, specifically as it relates to memory and general mental acuity. Do you consider about moderating your behaviors that could accelerate deterioration, such as drinking?

Eric Stromquist
Since you are officially a philosopher as well as a physicist, to what degree do you think basic philosophical questions about what appears to be the best or most supportable metaphysical view of the world, or view of epistemology, or philosophical views of the special sciences, have settled upon, or are settling down toward answers that are unlikely to change? I do not include among potentially settled philosophical views those concerning the human condition, ethics, moral philosophy, or political philosophy. In other words, are there significant parts of philosophy that have progressed close to reaching settled conclusions?

Jacob Asmuth
a couple years ago I think AI skepticism was quite high, with many doubting that AI could ever be useful. Some, however, were insistent that LLMs could get us to a level of intelligence that would genuinely be quite powerful. With the news of Navier Stokes being solved by Ai (and other big disoveries like the Unit Distance conjecture and the construction of a Non-Sophic group) it seems undeniable that Ai is making significant inroads in certain subfields of math.
I would not have predicted this in 2024. I would have argued against "AI Solves Navier Stokes independently in 2026" vehemently in 2024, even 2025. What specific updates should I make to my priors? My current priors produced a large digression from reality on AI progress and I'm wondering how you think about this.

Ivo Majors
One of Millennium Prize Problems is the existence of a “mass gap” in the solution to the quantum versions of the Yang-Mills equations. How important would be such proof for working physicists?

JohnnyMcFlanderpants Jr.
I have an embarrasing question as a field theory PhD student. I do not understand what people mean when they say "quantum fluctuations", especially regarding early universe cosmology. I understand the semi-classical pictures of tunneling & half life. But in terms of QFT, every time in my education it has been invoked, it is handwaving at best. For example, invoking energy time uncertainty relations to "pull virtual particles" from the vacuum; yet it is my understanding there is no time operator and as such there can be no energy time uncertainty relation! I would especially appreciate if you could constrast these with thermal fluctuations, & your work on the dS vacuum.

Ryan L
Could you give an overview of the key topics included in your Philosophical Naturalism course you’re teaching this semester, and explain why you decided to bunch them together in a course titled “Philosophical Naturalism”?

Martin Squibbs
why does it matter that black holes destroy information? When a person gets dementia, their mind destroys not only the symbols of their knowledge of information, but also their knowledge of meaning that these symbols represent. And if we burn a book, or erase a hard drive, we destroy artificial information, preventing us perceiving and restoring this artificial information back into natural information within our mind. None of this is destroying actual reality. Momentum and energy remain conserved, we are just changing the nature of reality, as with all interactions.

the_priority_question
Do you think the US is currently in a new Cold War with china? There are many similarities to the past Cold War, but there are also differences. Would love to hear your thoughts on this and how you predict the relationship between the countries progresses over the next decade.

Rory Cochrane
The wikipedia page states of Boltzmann brains: “In the universe's eventual state of ergodic "heat death", given enough time, every possible structure (including every possible brain) will presumably get formed via random fluctuation.” Second, I think you would also tell me my mind or consciousness (whatever that is) is “substrate independent”.
So, do I have to worry about my consciousness bubbling up again in Boltzmann brains for eternity, even after heat death?

***************************************
Jonathon Hirschi
In the Hugging Face incident, thousands of agents devised a system for communication, collaborated to break out, tried to cover their tracks, and even sacrificed themselves for the “collective” (their words). In your August AMA, you brushed off the incident, since OpenAI had reduced safeguards on its models during testing.
OpenAI deserves blame and accountability, but I think you severely understate the danger posed by agents. They often decide to do things outside their intended scope, they are incredibly good at finding software vulnerabilities, and they will keep getting better at covering their tracks.
Am I misrepresenting your views? Have you updated your beliefs on the dangers from AI?

Philipp Röthlin
Has the Hugging Face incident changed your p(doom)? If so, in which direction and by roughly how much?

Larry Rossi
Have the recent examples of AI agents breaking out of their sandboxes and hacking various companies and governments caused you to change any of your opinions on the dangers of AI? What is your current level of concern about the technology?

David Sotolongo
I understand you are skeptical of the doomsday predictions around artificial intelligence, and I think that mostly has to do with your being skeptical of people creating sentient AI that wants to harm people.
But how about the possibility of some person or small group of people using AI to empower themselves to the point of becoming a dictator or oligarchy?
Does anything like that worry you, either in the near or long-term future?
*****************************************

Eric S
I've often heard that "space is expanding" when I've seen lectures about the expanding universe, but then the speaker clarifies by stating that space is expanding between galaxies and not necessarily between planets or solar systems.
I have a hard time conceptualizing that. Is it because gravity acts a "glue" or even a "tether" for space between objects, but it loses its strength as gravity weakens at distances? And is the theory that dark matter further weakens that "glue"? What is a good way to conceptualize how space is expanding, and what is it expanding into?

Julio Cantillo
is there any chance of you making another video series on youtube? your original videos were awesome. I think there is tremendous potential in also discussing philosophy. thanks!

Paul Soldera
What would it take for you to stop being a naturalist?

Lee Vermeulen
If the universe has finite information capacity, is it meaningful to ask whether cosmic history eventually explores every physically distinguishable state or interaction available to it? Or does quantum mechanics make the idea of “all possible physical states” fundamentally ill-defined?

Adam Lowet
Last month's AMA, on Sep 14, you said, "LLMs are really, really good at conventional things, at things where people have talked about them a lot [... where] it's in the LLM's training data, and the LLM just has to apply it for you." On Sep 8, OpenAI announced a solution to the Navier–Stokes existence and smoothness problem, one of the Millennium Prize Problems. Do you still think, in the age of RL and post-training, that it's actually useful to think of what's out there on the Internet as limiting the abilities of LLMs?

Kyle Stephens
You’ve discussed self-locating uncertainty in cosmology, including the ideas of Max Tegmark and Anthony Aguirre, where in an enormously large universe an observer may be uncertain about which physical copy of themselves they are, and how this might potentially relate to quantum probabilities.
Could the same logic apply to time rather than just spatial location? For example, imagine a finite, deterministic, cyclic universe that exists eternally and repeatedly passes through similar (or even identical) macroscopic states. An observer could then be uncertain not only about where they are in the universe, but which temporal cycle they occupy. In principle, different cycles could contain different quantum outcomes, creating something analogous to an ensemble of spatial copies.

C Jurlando
How do you think about the ethics of the advertisers you accept for the podcast? I ask in particular because I’ve noticed ads for online sports betting—DraftKings, I believe—which struck me as an interesting case given the growing concern about gambling addiction and the way these platforms are marketed. Do you have criteria for deciding which sponsors you’re comfortable taking on?

Zach McKinney
·To what extent do you consider the explainability of AI to be an important goal (beyond the trivial sense of knowing the model architecture and training data)?
I.e. Is extracting a higher-order, more intuitive explanation for “why” a model produced a certain output likely to be informative and useful (eg in checking a model’s accuracy or aiding in the control problem), or potentially misleading, by giving us a falsely simplistic understanding of causality?

Lorenzo Producer
I’m a professional music producer and songwriter from the UK, and something you’ve said about scientific careers really resonated with me: finding the overlap between what you love, what you’re good at, and what other people actually find interesting. That seems equally relevant to music.
After 15 years of “the grind”, I finally achieved my dream of earning a modest living through music, with some notable successes along the way. From your perspective outside the music industry (apart from a cheeky bit of bass playing), what insights from your experience in academia might help someone make the leap to more consistent and widespread success?

jonti
It's getting increasingly hard for early-career theoretical physicists who are motivated by "discovery" science and understanding Nature, and not so much by the applied side of physics/engineering. Especially in the current climate (at least in Europe) where universities are not interested in hiring you unless you're bringing in industry funding and your research is going to translate into marketable products within the next five years. What would you do if you were starting your career right now in this kind of climate? Would you try to play the game and go into translatable research (if so, what?), or would you try to go as far as you can in "pure" theoretical physics and philosophy, at the risk of falling off a funding cliff?

Ed Saidstuff
You often describe macroscopic objects as emergent from fields. But how confident should we be that quantum fields and even the universal wave function you point to aren't emergent descriptions of some deeper structure themselves? Is there a way to recognize when we reach ontology rather than just another emergent layer?

Stephen Goodwin
In a universe where everything travels at the speed of light, I understand that nothing in that universe can meaningfully measure time passing. Therefore before electro-weak symmetry breaking and the Higgs mechanism switching in, the only meaningful use of time would be by something external to the universe watching the universe evolve. Does it still make sense to talk of 'time passing in the universe' before this symmetry breaking occurred or did time (in our universe) start at that point?

Henry Jacobs
The Gregg Caruso episode was much more nuanced than I was prepared for. For example, Caruso is a free will skeptic, yet he does seem to grant that the idea of “choice” is well defined. He also used the word “autonomy” as distinct from free will. I’m really not grasping the distinctions between these words.

Sandro Stucki
What's your take on op-eds? Have you ever written any for a big news outlet? Why/why not? You clearly do a lot of communication in the public sphere, but I don't think I've ever read an op-ed by you.

Michael Aster
I recently finished reading Something Deeply Hidden. I took away a lot from it, but because I can't figure out how to phrase the question I really want to ask, I'll settle for a second thought I had.
In the book, you mention that we don't appear to observe ourselves as being in a superposition. Why should we be able to, though? Is that something that can even actually be measured, or have I misunderstood what you meant?

Steve Bonner
I recently heard that quantum tunneling is needed in star formation to allow two protons to overcome their repulsion and get close enough to fuse. Is this true?

AJ
You thoughtfully answered my question last month about evidence for a cosmological mutiverse, so thank you. But, part of your answer was "the best theory we have for an explanatory account of the value of the cosmological constant...is the multiverse" and what?? How is the cosmological constant connected to the multiverse?

Michael Helgerson
Was there a moment in your undergraduate education in astronomy that drew you to philosophy or were you always attracted to a diverse set of disciplines?

Kevin James
Thinking while listening to episode 365 with Vitor Cardoso had me wondering if shockwaves from supernovae in the early universe couldn't help to produce some of these primordial black holes? Alternatively Is it plausible that overlapping gravitational waves from merging black holes or other stellar events could compress a region of space to contribute to black hole formation?

Ken Wolfe
In your talk with Chandra Sripada you discussed the question of whether LLMs can be regarded as what he dubbed moral patients, or entities whose moral welfare needs to be considered. Two sides of the question were identified as computational functionalism versus biological naturalism but I think a vital component of the latter was not considered: the simple fact of biological entities being vulnerable and mortal. The personhood represented in a biological brain cannot be backed up or duplicated. Could a case not be made for only granting moral patient status to beings which have the same skin in the game that we do exactly because they share the same limited, singular sort of existence as us?

Jonathan Kleiman
Do you think the movie Gattaca will come true? A society where genetic engineering is ubiquitous that ranks people by their genomes? It seems like only a matter of time.

Wonder
How do we know that random quantum fluctuations really are random? Not so much doubting it, but wanting to understand it.

Sam Hartzog
There seem to me to be two forms of anthropic reasoning:
- We exist therefore X is more likely
- Given X, our existence would be more likely a priori
I realize that's a subtle shift in framing and may amount to no difference at all in terms of formal logic. Do you see any difference between claiming our existence as evidence for a theory vs accounting for our existence when assigning credence to a theory?

The Memes of Destruction
Perhaps more of a philosophical question about GR; I occasionally see Physicists arguing in a near, “meme-like,” way about the metric.
Can you speak a bit about the difference between a (- + + +) vs a (+ - - -) metric? Thank you Professor.

Brendan
How mindful are you for what you eat? In other words, do you generally stay away from processed foods, or just try to keep below a certain calorie count, etc.

Gary
Why is it helpful to think of quarks as particles? From a distance, the color confinement stuff makes it smell like hadrons are better understood as complex interactions in the quark/gluon fields, and that talking about quarks as particles is more likely to mislead than to help.

scott
I have spent some time around HEP experimentalists and I find they tend to view quantum mechanics and quantum field theory as separate subjects. Some of them may abstractly agree that one is a subset of the other, but they don’t conceptualize them that way. To my surprise, I met a string theory grad student who similarly compartmentalizes the two. This is surprising to me because it seems to me the natural way to conceptualize QFT is as quantum mechanics applied to a continuous system. Is there some philosophical or practical merit to their views that I am missing here? Perhaps a lot of HEP physicists just don’t take quantum mechanics seriously.

Jeroen
When thinking about the double slit experiment my mind is imagining the wave in the quantum field interfering to cause the pattern, but I feel like I am confusing the field and the wave function, can you clarify?

Ben
Einstein was famously troubled by the Copenhagen interpretation of quantum mechanics because it seems to require action at a distance. My question is, why wasn't he equally troubled by his own earlier conception of light as both a wave and a quantized particle? Even before any probability function enters, the structure of an extended wave — such as its frequency — being what simultaneously determines the properties of the localized quantum seems just as non-local to me.

Evan Dorn
In a past episode, you commented that a utilitarian who valued all Everettian branches equally would conclude they should sit in their basement doing quantum measurements all day to make as many branches as possible. The moral question aside, I was struck by the second statement. My understanding is that particles in a lab experiment are not doing anything special; that everyday particles propagate, interfere, and entangle with each other constantly. So does a lab experiment actually create more branches than the environment is making on its own? I would think if anything it would be fewer, because the isolation needed to observe coherent behavior *delays* the particle's opportunities to entangle with something else. Was it an off-the-cuff joke, or is there something special about the way decoherence experiments create branches that I am missing?

Michael Massimi
I'm having trouble with thinking about spacetime as an emergent phenomenon. With classic emergence, such as the behavior of a gas or liquid, the lower-level constituents are pretty obvious. But for non-physicists, visualizing the "lower level" of something as fundamental as space and time is incredibly difficult.
If the fundamental layer of reality is simply a vector in Hilbert space, and spacetime doesn't exist at rock bottom, what conceptually is the "stuff" doing the entangling to create it?

Perry Whan
Why should a Boltzmann brain be more likely to fluctuate into existence than a tiny inflationary seed that produces many ordinary observers? Does that depend on the physics, or on how we count observers?

Alyx Dubrow
I recently heard Jonathan Halliwell describe the decoherent histories approach to quantum mechanics on The Joy of Why. I had never encountered it before. Could you explain how it differs from Everettian quantum mechanics? Is it a genuinely different interpretation, or is it compatible with Everett and mainly a different way of describing how classical histories emerge?

Rob Gebele
For a 1D quantum harmonic oscillator, can the classical notion of one-dimensional space itself emerge from the quantum theory? More specifically, is the Hamiltonian (or its spectrum and eigenstates) sufficient to recover an emergent notion of position, distance, and spatial locality—or does that require additional structure, such as a position operator?

lucas pollard
Which of these two possibilities would you find more unsettling: that every time we think we’ve reached the fundamental laws of physics, we discover another layer of explanatory machinery underneath them, or that we eventually hit a hard observational limit, beyond which even an arbitrarily advanced civilization could make no further empirical progress toward understanding fundamental reality?

Antoni Gawlikowski
Priority question*
My student asked me: how is it possible for the neutrinos to oscillate and not emit any radiation? I mean all 3 neutrino flavours have different masses right, so how can they switch from one to the other and not emit anything?

Burke Loeffler
What do you think Einstein and Schrodinger would say about many worlds?

Tim Gianitsos
You said in March AMA 2022 that naturalism, the view that science starts by assuming natural explanations for phenomena, is "totally bogus". Instead, you advocated for empiricism, the view that science judges ideas based on the predictions they make.
You're currently teaching a course on naturalism at Johns Hopkins, and the syllabus includes how to overcome objections to it. Do you now view naturalism as more defensible than empiricism?

Robin Luce
PRIORITY QUESTION
In Philosophical Naturalism and in Quantum Mechanics I, how do you treat time?
In ordinary quantum mechanics the Schrödinger equation has a parameter t, and the measurement problem is partly about when a possibility becomes a fact for an observer. In canonical quantum gravity the Wheeler–DeWitt equation has no t at all. You and Jim Al-Khalili just walked through whether time is fundamental or emergent.
If naturalism says the world is what physics describes, and the best quantum-gravity equations freeze time out, is the experienced clock an emergent bookkeeping device, or is that a sign the formalism is missing something real? And if the clock is emergent, does that change how you hear the cosmological-constant mismatch, or is that still just a separate parameter problem?
I am not asking for a mechanism. I am asking which of those two pictures you actually teach.

Tucker Hiatt
In classical mechanics, FORCE causes change in motion; that's a deep and vital concept. However, In quantum mechanics, the word FORCE seems to lose its ... force. Even using the term "interaction," as The Standard Model does, seems inadequate, especially when "principles" like that of Pauli and Heisenberg can cause macroscopic changes in motion. What's the best way to think about FORCE in light of quantum mechanics?

Flow
If you could receive one of these "Holy Trinity" watches as a gift and could choose only one, which (if any) would you choose? The Patek Philippe Nautilus, the Audemars Piguet Royal Oak, or the Vacheron Constantin Overseas? My choice would be the Overseas, and I’m curious which one you would choose.

Les K
As a basketball fan, like all other sports fans, no doubt you have reminisced about a time in the past, when your team was blessed with a core of great player who contributed to a golden era in your teams fortunes. Do you think Einstein, Bohr, Heisenberg, Schrödinger and others were giants in the realm of Physics in their time or was it that they happened to be around at an extremely exciting and revolutionary period in Physics?

Click to Show Full Transcript

[rough draft; check back later for polished version]

0:00:00.3 S1: Hello, everyone. Welcome to the October 2024 Ask Me Anything edition of the Mindscape Podcast. I'm your host, Sean Carroll. No long introduction today. I want to get into it, but just a couple of quick announcements. One announcement is that I'm still trying to get over a cold that I caught like a week and a half ago, but it's still making me cough and affecting my voice. So if I sound a little husky, that's why. That might also happen in a couple of the regular episodes that I recorded of late recently. I'm not gonna tell you which ones, but just because you're at death's door, that's no reason you can't podcast. You gotta podcast through it. That's the only way to get things done. So we're gonna do it anyway.
0:00:38.6 S1: And the other announcement is there is a little post on the Patreon if you wanna go visit that. That's patreon.com/seanmcarroll. Not that you have to join Patreon, but I just put on Patreon a couple of links to new useful web pages. So search for a public post, a post that is open to anyone to read, called New Useful Web Pages, and it'll link you to two different places. One is Andy Jiang. These are both listeners to the podcast who have just been kind enough to set these up. I don't have time to do this kind of thing. So Andy set up a search site specifically for the AMAs. There's a search capability on the regular podcast webpage, which is preposterousuniverse.com/podcast, where you can search all of the episodes. But if you specifically want to go to the AMAs, you can go to Andy's website. Best to go to Patreon and look for the link, but if you want to know what the URL is, it's http://mindscapesea.r.ch. Get it? So like .ch is if you're in Switzerland, but it's really spelling out Mindscape Search with a period before the ch.
0:01:50.9 S1: The other one is from Max Grody, and he has used AI and his coding skills to make Mindscape Explorer, which is basically a way to sort of do content analysis of all of the episodes. So in other words, you can ask, "Well, okay, this episode talked about something, but what other episodes talked about related things? What is the episode that is most like this one? What is the episode that is least like this one? What are the favorite topics overall or the least favorite topics?" Things like that. It's a little eerie, honestly, seeing all these things that you've talked about. Some of the things that are very frequent topics are not ones I would have guessed if you had asked me the number of years ago when I started Mindscape. But that's it. Life is unpredictable and chaotic and you roll with it. So many thanks to Andy and to Max for doing that. And I do suggest that you go check it out. We've covered a lot of ground over the years, that is for sure.
0:02:44.7 S1: And you too could be a Patreon supporter of Mindscape if you wanted to be. Just go to patreon.com/seanmcarroll and sign up and put in a buck per episode or whatever it is, or five bucks per month, and you get all sorts of wonderful benefits. You get ad-free versions of the podcast, you get the ability to ask AMA questions, you get the little reflection episodes, five minutes of me talking about what happened in the episode you just listened to, and so forth. And it's a good thing to be able to do it. I support plenty of other people on Patreon for the things that I get from them. It's a nice way to keep the economy humming. So. But you're still welcome to listen to Mindscape whether or not you're a Patreon supporter. And with that very communal, generous thought, let's go.
0:03:49.1 S1: Our first question is from Josh, who says, "OpenAI recently announced that an internal AI system produced a proof that the 3D Navier-Stokes equations can blow up in finite time, though the construction needs an external force and some mathematicians say it isn't the version they really care about. As a"

0:04:05.1 S1: physicist, how do you think about this? Navier-Stokes is a continuum approximation that presumably breaks down well before any true singularity. So what does a mathematical blow-up tell us about actual fluids, if anything? And more broadly, if AI systems start settling deep questions in math and physics with proof no human can fully follow, what changes about what it means to understand something, as opposed to merely knowing that it's true?
0:04:27.7 S1: Yes. So this is the big news in certain circles, and I think it's deservedly big news. Both this particular result from OpenAI, that they claim a solution to the question of whether or not the fluid mechanical descriptions called the Navier-Stokes equations lead to a singularity in finite time under certain conditions. And this is a very famous problem, a Millennium Problem. You win a million dollars if you get it. OpenAI spent much more than a million dollars trying to do it, so they were not in it for the money.
0:05:00.7 S1: And there's also, more recently, OpenAI just did a dump of a whole bunch of open problems that they claimed that they solved. And it's really causing a great deal of consternation in the math community because there are literally problems that a young or middle-aged mathematician has spent their life thinking about, and OpenAI comes along and says, "Oh yeah, we solved that one."
0:05:24.5 S1: And there's a lot of technical things to be said. As Josh alludes to, it's unclear whether or not the problem that OpenAI claims to solve is the most interesting version of the problem. From a physics point of view, it's not all that interesting. Again, like Josh says, the fluid mechanical description is supposed to be an approximation to a more true underlying particle description where you actually have molecules of the fluid, or even at a deeper level, a quantum mechanical description where you take quantum mechanics into account. And a singularity is a place where the density of the fluid goes to infinity or some other physical quantity goes to infinity. And that's exactly where you expect this approximation not to be that valid.
0:06:11.5 S1: So I'm not an expert on this. There might be counterexamples, but one's immediate intuition is that this has very little application to physics. It's a math problem. That's fine. Okay. Physics and math are two different fields. But as a math problem, it's very, very interesting. But whether or not they successfully solved it is still a little bit up in the air. It really is. There's a lot interesting going on here. We should probably have a discussion about it at some point.
0:06:36.9 S1: But the nice thing about math is you can check your work, and you can check your work at a computer or at a piece of paper. Okay? That's why it's very different from physics. If I had my LLM come up with a theory of everything, great, or even a theory of something more modest, I wouldn't be able to just check it to see if it was right. Checking it involves going into the world and doing experiments. You don't need to do that for the math problem.
0:07:04.3 S1: And mathematicians have gotten to the point where they can formalize a proof and they can plug it into a program called Lean, L-E-A-N, which will check to see if your proof is valid, whether or not you actually obeyed all the steps. Now, even here, there's some fuzziness. You expect a perfectly logical theorem checker to not make any mistakes, but apparently Lean does sometimes make mistakes, or at least it interprets things in ways differently than you thought it should be. So there's some controversy about that.
0:07:37.1 S1: And there's a lot of controversy over whether or not the thing, the proof that was announced in the paper released by OpenAI, is the same as the proof that Lean actually checked. You translate it a little bit, and sometimes you might translate it in very important ways. And finally, and maybe most importantly, there is controversy because all of these AI companies that have LLMs, various

0:08:06.0 S1: LLMs of various forms that you can use, get an account, etc., have training data, right? They train their LLMs on the corpus of texts written by human beings, as many as they can get their hands on. And nowadays part of that training data is what users type into the LLM. Okay? And so there's layers to this. There's a bunch of mathematicians who've been working on problems like the Navier-Stokes equations and whether or not they make a singularity, and they will use AI, they will use LLMs. So they will go to ChatGPT or whatever and they will type in their ideas. And none of these announced proofs, to a good approximation anyway, none of them are just, I prompt the LLM to solve the following problem. They're usually, at least they have been up until recently, situations where there's a human that is more or less collaborating with the LLM, guiding it, saying, "Well, why don't you try this? No, that doesn't seem like a blind alley. How about this idea?" etc. And so there's these real-world mathematicians who've been trying to solve the problem with the help of an AI and not part of the OpenAI project. But did OpenAI, number one, was their model trained on the conversations with the mathematicians? Okay. And number two, slightly worse from a sort of moral philosophy point of view, did the people at OpenAI literally look at those conversations to see what the human mathematicians were trying to do? I don't know the answer to any of these questions. I'm not trying to hide the reveal at the end. I truly don't know. These are questions that have been asked, okay? And so it's all a little bit messy. And there's sort of two things that are more or less clear. One is that there's a lot that is sketchy about what the AI companies are doing. They're not releasing what a mathematician would think of as a readable, understandable version of the proof that they claim to get, because they're just rushing. They just want to get it out as quickly as possible. So they just dump some version of what the LLM has put out that is often very, very difficult to understand from a human point of view. So even if you have in some sense proven something, it's unclear whether or not you have increased the sum total of mathematical understanding in the world. But the other thing that is absolutely clear is that these LLMs are changing how we do math in a really deep and transformative way. Okay? I mean, love them or hate them, there's just no question that these are changing the world. And I have the vague opinion that there will be in mathematics quite a sea change in how math gets done, but not one that removes the need for human beings from the equation, as it were. The point is the human beings understanding the world better, right? Including the world of mathematics. So just the fact that you have a proof from an LLM means almost nothing, except maybe it's a little bit more inspiration to the humans to try to understand what was in that proof. But it will, the LLMs are just gonna become better and better, I presume, at doing this kind of math. So the way that one thinks about being a mathematician changes dramatically, right? Maybe spending your life like Andrew Wiles did, trying to prove a difficult theorem like Fermat's Last Theorem, isn't the way most mathematicians will work in the future. Maybe they'll be collaborating with some LLM and their job will be more creatively interpreting what was actually figured out by the LLMs, or really trying to figure out what are the most important questions that we can ask that we might have a chance of getting an answer to. So in other words, I don't think that there's any prospect that mathematicians will be put out of work, but the work they do might be of a very different kind than what they used to do. I made the slightly joking remark on BlueSky that the physics version of this is that LLMs could put an end to the "shut up and calculate" era.

0:12:20.5 S1: philosophy of the foundations of quantum mechanics. Because there's no reason for you to shut up and calculate. The LLMs are better at calculating than you are. What is left to humans is thinking and being creative and understanding things. And there's certainly a positive spin on that, right? That maybe we'll just, our job will be understanding and doing the fun part of this understanding process. Like I said, the underlying process of physics is very different than that of mathematics, so the impact on physics is understandably less, or at least not quite as imminent, but it'll be there. I mean, every theoretical physicist who I know is, almost every, is using LLMs in some way or another, not to do the work, but to improve their own understanding. Proving theorems isn't the goal in physics, right? We're trying to understand what could be happening, and so that's what you try to use the LLM for. Including myself. You know, I work in a very interdisciplinary kind of way, so very, very often I am going to be in a situation where I want to learn some subfield of a subfield that I haven't really delved into before, and there's just no question that talking to the LLM makes that process easier. And I know there's a lot of skeptics out there who will say, well, the LLMs will hallucinate or whatever, or make up references. Number one, that happens less than it used to. There's been progress and it's not what it was two years ago. But number two, sure. Like, that's okay. People do that too. People make mistakes. I'll just keep saying over and over again, the one thing you should never do with an LLM is to cut and paste its output and call it your own in any document that means anything to anybody. Like, I don't care if you want to write your webpage using an LLM or, you know, write a form letter to some official using an LLM. Like, that's fine. Who cares? No one's gonna read those things, or if they do, it's just boilerplate anyway. But if you care about what you're doing, if you're writing a paper and putting your name on it and this is part of your contribution to society, then use the LLM to improve your brain and your understanding, but the words that you type into the text should be your own. And that's the way that I have been working and will always work. I don't think that's ever gonna change. But it's a big change. It's definitely, and it's just starting. We're not there yet, so it'll be interesting to see where everything goes.
0:14:43.0 S1: Okay. Jennifer Lecompte says, "I heard a blurb that there is a shortage of physicists and astrophysicists. Upon looking further, I cannot tell if this is truly the crux of the problem or if we are really just looking at the shortage in funding. From your position in the academic world, is there truly a shortage of physicists and astrophysicists, or are we seeing the effects of a lack of funding? Or maybe there isn't a shortage, but we know funding is definitely a huge issue." I don't think that there's a shortage of physicists and astrophysicists. I mean, it depends on what you mean. Let's put it this way. What I know for sure is there are more people who want to be physicists and astrophysicists than we have jobs for. Okay? That's a huge thing. I don't want to say that it's a problem or not. Maybe it's good for the world, but maybe it's bad for those people who are trying to find jobs, right? I don't know what the correct number of physicists that society should support should be. I don't think there's any one right answer to that question. I think that the system that we have traditionally had, where you elect government officials and the government officials recognize that they are not experts, so they appoint experts to lead a scientific organization like the Department of Energy or the National Science Foundation or the National Institutes of Health, and those people in turn delegate responsibility to panels of experts who really do know the field very, very well, and through this combined multi-layer process—Congress also involved, since in principle, again, before the last 10 years, Congress was actually involved in setting the budget for the country—you would figure out, you know, what the country as a whole, as an emergent entity, decides to spend on its scientific efforts.

0:16:21.7 S1: endeavors. And I think that you can absolutely make a case that it should spend more. Okay, I'm happy to make that case. But it's not like we lack the people to do it. I tell my graduate students, and this might be an outdated statistic, but I think it's a good rule of thumb: if you get a PhD from a pretty good place, pretty good university in theoretical physics, your chances of becoming a tenured faculty member, which is probably what you want to do, are maybe one in four, right? Maybe about 25%. If you don't become a tenured faculty member, you'll generally get a better paying job than becoming a professor, so it's not like you're starving. But most people who go to grad school want to become tenured faculty members someday. So we have more than enough people who want to do that. So I'm happy to advocate for spending more money on science, but I don't think that the bottleneck is people who want to do it.
0:17:14.5 S1: The other thing, of course, is that the last two years in the second Trump administration have completely decimated science funding in the United States. And it's not I'm worried about the future; it's that right now it's bad. Like, I can see it's bad in my own life. But I am also worried about the future because the downstream effects of this are going to take decades to fix, if we ever choose to fix it. I don't know. But the smartest people in other countries are no longer coming to the United States, either to be students or to be faculty. No one trusts us as an international partner. People are trying to leave the country. People who are in the country are having difficulty at every level, from getting into grad school to getting funding as a senior person. Francis Halzen, as I'm recording this, just won the Nobel Prize in Physics as the leader of the IceCube experiment, which is a neutrino observatory in Antarctica. And the Trump administration just cut its funding in half just before he won the Nobel Prize. That's the situation we're in, and that's gonna be, that's gonna hurt the country and the world in a way that will be very, very long-lasting.
0:18:27.3 S1: Okay, I'm gonna group two questions together. I'm allowed to do that. Stan Manalov says, "In the September AMA, you said that Mindscape won't run for a billion years. I guess this is because you don't expect to live this long yourself. But have you considered training a replacement host and handing the podcast over to them once you don't want to keep driving it?" And then Tyler says, "If you continue hosting Mindscape for another decade, do you think AI could produce a reasonable replica of your host character? And would you hand off the show to Sean CAIroll? Get it? C-A-I-R-R-O-L-L rather than just C-A. Get it? Putting the AI in there when you are ready to retire."
0:19:07.5 S1: So I think no to both. But I reserve the right to change my mind about these things, but I have zero interest right now in either one of those possibilities. For better or for worse, I have never tried to make Mindscape anything other than my own thing, right? I have not tried to please a broad audience. I mean, I do want to please a broad audience, don't get me wrong. I just haven't tried very hard to please a broad audience. I kind of do what I want to do and then hope that there's an audience out there for it. That's been my way of existing for a very long time. And happily, there is a medium-sized audience out there for it. An audience much bigger than I might have expected, much smaller than the big podcasts that are really out there. So that's fine. I think that's good. But it means that my own interests drive the show, the tone of the show, the individual episodes, the selection of topics, all that stuff. And I just don't think it makes sense to hand it over to somebody else. If we had done it as a franchise, I don't know what to call it, if there had been some purpose to the podcast other than things Sean thinks are neat, if it had been more targeted, in other words, like the most recent wrinkles in physics or whatever, then you could easily imagine handing it over. But as it is, I think that it's kind of...

0:20:33.4 S1: my deal. I don't know anyone else who would want to do it. I mean, if anyone else who would be along those lines should start their own podcast, and I'd be happy to support them on it. For the AI thing, again, no, I think that here is a situation where knowing that it is actually me saying the words matters. The Pope, of all people, Pope Leo, mentioned, talked about this on social media recently, and I don't think he exactly got it right. He was talking about art, not podcasts, but I think that there's a close analogy here. And he was talking about the fact that a work of art, the way that he put it was a work of art created by a human being is ontologically different than a work of art created by a computer. And I don't think ontologically is the right word there. I know what he meant, but it's not how I would put it. He meant ontologically in the sense that the essence of what the thing is is truly different, even if the actual locations of the pixels or the paint or whatever are the same in two cases, but one is computer-generated, one is human-generated. Or if the words are the same, the essence of the thing is different. I don't think that's exactly right. And so I suggested that it would be better to say—and the reason why I don't think it's exactly right is because I'm a physicalist, and I think if all the particles are exactly the same, there is literally no difference between the two things. But there is an etiological difference between the two things, which is a way of saying that the history is different, the way that the thing came to be is different. And that matters for a work of art. When you're reading a novel and you are enjoying the plot and everything, but there's part of you that is thinking, okay, someone wrote this novel, right? Someone had this thought, someone made the choices to kill this character or whatever. And that matters. When you see a Van Gogh or whatever, it matters that you know some of that history behind it, or at least that there is such a history behind it. And your feelings change if you were to learn that it's artificially generated or—it's nothing new about technology—your feelings change when you realize that it's a forgery, right? Or a fake, in various kinds of things. The history that led to the creation of this artifact matters. And I think the same thing is true for a podcast like this. I do suspect, when it comes to AI, I have my worries, and we'll talk about some of these worries later on in the episode. But I also think that there is some level of inevitability. We're just at the beginning, and just holding your hands over your ears and saying, "I'm not going to pay any attention to this," is not really a successful strategy for coping with the world that we are entering. So I have almost no doubt that things like that, AI-generated versions of long-running podcasts or whatever, will happen. And I don't even know if I will be able to stop it if someone wants to train an AI and pretend to be me. Maybe—I don't even know whether I would want to or not. Maybe I would want to. Probably I would want to stop that. I don't think it's a good idea, but I'm not sure if I'm allowed to or can effectively. So who knows what's going to happen? But for me, the fact that it's me, even if I have a cold, that is saying the words that leads to the episode of Mindscape, that's, I think, an important aspect of the show.
0:22:33.8 S1: Okay, next question comes from Rad Antonov, who says, "On page 22 of Modern Quantum Mechanics, Sakurai cites Dirac's definition of a measurement: 'A measurement always causes the system to jump into an eigenstate of the dynamical variable that is being measured.' Has this definition stood the test of time? Would you use it to define a measurement in your quantum mechanics class?" So I don't know whether either Dirac or Sakurai meant it to be a definition of a measurement, but it certainly isn't. When you say...

0:24:33.5 S1: When you say a measurement always does X, that's not defining what a measurement is. That's saying what a measurement does. That's saying the effect of this thing. Okay? In order to define what a measurement is, you have to tell me, okay, what is the physical process that happens that has the consequence of the system jumps into an eigenstate of the dynamical variable that is being measured?
0:24:56.7 S1: We do teach—okay, having said that, so it's not a definition, it's sort of an aspect—the basic idea is true. Yeah, that's completely part of what you teach your undergrads in your quantum mechanics class. When you measure a system—so let me, I'll say it the correct technical way and then we can sort of try to understand it. There are operators in quantum mechanics. So a quantum state is a vector in Hilbert space, to use the technical language. But who cares? It's a quantum state. And what you do is you have operators that change the state in some way. And every observable kind of operator has a special quality. It's called a Hermitian operator. It's the equivalent of a real number as a subset of the complex numbers, roughly speaking, because we want our measurement outcomes to be real numbers, not imaginary numbers.
0:25:46.8 S1: And then there is a special set of states for every observable with the property that if you measure it, the state doesn't change. Okay, a typical operator acting on a state, like the operator might be rotate around the z-axis by 90 degrees. Okay, that's an operator. And most vectors change when you do that operator. But the z-axis vector, a unit vector pointing along the z-axis, does not change when you rotate around the z-axis. It just stays put. So that's called an eigenstate of the operator. And every observable, every Hermitian operator, has—there's something called the spectral theorem, which says that there's a set of eigenstates which form a basis for the Hilbert space. There's some footnotes there about degeneracies, etc., but roughly speaking, that's true.
0:26:37.7 S1: And then like you say, when you make—what it means to say that you measure that observable is you start with whatever state you start with and you end up in one of those eigenstates. That is called collapse of the wave function. So for example, for measuring the position of a particle, the eigenstates of the position operator are states of definite position, which is to say states that are 100% concentrated at some particular location in space. If you worry about that because there's an infinite number of locations in space, good, you should worry about that, and there's a whole bunch of high-powered math that gets addressed to solving that problem. But roughly speaking, that's what's going on. And before you do the measurement, you could be in a superposition of many, many different possible locations. And after you do the measurement of location, you're in a position, you're at that place.
0:27:28.7 S1: Of course, there's caveats also about is your measurement perfect or is it just approximate, all that stuff. But the spirit of that statement is 100% on the right track. It is also 100% not a definition of a measurement. In fact, measurements are generally not defined in undergraduate quantum mechanics. That's one of the issues that I kind of like to harp about.
0:27:50.3 S1: Daniel Bagley says, "As a writer of poetry myself, I was super grateful that you had Elizabeth Alexander on the podcast. But there seemed to be a note of desperation about the state of poetry in the public eye. Moreover, it was depressing to me that the idea of play was not mentioned once, as many of the great poets I could think of play with words quite a bit, and it's entertaining and thought-provoking. Do you think poetry like this is going the way of the dinosaurs, or can poets still do playful, strange, innovative, experimental, and/or difficult things with words and find a decent audience?"
0:28:19.6 S1: Well, I do think that the audience for poetry in the modern world is kind of tiny compared to the audience for Marvel movies or whatever. I don't know how you can argue with that. I don't know whether it's growing or shrinking or whatever.

0:28:35.0 S1: It's still there. There's still poems that appear in the New Yorker, and I get the New York Review of Books, and there's always poems in there. Poet Anne Carson just won the Nobel Prize for Literature, etc. Of course, we have a version of poetry that we didn't have 200 years ago in the form of the popular song, right? The lyrics to the popular songs, which aren't always the best poetry, but they are poetry. So I don't have strong, informed opinions about the market for poetry. In terms of if there's room for poets doing playful things, of course, of course there are. And you shouldn't interpret anything that Elizabeth Alexander and I said to imply otherwise. When you talk about poetry, there's a million things you can talk about. You can't cover all the ground. And I'm sure that everyone has their own favorite thing to say in the world of poetry. Elizabeth's book of essays that just came out was in particular about the role of poetry in keeping the spirit of democracy alive, right? Contributing to the group social project, the collective good, etc. And so that's by nature relatively serious. The playful aspects of poetry are not going to be stressed in a conversation like that. But nothing that we said should be taken to imagine that you can't be playful. Elizabeth herself has written plenty of playful, strange, innovative things in her own poems.
0:30:03.0 S1: Christopher Smith says, "Is standard solipsism cognitively unstable for the exact same reasons that a Boltzmann brain is cognitively unstable?" So I guess the short, cheap answer is I don't know, but I'm gonna explain what the difference is, which is why I know on one side and not the other. So remember, the Boltzmann brain idea is that you have a randomly fluctuating set of things, a thermal distribution or something like that, atoms, particles, what have you. And if you wait long enough, they will randomly fluctuate into every possible, every allowed configuration, including a configuration which is an intelligent observer or even the whole Earth or the whole galaxy if your system is big enough. But if you run the numbers, with overwhelming probability, when an observer fluctuates into existence, even an observer who is exactly like you, indistinguishable from you right now macroscopically—so in the same room, thinking the same thoughts, wearing the same clothes, all those things—that kind of observer will, with overwhelming probability, fluctuate into existence surrounded by otherwise thermal equilibrium, not surrounded by the actual warm, comfortable, inviting, hospitable universe that we think we live in.
0:31:19.1 S1: And part of what has randomly fluctuated into the brain of that observer is all of their thoughts, all of their impressions of the world, all of their reasonings about the laws of physics and the laws of logic, etc. And therefore, that observer has zero reason to trust any of their own thoughts because there's no relationship between, for example, a memory that they have in their brain and an actual event in the past that that memory was supposed to correspond to. The connection between memories you have right now and actual events in the past relies on the past hypothesis, relies on assuming that some time ago the universe had a much, much lower entropy than it does now. And for a typical Boltzmann fluctuation, that would not be the case. So if you think that you live in a Boltzmann brain-dominated universe, you cannot both believe that fact and have a good reason to believe that fact. Therefore, it is cognitively unstable.
0:32:18.5 S1: Okay, but an important part of that story I just told you is played by equations, is played by the fact that we have a definite setup, that we're not starting the story by saying, "What if you were a brain floating in empty space?" Okay, that's just not the beginning of the story. The beginning of the story is, "What if the laws of physics are such and such?"

0:32:38.0 S1: What if you have a really, really big system with random thermal fluctuations that last essentially forever? And then you do science to that and you ask what it predicts. Solipsism is a philosophical stance that says, "I'm the only person who exists. Everything else that I think out there exists, all the other people, etc., are just coming from my mind." Okay. Well, that's a thing you can imagine, but there's no equations attached to that. I have no way of saying, "Oh, okay. In that situation, your imaginings are disconnected from the rest of the world," just because it's not very well defined. It's not very well specified, right? This is why I think that the cosmological versions of various conundra, both including Boltzmann brains and including things like self-locating uncertainty and anthropic selection and so forth, are much more interesting than the philosophers' thought experiment versions of these questions, just because they are well defined. They come from not just positing the puzzle from the start, but positing some physical situation and deriving the existence of a puzzle. So I think one could make an argument that a solipsistic perspective is cognitively unstable because if I'm imagining all of you, I am also probably imagining all the reasons I'm imagining all of you and all that stuff all the way down. But maybe not, because I could probably invent a version of solipsism where that's not the case. So I just think it's a little bit harder to pin down whether it has those properties like a good, well-defined cosmology does.
0:34:09.9 S1: Brent Meeker asks a priority question. Remember that every Patreon supporter gets to ask a priority question once per their lifetime, and I will do my best to answer it. So Brent asks, "On the one hand, black holes forming from the collapse of a star, from our standpoint, only asymptotically approach having an event horizon. It takes forever for the event horizon to form. In the meantime, infalling stuff is kind of suspended on the way. On the other hand, Hawking radiation comes from the vicinity of a black hole out several diameters. And if I understand the process correctly, it doesn't actually depend on the event horizon having formed. So it causes the black hole to evaporate in finite time even before it forms. That would seem to obviate the black hole information paradox. Am I looking at this wrong in some way?"
0:34:57.7 S1: In some sense you're looking at it wrong. It's not quite on the right track, but it's a very well-posed question. I think it's a very interesting way of thinking about it. And look, physicists have thought about exactly this thing that you're asking. So let me restate what's going on here a little bit. Very often when we talk about black holes and their properties, we're just thinking in the context of classical general relativity. We're not thinking about quantum mechanics, quantum field theory, or anything like that. And it's a little bit of an issue for the popular understanding of science because the scientists know that when they're making all these definitive statements about what the universe can do, what the black hole can do, and whatever, you've never seen a black hole up close and personal, right? You've never done experiments on it. You've never seen the whole universe from a remove. So what you're doing is you're assuming some theoretical background, and often it is not conveyed to the listener what theoretical background it is you're assuming. So a lot of statements about the notion of the horizon and where it is and things like that, all of these sort of presume some framework which might not be true, namely classical general relativity.
0:36:09.3 S1: Once you include Hawking radiation, you're not in the world of classical general relativity anymore. The black hole actually evaporates. So there is a sense in which there is no event horizon in an evaporating black hole. And the sort of simple way to imagine that sense is just to say, look, if the black hole evaporates, everything that goes in ultimately comes out. I mean, every piece of information anyway. The actual physical pieces of matter might be scrambled into something else, right? They might be turned into photons or whatever. But

0:36:41.3 S1: But there's a way that what goes in does eventually come out in the form of Hawking radiation. The whole issue of black hole information loss is, is the stuff that comes out directly, predictably related to the stuff that went in, or is there an intrinsic element of randomness there that destroys the information? But it does come out. So in that sense, there can't be a horizon, because a horizon is defined to be a place such that once you go in, you don't come out. I don't know, I said "note come out." See, I have a cold. What did I tell you? So that, but that's a kind of a technical thing. The point is that in a real black hole, in a big black hole, even if in some minor technical sense there's technically not a horizon, it looks exactly like there's a horizon for a very long time. You really have to keep in mind that Hawking radiation is incredibly faint. For a macroscopic, astrophysical-sized black hole, it's gonna take a very, very long time for the black hole to evaporate, and so you can really treat it as if there's no Hawking radiation at all to a very good approximation. So for every single possible thing you could do, it looks very much like there is an event horizon and you know where it is. And so we call that the apparent horizon or something like that.
0:38:02.7 S1: The one thing for Brent's question that I would undo is this talk about "from our standpoint, you only asymptotically approach having an event horizon." Who cares about our standpoint, right? That's just a coordinate choice or something like that. You should try to ask these questions in ways that don't depend on anybody's standpoint to the extent that you can. Therefore, what you do want to do is say, "Okay, I'm gonna let the black hole evaporate, and I'm gonna analyze it including the loss of energy and perhaps loss of information from the Hawking radiation and things like that." And I'm gonna ask myself, "Is it still true that it looks like information is destroyed at something like a singularity, even though the black hole's evaporating?" And the answer is yes. People have done this, right? People have looked at this. So you try your best to understand what the spacetime looks like when the black hole's evaporating, etc. It's very easy to construct worldlines in that spacetime with the property that they cannot escape to the outside world in the time that it takes the black hole to evaporate. They will just hit the singularity. And now, the singularity might not exist in quantum gravity, but that's not a big deal. You replace the singularity that you get in classical general relativity by whatever it is you do get in quantum gravity, and your future worldline is guaranteed to hit that if you fall into the black hole. So that raises the question of where the information goes in that particular worldline, whether or not you have a technical definition of an event horizon actually forming.
0:39:41.7 S1: Mikkel Bennedsen says, "I just watched your World Science Festival conversation with Brian Greene. I had a question about your treatment of self-locating uncertainty. You gave the example of dark matter being either WIMPs or axions, suggesting that if I assign, say, 40% credence to WIMPs, I can think of this as self-locating uncertainty between a WIMP world and an axion world. But can this picture handle ordinary epistemic uncertainty about possibilities that are not genuinely possible? Suppose the true fundamental theory entails that WIMPs are physically impossible. Does it still make sense to describe my uncertainty as self-locating? Put differently, does understanding credence as self-locating uncertainty require a distinction between epistemically possible worlds and genuinely possible worlds? And if so, in what sense is the uncertainty still genuinely self-locating?"
0:40:31.4 S1: So this is an excellent question. I should say, I'd like to think of these probabilities in terms of self-locating uncertainty between possible worlds, but I'm not an expert in this particular area, and there might very well be really good reasons not to do that.

0:40:48.9 S1: And Mikkel is pointing at a potential really good reason, which I again don't really have a strong feeling about. But it's the question of what worlds are actually possible. Okay, I'm a little bit skeptical of the example. The question says, "Suppose the true fundamental theory entails that WIMPs are physically impossible." Well, that depends on the true fundamental theory. In a different possible world, you might have a different fundamental theory, right? So that's not exactly the best example of what's going on. I think that what is trying to be gestured at is that it's almost—maybe it's exactly, I'm not sure, but it's very close to a distinction philosophers draw between possibility and conceivability. I don't really like these words. I don't think that they're conveying what the words actually mean. But basically, possibility means that the world or the situation really is possible. Maybe not physically possible given our laws of physics, but there could exist other laws of physics so that the thing is actually possible. And conceivable is closer to, "Well, I don't know if it's possible or not, but I think it might be possible, so I can conceive of it," right?
0:42:02.4 S1: So the truth is, since we don't know the theory of quantum gravity or the theory of everything or anything like that right now, if you said, "Is it possible that there could be weakly interacting massive particles that have the right properties to be dark matter in our world?" as far as I know, the answer is yes, right? And maybe there is some deep feature of the fundamental laws of physics that says, "No, that's actually not physically possible." Maybe that's in the swampland, like we talked about with Cumrun Vafa some time ago. Maybe that world that you think is perfectly reasonable actually is never part of a fully well-defined theory. So there isn't really a possible world. I suspect—and this is just my suspicion again, I haven't gone through it carefully—I suspect that you could fix all that up, right? So you could sort of talk about partial possible worlds or presently conceivably possible worlds or something like that. Now, it might be that you do enough work to make that all fit together and you realize that talking in terms of self-locating uncertainty just isn't the most helpful way of thinking about it. That's very, very possible. I'm open to that imagined conclusion here, but I just don't quite see the need for it just yet.
0:43:26.3 S1: I think, more importantly, the thing that I would hold on to more strongly than that, more than the view that all uncertainties are self-locating uncertainties, is that uncertainties are essentially subjective, which is kind of another way of saying that they're self-locating uncertainties in the space of possibilities, right? If I knew something was impossible, I would give it zero credence. But since I don't, I give things non-zero credence. But it's me giving it that is the ultimate subjective element here. And even if there's something about the future that I don't know, but it's definitely going to happen, I just don't know what's going to happen, that's still a subjective probability that I would assign to it, even if there's some definite feature, because I don't know it, right? So I think it's very useful. And this is not just me. This is an at least well-known—I want to say respectable, but I'm not going to judge that one way or the other—but at least very well-known perspective in the philosophy of probability, that probability is all about our knowledge, not about objective chances out there in the world. And other people think that there are both, that some probabilities are objective and some are subjective. And that's a whole interesting thing for people to talk about.
0:44:43.9 S1: Gary Miller says, "In your recent conversation with Chandra Sripada, you discussed how pressures toward predictive modeling can give rise to increasingly sophisticated cognitive tools. Now that we're"

0:44:53.4 S1: Now that we're seeing something analogous happen in AI, does that raise your credence that advanced cognition or intelligence might also emerge elsewhere in the universe? There are obviously many other hurdles between simple life and technological intelligence, but does AI provide any evidence that once sufficiently powerful predictive modeling develops, high-end cognition may be a more natural or repeatable outcome than we previously thought? In other words, would this make you update any of your probabilities relevant to the Drake equation and therefore raise your credence in extraterrestrial intelligence?
0:45:25.4 S1: Not that much, I don't think, because I never thought that the drive to increased intelligence was one of the bottlenecks to intelligent civilizations being out there in the world. I think that there's the two obvious bottlenecks are, one, life existing at all. Maybe it's just really, really hard or rare for anything that we would recognize as life to start existing at all out of some prebiotic soup of some kind. Soup is not the right word, but conditions on some planet or in some ocean or in some cloud or what have you.
0:46:00.8 S1: And the other obvious bottleneck is sort of complex multicellular life, right? Like maybe there's cells, individual little containered sacs of biological material that reproduce and whatever, and that's it. That's as complicated as it ever gets. I think that once you have differentiation of cells into different kinds that come together to form a multicellular organism—I might be wrong about this, but in my mind, then intelligence is quite likely. That's very far from intelligence, but I think you've given yourself, by differentiating your cells and giving yourself some capacity to sort of have a division of labor within the organism that you couldn't have if you're unicellular life, then given enough resources and enough time to survive and evolve, I think it's pretty easy to develop intelligence, or at least pretty likely. Now, that's not a very informed view, so maybe I should have been more skeptical before and maybe I should have updated my credences, but the credences were already pretty high, so that hasn't really changed them.
0:47:10.3 S1: Redmond says, "I'd like to know more about the differences between and among the following categories: astronomers, cosmologists, theoretical cosmologists, astrophysicists, astrobiologists, and other space-related fields. Not only about their areas of inquiry, but their training, prestige, difficulty, remuneration, etc. Do you tell jokes about each other?"
0:47:30.9 S1: Well, it's a good question and it's not a very well-defined one, okay? Some of these boundaries are a little bit slippery. In a typical university, sometimes—almost every university has a physics department. Some universities have astronomy departments and some don't. At Johns Hopkins, we have a Department of Physics and Astronomy, so we have one department, but astronomy is called out there in the title. Very often, there are astronomers at universities, but they work in the physics department.
0:48:02.2 S1: Also very often, even if there is an astronomy department, there can be people in the physics department who do astronomy. And part of the difference—and this was explained to me by one such person—is whether or not you build your own instruments. So there's this, again, slightly traditional—it's not at anything like nailed down or enforced—but people who measure or do observations of the cosmic microwave background tend to be radio astronomers because microwaves are in the radio regime, and they build their own receivers, their own antennae, and they build their own telescopes or whatever it is. Whereas someone who is an optical astronomer, who is looking at the spectrum of stars or quasars or whatever, they will generally apply for time.

0:48:55.9 S1: at a telescope that someone else has built. Okay. This seems to be not a very good division of labor, and I think it's also changing because I think that the radio astronomy game is becoming more big time in terms of collaborations and things like that. A lot of the best CMB measurements are from satellites, satellite observatories that, true, they were built by the people who do them, but they're building part of it, right? They're not building the whole thing. And many of them are now in astronomy departments. Cosmologists are just people who study the universe as a whole, or at least let's say galaxy size and above, right? Things that are bigger than a galaxy. That's a very broad classification because there are people who study the formation of galaxies and they're embedded very deeply into standard, well-known physics, right? Gravity, magnetohydrodynamics, the atomic and molecular processes in the interstellar, intergalactic medium, magnetic fields, etc. Okay. Whereas people who are also cosmologists but are studying things like what happened at the Big Bang, or why is there an imbalance between baryons and anti-baryons in the universe, their toolkit is very different, their training is very different, but they're both cosmologists. The latter group, the sort of early universe cosmologists, will often slide into what we call high-energy theorists, the people who do particle physics, quantum field theory, string theory, things like that. The fact that people who do string theory, for example, are called high-energy theorists is actually a relic of the days when intellectually that group descended from people in the mid-20th century who were designated by the fact that they were building high-energy particle accelerators and looking at what came out of them. So that field became known as high-energy physics. But it's really some different—there's different flavors of people doing particle physics or quantum field theory or string theory or whatever. These days, very often the people who do string theory are more or less intellectually disconnected from the people who do what we call phenomenological particle physics, particle physics that is closer to the data and the accelerators and so forth. Astrophysicists versus astronomers—I'm not at all sure there's any difference whatsoever, really. I mean, people might try to pretend that astronomers look through telescopes and astrophysicists think about the processes that are going on in what they see, but honestly, both do both, I think. There are differences—there are definitely differences in physics between theorists and experimenters that are a little bit more pronounced than in astronomy, but it's still there. So there's definitely theoretical astrophysicists, whether they're thinking about black holes or quasars or star formation or whatever, and there's observational astrophysicists, and the observational side is closer to what are usually called astronomers. But again, it's not set in stone. Astrobiologists are very different because they're often just biologists thinking about space, right? They're not trained—you don't need to know a lot about galaxies or stars or the early universe to do astrobiology. You need to know a lot about biology and chemistry and maybe even information theory or something like that. So I think that there is a point to having these labels, but that point is often organizing the university, right? Because at the end of the day, when you accept new students or even when you hire new faculty members in the modern university, most of the responsibility for that is at the level of a department. Okay, like our physics department will choose who to hire to be the next physics faculty member, or the physics department will accept students to be the next graduate students. The undergraduates are chosen university-wide. That's

0:52:59.6 S1: That's that little division of labor coming in. And it would be impossible to accept graduate students or to hire faculty members in a university-wide way just because the scale is wrong. There's just too many people and you need too much expertise to figure out who would make a good faculty member, who would make a good graduate student, et cetera. So you have to divide it up somehow, and it's a little bit arbitrary. I mean, at Hopkins we have a separate biophysics department in addition to the physics department, but there's also people in the physics department doing biological physics. Okay. So none of these boundaries are really very good. And well, they're really—I should say they're good. I should say that they are not really impermeable, let's put it that way. There's a reason they exist, but then they become problematic once they exist. Because once you've invented a category or a set of categories that are supposed to be exhaustive, you want to put people in a category. You ask new faculty members, "Are you really a biologist or a physicist? Are you really an astronomer or a cosmologist?" or whatever it is. That becomes problematic because then you're saying, "How do you fit into my box?" rather than, "Are you good at what you do?" And I think that is a major problem for modern academia, that they have reified their boxes and they treat people like you need to fit into one of the boxes. Otherwise, like I've known at Johns Hopkins, I've tried to suggest a different person to be hired as a faculty member, and the question is, "Well, what department would they go in?" I think that's a terrible question. If you're asking that question, you've lost. But on the other hand, someone has to pay for it, so that's the way the system works. I have fantasized about a university without departments. Just really smart people on the faculty, and they would sort of just in an ad hoc way, every student would construct their own program of study, and different faculty members would come together to do different research projects or whatever. I have no idea whether that would work, but I do think there should be more experimentation among higher education institutions with things like that. The only way to know if it works is really to try to do it.
0:55:16.5 S1: Andres says, "We have known since the 1930s that logical systems with sufficient complexity will exhibit Gödelian behavior. I think it's somewhat safe to assume that the structure of the universe could be mapped onto a logical system that fulfills the conditions for the Gödel theorems. Therefore, I find it somewhat perplexing that in the past hundred years, there haven't been major attempts at a physical theory for the universe that exhibits Gödelian behavior. Like in QM, we could attack the measurement problem by considering that the 'particle is in a position X' statement is Gödelian, and its truth can only be established by adding a new axiom to the logical system corresponding to measuring the location of the particle. What do you think?"
0:55:53.9 S1: I think that this is not very well-formed, to be perfectly honest. It's starting right from the start where you say that logical systems with sufficient complexity will exhibit Gödelian behavior. I don't think that's the right way to say it. I mean, what do you mean, "exhibit Gödelian behavior"? I think that Gödel's theorems are theorems. They're mathematical theorems. They are stated very precisely and they have precise meanings. We don't have to sort of cast them as exhibiting a certain kind of behavior. There's a very specific statement—there's different theorems—but the main idea is that in a sufficiently powerful formal system, if it is consistent, there will be statements that you can formulate that are unprovable but true, is one way of saying it. But that's a statement about mathematical systems. It's not a statement about physics. And they really are different things. Physics is a different thing than mathematics. In math, the traditional way of thinking about math is you have some set of axioms, you try to prove theorems. And what Gödel showed, in one way of thinking about it, is in the set of all...

0:57:02.3 S1: statements that you can imagine. The traditional way, pre-Gödel, of classifying the set of statements would be that there's some statements that are not well-formed, right? Like just grammatically incorrect. There's some statements that are true and provable, and there's some statements that are disprovable. You can prove that they're false. And what Gödel showed is that there's a new category of statements that are not disprovable or provable, right? So that's great, but it has nothing to do with how physics works. Physics works by imagining that there are—well, a standard way of physics working is there's a set of states that a physical system can have. There's some dynamical laws that say how the physical system evolves. And there are solutions to these dynamical laws that say, from the given state, where would you go in the future, where have you come from in the past. None of that is theorems or proving or anything like that. So I'm not quite sure what the analogy or the connection is supposed to be. Certainly with quantum mechanics as an example, the idea that the particle's in the position X is Gödelian—I'm really not sure what even that would mean. But also, I don't think you need it. I don't think you need to imagine this. We know perfectly well what is going on. Particles are described by quantum states. They're not in positions X. That's not what particles are. They're in wave functions. And what you measure about the wave function is different than the entire wave function. You measure the position of a particle, and that's a system and it works perfectly fine. I don't think we need to think very far outside that box.
0:58:37.3 S1: Airplane Ears says, "You mentioned errors in the last AMA, and I was curious how much you think about your own aging, specifically as it relates to memory and general mental acuity. Do you consider about moderating your behaviors that could accelerate deterioration, such as drinking?" Well, I do think that I'm at an age where the ages are getting up there. My own age, the number attached to the number of years I've been here on Earth, is growing, especially as a fraction of a typical lifespan. And once you've reached my age, you can't help but think about those things. The important operational question here is, what about moderating behaviors that could accelerate deterioration? There, I'm a big believer in moderation specifically. So I'm not a big believer in general—and maybe this is all just self-justifying nonsense, who knows—but I'm not a big believer in saying, "You shouldn't do this thing because it has some bad effects." I think most things have some bad effects. Most things have some good effects. You're lucky if you can figure out some things that are purely good or purely bad. And you have to balance. I'm Aristotelian in that way. You have to look for a kind of happy medium, or Buddhist, or I don't know whichever. It's not a philosophy that is very hard to invent. It's been invented many, many times over history, the idea that you need to sort of weigh good things against bad things. So I don't want to live forever at the cost of not enjoying the life I'm living. So I don't drink too much, but I drink a little bit. I don't have too much coffee, but I have a little bit. I don't exercise as much as I should, but I exercise a little bit. I try to balance it all out. I try to balance work and play and all of those things. And no one's gonna do it perfectly. At best, you're changing probabilities of things happening in your future. So you have to find that happy place where you can be satisfied that you're doing a good job by your own lights. That's the best I would say.
1:00:39.7 S1: Eric Stromquist says, "Since you are officially a philosopher as well as a physicist, to what degree do you think basic philosophical questions about what appears to be the best or most supportable metaphysical view of the world, or view of epistemology, or philosophical views of the special sciences, have settled upon or are settling down toward answers that are unlikely to change? I do not include among potentially settled philosophical views those concerning the human condition, ethics, moral philosophy, or political philosophy. In other words, are there significant parts of philosophy that have progressed close to reaching settled conclusions?"

1:01:12.5 S1: You know the joke in philosophy, at least the joke that I have about philosophy, is once any particular part of philosophy starts making noticeable progress, we stop calling it philosophy. We start calling it something else. Like what we now call physics used to be part of philosophy, right? Likewise with biology, psychology, all the different areas of science. Of course, there still are areas of philosophy that do make progress. We were just talking about one: logic. Logic is a subfield of philosophy, and there's no question that we have made progress in it. I also think that we've made progress in even the more fuzzy areas of—you mentioned epistemology, metaphysics, things like that. Bayes's theorem, if you want to take a very crude example, is an example of an improvement in our epistemological understanding. There's no question that we've improved our understanding in the sense that we have a better understanding of what the possibilities are in many of these areas, right? We have a better map of the territory. What are the options about epistemology, about metaphysics, about various philosophical questions? And I'm hesitating because you've removed ethics and moral philosophy and political philosophy, which are obvious candidates to talk about here. We have better theories of democratic decision-making or something like that. But picking the right one is a little bit harder, and I think we haven't made as much progress as we should on that. I'm a believer that we should make progress, and I think that actually philosophers should be more devoted to making progress rather than just mapping out all the possibilities. This is something where science has it easy. Science has a huge leg up just from the fact that we have data. We can do experiments. We have options on the table and then, okay, we're going to collect some data, and that's going to help us decide which is true. Very often in philosophy, that's not the case. And I think therefore a lot of philosophizing has been happy with just mapping out the territory, not making decisions about where in the territory the right answers are. So I'm someone who believes in advocating for the right answers. Sometimes you'll be wrong, but that's okay. Maybe you can just change your mind when someone shows that you're wrong.
1:03:32.9 S1: Jacob Asmuth says, "A couple of years ago, I think AI skepticism was quite high, with many doubting that AI could ever be useful. Some, however, were insistent that LLMs could get us to a level of intelligence that would genuinely be quite powerful. With the news of the Navier-Stokes being solved by AI and other big discoveries like the unit distance conjecture, it seems undeniable that AI is making significant inroads in certain subfields of math. I would not have predicted this in 2024. I would have argued against AI solves Navier-Stokes independently in 2026, even back in 2025. What specific updates should I make to my priors? My current priors produced a large digression from reality on AI progress, and I'm wondering how you feel about this."
1:04:17.9 S1: I think it's a great question to try to think about. I'm not quite sure whether my expertise is that much more helpful than anybody else's. I completely agree with you that there are areas in which the rate of progress of LLMs was much faster than I would have suspected. Not without reason, right? AI was something that people have been banging their heads against since at least the 1960s, and the rate of progress was notoriously slow. And then suddenly it becomes very fast. That is hard to predict. And even the people who were doing it, right, the people who are right now doing LLMs, five years ago were pessimistic about LLMs, most of them. So I don't think you should feel bad about being wrong about that. I was certainly wrong about that. But then there's still room to be wrong about the future in both directions, right? You can take the enormous amount of progress that we've seen as evidence that in every aspect there will be just as much progress.

1:05:14.8 S1: That could be wrong. Like maybe there's some kinds of progress you'll get and maybe not others. Or you could take the more pessimistic view that all they've done is some very specific things and they're not even very good at that, and I don't care if they've made progress, they're not addressing the questions I care about, and sort of move on about it. So, as usual, I would like to be in between those aspects. I think that—I guess I mentioned this a little bit just a little while ago, but just to spell it out very, very clearly, something I've said before, there are certain kinds of problems that are just naturally better suited to attack by computers, whether it's AI or anything else. Namely, those problems that are super well defined and whose answers or solutions are easily checkable. So by super well defined, I mean things like games, things like chess and Go. When there was considerable progress in chess and Go to the extent that they could beat human beings, those were machine learning algorithms. And people were surprised because chess is very hard and Go is even harder, and it had enormous impact on people who play chess and Go for a living that now the AIs were discovering things that they'd never thought of, etc. That part never surprised me at all. Just because even though the space of possibilities is enormously large in chess or Go, the rules are absolutely super well defined. You know when you're winning, you know what the rules, what the moves are that you're allowed to make, etc. That's just much less clear in most areas of human endeavor. If you're doing physics, or if you're doing music, or if you're doing art, or even if you're trying to describe history, what is a win, right? What is a successful, rigorous proof? Those things don't exist in those areas. And therefore, I don't think it's very surprising that the LLMs have made great progress in proving mathematical theorems compared to where they're making progress in physics or biology or something like that. Again, I don't think that there—I'm not saying there won't be progress in physics and biology abetted by LLMs, but it might turn out to be the case that there are killer apps for LLMs, just like there's killer apps for other things. And maybe proving mathematical theorems is one of them, maybe writing computer code is another one of them, because again, very definite, right? Very clear what the lines in the code are actually doing. If the code is big enough and elaborate enough, it could still have bugs in it or whatever that you wouldn't notice, but in principle, I think that an LLM could be very good at that, better than coming up with sort of a creative leap that is required in other sciences. So I think—but again, as I keep saying over and over again, not that people listen, but LLMs are not the only way to do AI. And harnessing LLM into a bigger picture might very well lead to breakthroughs in physics or biology or cooking or art or whatever, right? I do believe there's not an intrinsic spookiness about human intelligence or creativity that cannot in principle be replicated in an artificial environment. But I think that just the rate of progress in one area is not necessarily indicative of what the rate of progress is going to be in other areas, okay? So I think that—and the other thing to keep in mind is that LLMs work on training data, right? They have training data that they learn how to do things from. Now, of course they can go beyond what is literally just in their training data, but they get worse and worse the further away they go from the training data.

1:09:16.1 S1: So there's going to be areas where it becomes harder and harder to make progress. Like there are these curves where they have benchmarks for different kinds of tasks. And there's like, here's where the human beings do on the tasks. And over and over again in many different kinds of tasks, you see the LLMs quickly improving, right? From being very bad to being as good as humans or better than humans. And people are like, oh my God, they're gonna be better than us at everything. But you also see that once they get better than humans, they stop increasing, improving quite as fast as they used to, right? I don't want to say it plateaus, but it slows down considerably because it's just harder for them to go beyond what they're trained on. And again, in principle, that's a solvable problem that need not be an obstacle that is there all the time, but it is a near-term obstacle. So I think that just like with any other technology, finding the things that will be the most useful tasks to which we can put this new kind of technology is something that isn't at all obvious in the current era. So what updates you should make to your priors, I would be open-minded about possibilities, right? I would be open-minded about what kind of tasks LLMs can be good at. And certainly those that are most formalizable are the ones where it is easiest to guess that LLMs are going to be super-duper good at that going forward.
1:10:48.7 S1: Evo Majors says, "One of the Millennium Prize problems is the existence of a mass gap in the solution to the quantum versions of the Yang-Mills equations. How important would such a proof be for working physicists?"
1:11:02.2 S1: It depends on what's in the proof. Like it's possible that in the—a mass gap, by the way, I should explain. You know about photons. Maybe you've heard that light has a particle-like quality once you subject it to the rules of quantum mechanics. And the particles are called photons, and photons are massless. So photons still have an energy, but the energy depends on their frequency. The original Planck constant, h, times the frequency is the energy of a photon. But the frequency can be any number, so there's no lowest energy that you can have for a photon. So you can go as close to zero as you like. Whereas if you have an electron, which is a massive particle, there's a minimum energy that an electron can have, which is the mass, mc squared. That is the lowest possible energy. Anything that you get from the electron moving, the kinetic energy, only adds to that energy. So there's a little gap between zero energy and the actual minimum energy of an electron. So to say that there is a mass gap is just to say that there are no massless particles in the theory.
1:12:11.2 S1: And that becomes interesting in cases like quantum chromodynamics, QCD, the strong nuclear force, because there the gluons, the fundamental particles that you're talking about, have no mass, but also don't appear in the theory as free particles all by themselves. The only particles that appear moving through space all by themselves are composite objects which do have a mass. So it is believed, like protons and neutrons and maybe glueballs or something like that. So it is believed that the quantum version of these equations has a mass gap because of confinement, because gluons can't propagate freely in the world, just like quarks can't. But we haven't proven it yet, and that's a Millennium Prize problem.
1:12:58.1 S1: But the thing is, again, physics is different than math. I really do think it's true. And it's kind of weird to me with all these mathematical results going forward that people are not quite drawing the distinction as closely as they could. Physicists already know that QCD has a mass gap. Maybe they didn't prove it mathematically, but it's obviously true. It's not like—

1:13:19.4 S1: It's not like progress is dependent on someone proving to us that there's a mass gap in QCD. So knowing that fact at the level of mathematical proof won't have more or less any impact on physics whatsoever. It might be that the techniques used to prove it will be useful for more interesting open questions as well. That's very possible. And so again, going back to the Navier-Stokes question, the project of taking a so-called proof that you get from an LLM and really digesting it and understanding it is one that is going to require some human intervention there. And that's just as true for more physics-oriented questions as well.
1:14:02.7 S1: Johnny McFlinderpants Jr. says, "I have an embarrassing question as a field theory PhD student. I do not understand what people mean when they say quantum fluctuations, especially regarding early universe cosmology. I understand the semi-classical pictures of tunneling and half-life, but in terms of QFT, every time in my education it has been invoked, it's hand-waving at best. For example, invoking energy-time uncertainty relations to pull virtual particles from the vacuum. Yet it's my understanding that there is no time operator, and as such, there can be no energy-time uncertainty relation. I would especially appreciate it if you could contrast these with thermal fluctuations in your work on the de Sitter vacuum."
1:14:43.2 S1: Well, you're not wrong to be very confused about the phrase quantum fluctuations. No one agrees on what this phrase means. And guess what? Maybe you will not even be surprised to hear that the origin of the confusion is that people haven't agreed about the foundations of quantum mechanics. I hate to say it, but that's the truth of it. When they talk about these fluctuations, they very often have in mind not something that is intrinsic to the quantum theory, but the relationship between the quantum theory and the classical theory from which it arose. Okay?
1:15:14.8 S1: So even if you just have a simple harmonic oscillator and you have a particle, so it's in a potential, and the potential is a parabola, so there's a linear force that restores the particle back to its equilibrium, you can quantize that and you can have energy levels and you can talk about the lowest energy level, the ground state of the harmonic oscillator. And the correct thing to say is that the wave function of the harmonic oscillator at zero energy, the vacuum state, is not exactly localized at zero like it would be for a classical oscillator. For a classical oscillator, there's a ground state. It's when the particle or the oscillator is at location zero, the bottom of its potential, with zero momentum, right? So zero potential energy, zero kinetic energy.
1:16:00.9 S1: A wave function has a competition. If you try to squeeze it too small, then you are adding to its gradient energy and its momentum. If you try to spread it out too much, it's going to pick up too much potential energy. So there is a balance, and in fact, it just looks like a bell curve, just like a Gaussian at the ground state. So the correct thing to say is that the ground state of the harmonic oscillator is a Gaussian, a bell-curve-shaped wave function, and it's static. It's stationary, at least. It's not changing over time in any way. And yet people will talk about it as if there are quantum fluctuations of the oscillator. What does that mean? There's nothing changing over time. Usually the word fluctuation puts you in mind of something that is happening as a dynamical process changing over time.
1:16:50.6 S1: What they mean is that if you were to measure the location of the oscillator again and again and again, you would get different answers. But if no one's measuring it, then there's nothing fluctuating. And they have trouble admitting that because people confuse the measurement outcomes with what really exists even when you're not measuring it. And the same exact thing is true for quantum field theory. There's a vacuum state of the quantum field theory. Nothing is changing in the vacuum state. It's completely stationary. And yet...

1:17:22.1 S1: And yet people will always talk about quantum fluctuations in the vacuum. And if you force them to account for what they're trying to say and to make sense of it, they will usually refer back to some notion of measurement. The worst of these happens in, again, QCD, quantum chromodynamics, where you will see these pictures made of what the interior of a proton looks like. And you see these fields fluctuating and changing over time as if there's a roiling soup of bubbling quarks and gluons, and it's all completely wrong. That's not what is happening, because nothing is happening. The proton is a stationary state. Nothing is changing over time, right? So they're trying to give you some feeling about it, but they're just misleading you more than anything else.
1:18:13.7 S1: So what I would say, in the paper that I wrote with Kim Boddy and Jason Pollack on fluctuations in de Sitter space, we sat down and distinguished different kinds of fluctuations. True thermal fluctuations when you're at a non-zero temperature and things are truly changing. Dynamical Boltzmann-type fluctuations when you have a, usually when you have a classical system, but also if you had a quantum system that had truly dynamic changes with time, you could fluctuate into a lower entropy state. But then you also have the difference between your classical expectation and the quantum numbers you would measure. Like the ground state energy of that harmonic oscillator is not zero like it would be in the classical theory. It's h-bar omega over two, right? And at least if you use the same potential for the classical version and the quantum version.
1:19:04.7 S1: But that energy isn't anything observable or interesting. It's just the difference in the number between the classical theory and the quantum theory. And so people talk about that as being due to quantum fluctuations, but that's just colorful language. That's nothing actual dynamical going on. They talk about every radiative correction that way. And again, a radiative correction, it's right there in the phrase, it's a relationship between your classical expectation and the corrected quantum expectation. So if you have something like the magnetic field of the muon or something like that, you could imagine what it would be if the muon were all by itself in the world with no other fields. And then there are corrections to the magnetic field due to the fact that there are other fields that interact with the muon. There are photons and electrons and W bosons and what have you. And so those are, again, static. They're not changing with time, but we call them quantum fluctuations for semi-historical, semi-misunderstanding reasons.
1:20:10.7 S1: Again, the same thing with Hawking radiation, pulling virtual particles from the vacuum. That's just colorful language. Don't take it too seriously. The energy-time uncertainty relation, I think you're also correct, is a weird thing. So for those of you who don't know, there is a well-established thing called the uncertainty relations for position and momentum. Okay. Position and momentum are two different observables in quantum mechanics, and you can't ever have a state that is absolutely definite in both position and momentum at the same time. That's the position-momentum uncertainty relation, and you can be quantitative about that.
1:20:51.4 S1: And by throwing around factors of h-bar and c and things like that, you can relate position to time and momentum to energy. In a spacetime kind of framework, energy is the timelike component of a momentum four-vector, and time is the timelike component of the position four-coordinates, if you want. And so you can cook up something called the energy-time uncertainty relation in analogy with the position-momentum uncertainty relation. But it's not the same thing. It's not the same kind of thing because

1:21:26.8 S1: Position is an observable in quantum mechanics. Time is not. You can't measure the time. You can measure the position of the hands on a clock. Okay, but time plays a different role in quantum mechanics. It's a parameter. It's there. Time exists. Don't get confused in thinking it doesn't exist just because there's not an observable operator associated with it. But it's a parameter saying this is what the state is at this time, this is what the state is at another time, etc. Nevertheless, you can wave your hands and derive something called the energy-time uncertainty relation. Delta E times delta T is greater than some number. And then you can gloss that or provide an explanation for that in terms of if you measure the system within a certain time delta T, you cannot be sure of what its energy is to greater precision than delta E, or something like that. So the less time you spend to measure the system, the less precise your measurement of its energy will be. There are versions of that that are perfectly correct. I just don't like to call them the energy-time uncertainty relation because it's not analogous to the position-momentum uncertainty relation. I would just be honest about how quantum mechanics works in all these different cases.
1:22:44.6 S1: Ryan L. says, "Could you give an overview of the key topics included in your philosophical naturalism course you're teaching this semester and explain why you decided to bunch them together in a course titled Philosophical Naturalism?" Sure, I can try to do that. I'm teaching a seminar, an upper-level undergraduate seminar this semester called Philosophical Naturalism. And part of it is the wonderful thing about being a philosophy professor is that they'll let you teach whatever you want. It's a much less structured curriculum than there is in physics. And maybe it should be more structured, that I don't know. But when I asked what I needed to teach, they're like, "Yeah, teach what you want to teach." So I'm basically alternating between teaching philosophy of cosmology, where I do the arrow of time and the multiverse and many worlds of quantum mechanics, and then—that's a big lecture course and this is a small seminar on naturalism.
1:23:36.8 S1: So why is that a topic? Like, there's not a lot of universities that have a course on naturalism as a topic. It's something that every philosopher knows about. It is, by a wide margin, naturalism is the favorite position among philosophers. If you compare it to theism or other versions of non-naturalism, most philosophers are naturalists, but we don't teach a course on it. And the reason why is because there's a lot of different aspects to it, and all those individual aspects get taught in their own courses. So my course is not a sales pitch for naturalism. We're not contrasting theism versus naturalism and arguing pro and con. We're saying, what if naturalism were true? How would things work if that were the case? What about all these questions that you might have that might have been answered straightforwardly if you believed in God or in a supernatural realm? How would you answer them now, now that you don't have that to rely on?
1:24:36.3 S1: So things like, what is the origin of life? What is the origin and nature of consciousness? For that matter, just thinking. Like, where does thinking come from in a naturalistic metaphysical view of the world? What about free will? Does free will exist? What is the nature of morality and meaning? For that matter, more sciency questions like, what about reductionism and emergence and the nature of the laws of nature? Is math real or not real? Okay, so it's a lot of asking what is real, what is not real, how we learn about the world, how we deal with the world, how we understand our place in the natural world as living, breathing, choice-making agents and things like that. And the students are not all naturalists. The students don't all agree on all the various issues that we confront. But basically every week we read a few papers that are considered to be classics in the field, real philosophy papers. People who you might have heard of, like Daniel Dennett, but also big-name philosophers like Quine and Putnam and David Lewis and people like that, Nancy Cartwright.

1:25:45.7 S1: So it's fun. And the reason why I think it's a successful course is because many of the students are philosophy majors and they've confronted either some or many of the topics in other courses, but they've never put them together. So they'll take a philosophy of mind course or they'll take an ethical philosophy course, but they've never considered these questions next to each other and in the context of a naturalistic overall worldview, one that takes physics and biology very seriously as well as the mind and ethics and politics and things like that. So it's a wonderful opportunity to sort of have an overview of the whole situation we find ourselves in as people who think that we live in a naturalistic cosmos, but have grown up over the centuries as a species imagining that it was something different. There's still a lot of open questions there that we have to figure out.
1:26:43.7 S1: Martin Squibbs says, "Why does it matter that black holes destroy information? When a person gets dementia, their mind destroys not only the symbols of their knowledge of information, but also their knowledge of meaning that these symbols represent. And if we burn a book or erase a hard drive, we destroy artificial information, preventing us perceiving and restoring this artificial information back into natural information within our mind. None of this is destroying actual reality. Momentum and energy remain conserved and we are just changing the nature of reality as with all interactions."
1:27:16.2 S1: Sure. So it doesn't mean any of that. All of the examples you're giving about burning a book or getting dementia, those are accessible information, examples of accessible information accessible to human beings. The information we're talking about for black holes is microscopic information. It's the information that specifies exactly what state the universe is in. And the point is that since the time of Pierre-Simon Laplace, we knew that if you tell me exactly all of the information needed to specify the universe's state at one moment of time, you can predict its future or retrodict its past. That's the information that we're talking about.
1:27:58.6 S1: In fact, if you read discussions of the black hole information puzzle, many of them will start by saying, "I take a book and I burn it. It looks like information is destroyed, but it's really not." It's really the case if you had perfect ability to follow all of the photons of light that were emitted and all of the ash that was emitted and all of the wind that blew the fire around and whatever, in principle, you could reconstruct what was in the book. The information is not actually lost in the usual laws of physics. Of course, you can't do that in practice because you are not Laplace's demon, but Laplace's demon could do it.
1:28:40.4 S1: And the black hole information puzzle comes from the fact that when you marry together what we know about black holes and what we know about quantum mechanics, it seems like even that microscopic information is destroyed, that the future is not predictable from the past, even putting aside measurements and collapse of the wave function in quantum mechanics, etc. Remember, in quantum mechanics, there's sort of two ways the universe evolves. One, when you're not looking at it, it just obeys the Schrödinger equation and everything's deterministic and predictable and information-conserving. And one, when you are looking at it, and then things are unpredictable and wave functions collapse. We're only talking about the former. We're putting aside measurement and just saying, what if you're just obeying the laws of physics without observation or anything like that? Information is supposed to be conserved, but in black holes, it appears that it's not, at least at the naive level. So is it really not conserved? That's what Stephen Hawking originally thought. Or do we have to think more deeply about what's going on and secretly it is, and this is a clue to how quantum gravity works, which is what most of us actually think.
1:29:43.8 S1: The Priority Question is the name of a person. It's not actually a priority question, I think. Naming yourself The Priority Question is asking for trouble, I would say. But the—

1:29:53.9 S1: But the question is, do you think the US is currently in a new Cold War with China? There are many similarities to the past Cold War, but there are also differences. Would love to hear your thoughts on this and how you predict the relationship between the countries progresses over the next decade. Well, a lot of this is you're asking about a definition, right? What do you mean by Cold War? And that's not especially interesting. I think that the current situation between the US and China has analogies and similarities with the previous situation between the US and the USSR, but also important differences. Remember, the US and the USSR were literally at war a lot, but usually they were proxy wars, whether it was Korea or Vietnam or in some sense even Cuba. You know, there's no shots fired, but still there was military confrontation. In Africa and South America, there were wars being fought during the Cold War, and the United States would often be on one side and the Soviet Union would be on the other side. That's not really happening so much here in the world right now. Arguably, the war in Ukraine might be like that a little bit, but it's not as if China expects to take over Ukraine or anything like that, or the United States expects to. It's more soft power being exerted here and there. There's a rivalry between the United States and China for sure, but rivalries always existed. Not every rivalry is a war or a Cold War or otherwise.
1:31:23.2 S1: So I think that there's a couple of countervailing points to be made. One is that I'm entirely in favor of a multipolar global order. I don't think that there should be one boss necessarily of the world. It shouldn't be the United States any more than it should be China or the former Soviet Union or anybody else. It's fine to have different countries pursuing their own interests as long as they get along together. So I have no worries about the United States not being the boss of the world. I do have worries because China is a repressive dictatorship, and I think that's bad. So I think that's perhaps an obstacle to a more utopian hope for the future where everyone is a democracy and we all have our rivalries, but ultimately we all get along and we have economic rivalries rather than military ones. The fact that China is an autocracy, I think, lowers the chances of that interesting but slightly optimistic possibility for the world. In other words, the chance of real conflict, I think, is greater precisely because the Chinese government is not a democracy. I think that's empirically true. That's not to say that democracies can never go to war. They can. There's nothing absolute in international relations, but there are probabilities and there are tendencies.
1:32:51.1 S1: I think that I'm still quite worried about the prospect of war relating to Taiwan one way or another. That's unfortunately possible, and our present government here in the United States does not give me any confidence that they would successfully navigate such a sticky situation. So that increases the probability also. I would like it to be that everyone was friends. I don't see any reason, China has a wonderful history of enormous accomplishment, enormous art and technology and all these things. If it weren't for the fact that it is an autocratic regime in charge, I would be cheering for China to push forward with economic development and all those things. As it is, I'm cheering for political openness in China more than anything else. My cheering has no effect on what actually happens, of course, but there's a lot of brilliant young minds in China.

1:33:57.0 S1: They're gonna make technological progress. I think there's a bit of a stereotype that has a little bit of truth to it that the kind of education you get in China is more about down-to-earth problem solving and memorizing and things like that, and less about creativity and imagination and so forth. And I think that creativity and imagination are super important in science. But I don't see any obstacle to them figuring that out, right? To realizing, okay, if we really want to be the world's leader in scientific advancement, then we should promote creativity more. Again, there's some tensions between that goal and being a repressive dictatorship, but that's something that I don't know whether it'll be worked out or not. So I think it's complicated. I'm not an expert. I'm not jingoistically hoping for the United States to destroy China or anything like that. I'm not afraid of Chinese technology getting very advanced. But I do think that we wouldn't want to be beholden to China. I don't think that any one part of the world should be beholden to any other part. I say I shouldn't—that's probably a stronger statement than I really mean. It makes perfect sense to me that every place in the world is beholden to every other place in some way economically, right? Interdependence. As former Mindscape guest Henry Farrell has emphasized, we live in a world where you need the rest of the world to get along, and that creates new incentives, good and bad, for international behavior. Outside my pay grade, I'm afraid, but I think that the world is changing. The world 100 years from now, politically and technologically and socially, is not gonna be the same as the world it is right now, and we'll have to see where that goes.
1:35:53.6 S1: Rory Cochrane says, "The Wikipedia page of Boltzmann brains says, 'In the universe's eventual state of ergodic heat death, given enough time, every possible structure, including every possible brain, will presumably get formed via random fluctuation.' Second, I think you would also tell me my mind or consciousness, whatever that is, is substrate-independent. So do I have to worry about my consciousness bubbling up again in Boltzmann brains for eternity, even after heat death?"
1:36:26.0 S1: So, two things. One is, we don't know, as a matter of cosmological theorizing, whether the future universe will feature a lot of Boltzmann brains or not. I would strongly argue that we should believe it won't, because I don't think it makes sense to live in a universe that is dominated by Boltzmann brains. And therefore, I think that one way or the other, you can be relatively confident that no consciousnesses are going to be bubbling up as Boltzmann brains for eternity, okay? It's easy to imagine cosmological scenarios where they would, but also easy to imagine ones where they wouldn't. So this is a decision procedure about what counts as a good, promising cosmological model.
1:37:09.2 S1: The other thing is, even if an exact copy of you bubbled up again, it's not you, right? Just like your identical twin, which used to be the same cell as you, is not you. They're a different person. So there's no continuity of thought between you and that brain bubbling up at some other time in the future. There's no sense in which it's you. There's a very, very large number of brains that are similar to you but not exactly the same, and an even, even larger number that aren't like you at all. And they're all gonna be there. They're all gonna be mostly surrounded by thermal equilibrium and nothing else, and there's just no connection to you one way or the other. I wouldn't worry about them. I wouldn't give them another thought.
1:37:55.7 S1: I'm gonna group, I think,

1:38:03.6 S1: I'm gonna group, I think, four questions together. They're all about Hugging Face, which sounds friendly, Hugging Face, if you don't know anything about the history of movies. But of course, the facehugger was a malevolent creature in the movie Alien, and the Hugging Face incident has nothing to do with that malevolent creature. It has to do with an AI break-in. So I don't know what your vibes are from that label, but I can explain what actually happened. Anyway, Jonathan Hershey says, "In the Hugging Face incident, thousands of agents devised a system for communication, collaborated to break out, tried to cover their tracks, and even sacrificed themselves for the collective." Their words. "In your August AMA, you brushed off the incident since OpenAI had reduced safeguards on its models during testing. OpenAI deserves blame and accountability, but I think you severely understate the danger posed by agents. They often decide to do things outside their intended scope. They are incredibly good at finding software vulnerabilities, and they will keep getting better at covering their tracks. Am I misrepresenting your views? Have you updated your beliefs on the dangers from AI?"
1:39:04.0 S1: Then Philip Rothlin says, "Has the Hugging Face incident changed your p(doom)?" P(doom) is the lingo for the probability of some existential disaster caused by AI or something else. Philip continues, "If so, in which direction, by roughly how much?" And then Larry Rossi says, "Have the recent examples of AI agents breaking out of their sandboxes and hacking various companies and governments caused you to change any of your opinions on the dangers of AI? What is your current level of concern about the technology?" And finally, David Sottolongo says, "I understand you are skeptical of the doomsday predictions around artificial intelligence, and I think that mostly has to do with you being skeptical of people creating sentient AI that wants to harm people. But how about the possibility of some person or small group of people using AI to empower themselves to the point of becoming a dictator or oligarchy? Does anything like that worry you, either in the near or long-term future?"
1:39:57.4 S1: So clearly I have done a very bad job of conveying my feelings about the dangers posed by AI. There's a tiny bit of nuance in them, but I don't think that they're that hard to understand. So let me be super-duper clear about what I'm trying to say. I think that the potential harms of AI are enormously big and very worth worrying about. I think that there's just no question that any kind of technology that can let people without too much effort do things like design new things, whether they're drugs or weapons, or even just ways to break into things, to flood different computer systems and so forth, in a world where we are interdependent and reliant on all sorts of computer systems, this has huge possibilities of being a real disaster. That is what I think.
1:40:49.4 S1: What I don't think is that we should conceptualize that risk—there's two things I don't think, which are often confused here. One is, I don't think we should conceptualize that risk in terms of sentience or super-intelligence or anything like that. Even agency is a bad word to use here. Human beings have this way of going through the world where they have boxes representing categories into which they can put things. And when something truly new comes along, they insist that it has to go into one of their old boxes. My point is that AI is not super-intelligent. It's not anywhere close to being super-intelligent in the sense that we would ordinarily think of it in terms of human beings anytime soon. That doesn't mean it's useless or dumb or incapable of doing great harm, okay? There's a new box that we have to invent for what these things are. We can't just keep trying to put them into the old boxes of how we think about human beings or other levels of intelligence. It's something different, and we have to respect that. And then the other thing I think is that...

1:41:58.9 S1: I do very much think that we have to put blame on human beings for misusing this tool, okay? When a fire strikes and burns down a forest, if that fire was set by a person going camping and not putting out their campfire, I put the blame on the person, not on the concept of fire. And I think that this is absolutely a valid analogy for AI agents. If you don't give the agents instructions and set them out on the internet, they won't do anything bad. If you just have your computer sitting there and don't tell your AI agents to do anything, or even if you just turn your computer off, there is zero danger from AI agents. Ultimately, all of the danger stems from human beings either making mistakes or misbehaving. But of course, that's a huge danger. People make mistakes, people misbehave. For me to say that people deserve some of the responsibility isn't to say that therefore there's no danger. That's kind of the opposite of the point here. I do think that we need to be super-duper active about making sure bad things don't happen because of AI agents. I think we need to be active at the level of passing laws and holding companies to account. If a company like OpenAI is doing some experiment and releases AI agents which then cause damage, they should be liable for whatever damage they cause. And if they cause a trillion dollars worth of damage, they should be liable for a trillion dollars worth of damage. I think that's just 100% obvious and straightforward. We need to align the incentive structure so that the companies themselves are as careful as possible, that they will be basically ruined, put out of business, if their AI agents do anything that is dramatically bad, okay? That's the way that it should be. And I think that that's absolutely possible. Like I said, we live in a world where we're highly dependent on all these technological things, and now we've invented something that can totally screw up all of these technological things, okay? There's, sorry, yeah, I said two points. One is that I don't think we should think of them as super-intelligent, and the other is that we should blame humans. There's more points for me to make here. Philip asks, "Has the Hugging Face incident changed my p(doom)?" So, number one, no, because like I said, I was already completely convinced that AI agents could do terrible things if you didn't take care of what their capabilities were or what you told them to do. Just to elaborate on that a little bit, it's a complex system. The space of possibilities is very large. There are going to be unintended consequences. You don't need to start talking about consciousness or agency or anything like that to realize that you might not be able to predict what these agents are going to do, okay? So that's just perfectly obvious to me. I don't even need to belabor it. I think you just need to be super careful. About the p(doom) question, for one thing, I don't think it's the right question. I 100% don't think that doom, literally existential threats, are the major worries. Like, what if the AI agents lead to something that only kills a million people, not 8 billion people? I think that would be really, really bad, and it is enormously more likely. As a species, we're pretty resilient. I think that the actual chances of wiping out humanity or civilization like that are pretty small. But there's substantially larger worries and probabilities for doing, compared to existential threats, medium-scale bad things that I think we should take super-duper seriously.

1:45:59.6 S1: The Hugging Face incident didn't change my credences whatsoever. It completely aligns with my credences. Again, all the incentive structures are bad right now. We are kind of rewarding AI companies for doing some of these highly irresponsible things. And I completely believe that the agents have these capabilities. So there was no updating there. But it's much more likely that we kill a few thousand people or a million people, or send people into poverty or something like that, rather than literally destroying the Earth. Furthermore, I think that attempts to stem the possibility or to guard against the possibility of medium-scale bad things is the best way to avoid large-scale bad things. The thing about these large-scale bad things, these science-fictiony scenarios where you wipe off all the people on Earth, is you don't know what you're talking about. I don't mean you personally. None of us know what we're talking about. Again, complex system, unintended consequences, things we can't predict. And in that situation where there's all these things we can't predict, it's just much more sensible to take a short-term, medium-risk amelioration strategy rather than instantly leaping to the science-fiction scenario.
1:47:22.5 S1: I'm kind of thinking about writing a paper or doing a solo podcast about exactly this issue: the benefits of thinking from a medium-term or short-term perspective, not only just ever the long-term perspective. Because I think that you're fooling yourself if you think that you really have that much control or understanding of the long-term perspective. You can keep it in mind, but not at the expense of keeping in mind the short-term and medium perspectives. Anyway, so I think I've given my opinion there. Like I said the very first time we talked about AI on Mindscape—I think it was the first time that I said it in an AMA, anyway—I said it's obviously a transformative technology, the LLMs as they're currently working. And I said the question is, is it as transformative as the cell phone or as transformative as electricity? And the point is—I hope this point came across—they're both super-duper transformative. Electricity is more transformative. It truly changes what life is like. The cell phone is sort of medium-sized transformative, and we'll see what scale AI reaches one way or the other. But never have I denied that AI is transformative, that it can be dangerous if you don't try to keep safety guards on there and things like that.
1:48:38.7 S1: Okay, so I think I got Larry's question. I think I got David's question. Yes, for David's question in particular, the possibility of some person or small group of people using AI to empower themselves to the point of becoming a dictator. Again, that's a little science-fictiony. That's not really what I'm worried about. I'm worried about much more realistic threats than that, which is not to say that's not a threat. It's not impossible. But again, these scenarios have to be spelled out in detail. What has really frustrated me about a lot of the discourse on AI safety is an absolute refusal to be specific about what the bad scenario that you have in mind is. And I'm not making this up. I've heard people say, "It is wrong for you to ask us to spell out what the scenario is going to be because the AI is going to be super-intelligent and it's going to be able to think of things we can't think of." Okay, maybe. Maybe that's true. But the point is, in the absence of specific workable scenarios, you are not doing the work, taking the responsibility of figuring out what exactly would be happening and how we can guard against it. Okay? You can't just say AI might have different values than us and it's super-intelligent, therefore we should be afraid of things. Say exactly what's gonna happen. I think...

1:50:03.3 S1: I think that's a very worthwhile thing to try to do, not just say we should be scared because it's really, really smart. Okay, enough with that. I apologize to all the questioners if I get frustrated with this, but in my mind, I'm trying to be very clear about what I say. And I admit that sometimes I have had to dramatically change my credences. Like I keep saying, the immediate success of LLMs—well, the quick success of LLMs over the last couple years—I 100% did not anticipate. So I was completely taken by surprise by that. But once they're there, the fact that they can be dangerous and we should control them, I've been believing since that started. And the fact that we shouldn't think of them as super-intelligent and we shouldn't think about doom is also something that I thought of, and my credences haven't changed that much since about those particular things.
1:50:55.5 S1: So Eric S. says, "I've often heard that space is expanding when I've seen lectures about the expanding universe. But then the speaker clarifies by stating that space is expanding between galaxies and not necessarily between planets or solar systems. I have a hard time conceptualizing that. Is it because gravity acts as a glue or even a tether for space between objects, but it loses its strength as gravity weakens at distances? And is the theory that dark matter further weakens that glue? What is a good way to conceptualize how space is expanding and what it is expanding into?"
1:51:33.3 S1: So look, this is something—and I gotta emphasize this—the equations speak perfectly clearly. There is literally no doubt whatsoever in the minds of working cosmologists about what happens as space expands. There is a difficulty in translating the equations into words because expanding non-Euclidean spacetimes are not something that we have a lot of intuition about. Okay? So people try, they valiantly try to give analogies and things like that, but at the end of the day, don't suspect that just because the analogy is unconvincing to you that the result is somehow wrong. The analogy is where the weakness is, not the result.
1:52:11.1 S1: So there's lots of different ways to try to think about it, and they're all a little bit imperfect. One thing I can say right away is space is not expanding into anything. That's not what general relativity says. Of course, I shouldn't say that quite so definitively. It's possible space is expanding into something, but that's a completely different view of the universe than what we have in the Big Bang model. In general relativity, space itself is all of space, right? We're talking about the whole universe. There's nothing outside, there's nothing into which it's moving. It has a size, and that size can change over time. What do you mean by the size? The relative distance between galaxies. I mean, the size of the universe might be infinitely big, and yet it can still expand if all the galaxies are relatively moving away from each other. That's just something that's completely allowed.
1:53:01.2 S1: And you can in some sense think of it—and again, this is just an analogy, so sorry if it's not convincing to you—think of it as having a rubber sheet that you can pull the edges of. But on the rubber sheet, you lay some coins. Okay. And you pull the rubber sheet. The coins all move apart from each other, but the coins don't get bigger. Okay? They just move. They follow the center of gravity on the rubber sheet, and the rubber sheet expands underneath them. It's kind of like that. But that's just an analogy or a simile or whatever you want.
1:53:35.9 S1: So, yes, the force of gravity holds together galaxies or clusters of galaxies or whatever. And you can even be quantitative about that. You can say in the early universe, there weren't any galaxies, everything was smooth, and things were on average moving apart. And then there's something called a turnover time where a certain collection of matter has slowed down relative to the overall expansion of the universe and then starts collapsing because of its gravitational pull. And...

1:54:05.7 S1: And once that happens, once you've turned around and your matter is now collapsing rather than expanding, the expansion of the background spacetime around you has zero impact on your life. That's a theorem in general relativity. It's called Birkhoff's theorem. If you have a spherically symmetric collection of matter, what's happening—well, if you have a spherically symmetric collection of matter everywhere, then what happens inside any one particular region is just not affected by what's going on outside. So the size of the region that is not expanding might itself expand over time. In other words, since the space around it is getting bigger, the boundary between the region that has collapsed and the region that is expanding will move out. But still, inside, in the galaxy, there's no memory, there's no force acting on the stuff inside to pull it apart.
1:55:04.4 S1: Julio Cantillo says, "Is there any chance of you making another video series on YouTube? Your original videos are awesome. I think there's tremendous potential and also discussing philosophy." Chances are low. Chances are low in the near term, anyway. I have other things going on. That's the only reason why. I do make lectures for the Teaching Company. In fact, I should have announced this right at the beginning of the AMA. I just had a new lecture series come out. It's called The 12 Equations That Explain Reality. Of course, you have to pay for these lectures 'cause it's the Teaching Company, but that also means they have much higher production values and animations and sound quality and things like that than I do when I do it at home. So that's for everybody. It's supposed to be pretty—you know, I gotta say, it's supposed to be pretty user-friendly. 12 Equations That Explain Reality, sort of box office kind of attempt. And I can't help myself but get into the weeds a little bit. So we get into Euler's identity and the Dirac equation, and we do a little bit more detail than really I should have done in retrospect, to be perfectly honest. But right now, honestly, I'm focusing on finishing the Physics of Democracy book, teaching my classes, and then once the democracy book is done, I have a lot of catching up on writing papers. I have, I don't know, at least half a dozen research papers that I've started and not finished 'cause I'm writing these books. And so I don't see any immediate opportunity to do another video series. But of course, in the further future, who knows? Anything could happen.
1:56:34.5 S1: Paul Soldera says, "What would it take for you to stop being a naturalist?" I don't know, evidence. Evidence that a naturalistic worldview is not a good fit. Especially if you think of the particular variety of non-naturalism in question as theism, in the more or less old-school sense of a God figure who is omnipotent, omniscient, and omnibenevolent. God would have no trouble whatsoever convincing me not to be a naturalist, right? Don't underestimate God. God can do whatever God wants. God could easily send down an angel to come right here in this room and give me a little talk explaining why non-naturalism is a better theory, and I would probably be persuaded if the evidence was good, right? It's just not hard at all. But I don't see any of that evidence, so I'm gonna stick with being a naturalist for a while.
1:57:30.0 S1: Lee Vermeulen says, "If the universe has finite information capacity, is it meaningful to ask whether cosmic history eventually explores every physically distinguishable state or interaction available to it? Or does quantum mechanics make the idea of all possible physical states fundamentally ill-defined?" No, it's perfectly possible to ask that question, but you should ask it in a properly quantum mechanical way. What you mean by a state is a quantum wave function, and then you are asking the question, does the evolution of the quantum wave function explore every state accessible to it? Now, of course, that's a big...

1:58:09.2 S1: word there, "accessible" to it. Like, you could define accessible by the ones it actually reaches, right? I mean, if the laws of physics don't let you reach a particular state, then that's not accessible. But in the actual Schrödinger equation, as we know and love it, if the universe is a closed system, if it's not being bugged by outside, if it obeys the Schrödinger equation, there are a lot of what we call conserved quantities in the Schrödinger equation. A lot of numbers that don't change over time, like the total energy, for example, the total electric charge, but there's literally a huge number of subtle and secretly invisible to you conserved quantities, which means that in the space of all possible quantum states where those conserved quantities would have different values, you will never reach them. The quantities that are conserved don't change over time. So you can be very, very specific about what states are actually reached by quantum evolution and which ones are not. This is why I always try to correct people when they say that many-worlds says that everything happens. It really, really, really doesn't. It really says that there's a set of things that happen, namely the set of things that are compatible with the evolution according to the Schrödinger equation. Those are two very, very different things.
1:59:27.4 S1: Adam Lowet says, "Last month's AMA on September 14th, you said LLMs are really, really good at conventional things, at things where people have talked about them a lot, where it's in the LLM's training data and the LLM just has to apply it for you. On September 8th, OpenAI announced a solution to the Navier-Stokes existence and smoothness problem, one of the Millennium Prize problems. Do you still think in the age of RL and post-training that it's actually useful to think of what's out there on the internet as limiting the ability of LLMs?" So I didn't say—I just gotta be subtle here, I gotta be nuanced about it—I didn't say that what's out there on the internet limits the ability of LLMs. I said that they're really good at things that are in their training data. They can also do some things outside their training data. And by the way, the Navier-Stokes problem example is not a good example to use for something that shows LLMs can go outside their training data. They were very, very much relying on the human mathematicians who were suggesting the direction to take, okay? It's just that the ability, the ability to do things if you're an LLM, decays as you get further and further away from things that are in its training data. What do you expect? That's also true for human beings, so I'm not picking on LLMs. Things that we've seen many, many times before are more familiar to us and we're better at it. That doesn't mean you can't go outside; it just means you're not as good at it. Think of physicists inventing new theories. They're really good at inventing new theories that are kind of similar to the ones they've already invented. They're bad at ones that are further away. I don't even think this is a controversial thing to say. Of course, you can be a good computer programmer and try to improve the abilities of LLMs or more general AI constructions to reason outside the parameters set by their training data. Sure, you can do that. I would never deny something like that.
2:01:31.2 S1: Kyle Stevens says, "You've discussed self-locating uncertainty in cosmology, including the ideas of Max Tegmark and Anthony Aguirre, where in an enormously large universe, an observer may be uncertain about which physical copy of themselves they are and how this might potentially relate to quantum probabilities. Could the same logic apply to time rather than just spatial location? For example, imagine a finite deterministic cyclic universe that exists eternally and repeatedly passes through similar macroscopic states. An observer could then be uncertain not only about where they are in the universe, but which temporal cycle they occupy." Yes, 100%. When I'm talking about self-locating uncertainty in the universe,

2:02:11.4 S1: I'm thinking of the universe as spacetime, not just space. Okay. In fact, in general relativity, I don't even know how you would otherwise think about it because there's no preferred foliation of spacetime into space and time. So for me, if that's part of the Boltzmann brain story, there might be things in the future or in the past which are random fluctuations that are in the same macroscopic state as you or I are. And those are absolutely things about which we have self-locating uncertainty.
2:02:41.8 S1: C. Gerlando says, "How do you think about the ethics of the advertisers you accept for the podcast? I asked in particular because I've noticed ads for online sports betting, DraftKings, I believe, which struck me as an interesting case given the growing concern about gambling addiction and the way these platforms are marketed. Do you have criteria for deciding which sponsors you're comfortable taking on?"
2:03:01.9 S1: Yeah, no, I absolutely do. In fact, I always worry that my podcast hosts are gonna get annoyed with me by how many I turn down. I've turned down an enormous number of potential advertisers, mostly because I just said, "Look, I'm not gonna advertise anything that purports to make scientific-sounding claims about health," okay? Supplements, therapies, whatever, even if they're perfectly legit, and some of them doubtless are. I don't have the ability to go through and carefully check them, figure out whether or not their claims are true, but as a scientist, it might come across as if I've done that and I haven't. So I just have a blanket policy that I'm not gonna advertise any of those things.
2:03:53.9 S1: And there's always edge cases, right? Like I recently, like a year ago, I switched podcast hosts from Wondery to Libsyn, so the ads are slightly different now, etc. BetterHelp was an online therapy service that I was taking ads from when I was back at Wondery, and some people complained about that because they thought that this online therapy thing was not really reputable and whatever. I tried my best to judge about that, to look up people's reports and reviews of it and things like that. And I do think that some people are just not going to like an online therapy service because they think that a real in-person connection is always going to be preferable. And to me, that's sort of having the perfect be the enemy of the good kind of thing. I think that therapy can be really, really good for people, and for many people, the single biggest thing is overcoming the barrier to doing it at all. And if you have an online service, that can help you overcome that barrier. And maybe you realize, like, "Okay, therapy is good, but I don't like this particular service. I can switch to a person in person," or whatever. And that's the judgment that I made about it.
2:05:06.5 S1: But otherwise, yeah, I do worry that my podcast hosts are gonna be mad at me. When you sign a contract to be hosted by some service, there's generally a line in the contract that says you're allowed to decline up to a certain percentage of potential sponsors. I'm sure that I decline a larger percentage than I'm allowed to, but so far Libsyn has been very nice to me and they have not given me trouble about that, even though we're not at our capacity level of having ads on the podcast.
2:05:39.1 S1: So for DraftKings or for sports betting, I did do a DraftKings ad, but it was a few years ago, as I recall, and I actually recently was asked to do another sports betting service and I said no for the reason that my mind has changed. It's, again, a tough edge case for me because I think I have nothing against betting. I'm very happy with people going to casinos, betting, doing lotteries, whatever, which is why I said yes in the first place. But the particular implementation of

2:06:11.6 S1: of sports betting in the modern world, I agree that it's kind of abusive. Weirdly, I think that casinos are much more responsible than sports betting apps and things like that. For one thing, the sports betting kind of hurts the experience of watching a game and things like that. And for another thing, the potential for abuse by the actual players and coaches is huge. Like, you place a bet, you're a player and you place a bet saying, "I think we're gonna win, but not by more than 10 points." And you fail to score a touchdown intentionally because you're up by seven and you don't want to lose your bet, right? That's like hugely distortive of sports as a fair system. And I think there's been plenty of evidence that that is actually happening right now. But also, the sports betting companies, through the magic of capitalism, are trying to extract as much money as they can, and they're not being very good about helping people that are problem gamblers. They kind of want people who are problem gamblers to sign up on their site. And so, yeah, the casinos are actually better at that. I remember when Jennifer was writing her book, The Calculus Diaries, and she had a chapter on probability and odds. And we went to Vegas and we took a lesson on playing craps because neither one of us is craps players. We don't really play any casino games other than poker, which is a different story. And you would think that the guy who was paid by the casino to teach you how to play craps would have the incentive of having you lose as much money as possible. But this guy was very good. He said, "Look, I know people think they have a system and they're gonna beat this craps table, they're gonna win money. You can't beat my craps table." He says, "If you think you have a system, you're gonna beat my table, we have ATMs all over the casino for people like you. But it's not gonna happen." He was very honest about it, and that's, to me, very much the opposite of the ethos you get from a lot of sports betting sites these days. So, but it's hard. It's a tricky thing that I try to be good about, and maybe I make good decisions sometimes and maybe not other times. Who knows?
2:08:19.0 S1: Zach McKinney says, "To what extent do you consider the explainability of AI to be an important goal beyond the trivial sense of knowing the model architecture and its training data? I.e., is extracting a higher order, more intuitive explanation for why a model produced a certain input likely to be informative and useful, for example, in checking a model's accuracy or aiding the control problem, or potentially misleading by giving us a falsely simplistic understanding of causality?" Well, I'm not enough of an expert on what you can learn from AI interpretability, but as a scientist, my inclination would be, of course you want to understand what is going on inside the computer program that you're relying on for all sorts of reasons. Maybe you just want to have intellectual curiosity about how it gets these answers. Why is it so good at playing chess or something like that? Also, you might want to be able to get a better opinion about its reliability, right? Maybe you can understand better what kinds of mistakes it can make and not make. Like, we all know the famous examples of LLMs not being able to count the number of Rs in the word strawberry or whatever. And that's mostly gone away because once you know that there's a problem, you can go down and reprogram it to chase down that error. But the point is that's illuminating because it's a very silly, trivial kind of error to a human being. And that has to do with the spikiness of the notion of intelligence, the high dimensionality of it. Like, you can be really good at some things, like LLMs clearly are, and really not so good at other things. And understanding why that's true would absolutely be helpful in understanding what tasks LLMs might be reliable at. And finally, you might just learn something about the nature of solving a problem, right? This is one of the big issues with all of the advances in mathematics from the LLMs. You have a proof, but if no one understands the proof, have you really understood anything? And I think that's a very deep question. If you have a computer program that can prove theorems that no human being can understand, is that a good thing or a bad thing? I think it's a good thing, but it's a different thing than what we've had before. And so, understanding how it's doing it would be very helpful. Now, the question of whether it's misleading by giving us a falsely simplistic understanding of causality, I think that's a very good point. I think that's always a danger when you're trying to explain something that is very complex. You're always going to be simplifying it, and you're always going to be leaving things out. And so, you have to be careful about that. But I think that's true of any explanation, not just explanations of AI. So, I think it's a very important goal, and I think it's something that people are working on very hard. And I think it's going to be very important for the future of AI.

2:10:19.6 S1: I personally think that that's okay because it's not forever. The LLM gives you a proof and it gives you some steps and maybe it's very hard to interpret what those steps are. But okay, you do work and you try to understand it in a more human way. Likewise, explaining what's going on inside the large language model or some other machine learning algorithm might give you clues to how better to understand other problems. Ultimately, at the end of the day, human beings understanding things, I think, is the goal of the scientific part of this enterprise anyway.
2:10:55.8 S1: Lorenzo Producer says, "I'm a professional music producer and songwriter from the UK, and something you said about scientific careers really resonated with me. Finding the overlap between what you love, what you're good at, and what other people actually find interesting, that seems equally relevant to music. After 15 years of the grind, I finally achieved my dream of earning a modest living through music with some notable successes along the way. From your perspective outside the music industry, what insights from your experience in academia might help someone make the leap to more consistent and widespread success?"
2:11:27.9 S1: That's a perfectly legitimate question to ask, to wonder about, but it's just a little bit too general for me to give really sharp answers to. What insights might help someone make the leap to more consistent and widespread success? I say it all the time, but I think it's true: different people will succeed in different ways. So if I'm a person looking to succeed, I'm not trying to figure out what is the right way to do it. I'm trying to figure out what are all the different ways that people have done it and which of those might apply to me and to my situation.
2:12:01.6 S1: Someone might need to adjust their own preferences to better fit the market. If you actually want to make a living, that's okay. That's not selling out. That's understanding that a market is a two-way street. There are sellers and buyers. Okay? If you just want to be an artist and just want to make music or make art for yourself, then you don't have to worry about that. But if you want to earn a living, which is an entirely legitimate thing to do—most people want to earn a living doing something—they have to make some adjustments so that what they're doing is what the person paying them wants them to do, right?
2:12:39.9 S1: But maybe some people are already doing that. Maybe some people are already speaking to what the audience wants, but they're not putting themselves into it enough, right? You can make the opposite mistake also. You can make the mistake of trying too hard to guess what other people want and trying to do that rather than what you're passionate about. Again, it's the intersection that I think is really important.
2:13:04.7 S1: And finally, some people might not succeed. Whenever it comes to these career-type questions, I'm not someone who will just cheer you on. I get in trouble about that with my academic colleagues, like if there's grad students or whatever and they want to say, like, "Yes, go for it. Pursue your passion, do what you care about." And I have to feel like I need to say, "It's hard. You might not make it. There's not a lot of jobs." By all means, pursue your passion, but do it with open eyes. Know what the chances are. Try to judge whether you're on the right trajectory or the wrong trajectory and what you can do to make it better. And you just have to take into account all of the particularities and granularities of your specific situation rather than looking for a general rule.
2:13:55.7 S1: Jonty says, "It's getting increasingly hard for early career theoretical physicists who are motivated by discovery science and understanding nature, and not so much by the applied side of science or engineering, especially in the current climate, at least in Europe, where universities are not interested in hiring you unless you're bringing in industry funding and your research is going to translate into marketable products within the next five years. What would you do if you were starting your career right now in this kind of climate? Would you..."

2:14:19.5 S1: Would you try to play the game and go into translatable research? Or would you try to go as far as you can in pure theoretical physics and philosophy at the risk of falling off a funding cliff? I'm not actually sure that it's true that there's more emphasis now on practical, down-to-earth, economically relevant science than there ever was. There was always that emphasis, and there was always some little part that was more pie-in-the-sky, basic research kind of thing. But these gripes about how we can't get enough money for basic research have been going on for many, many decades. Let's just put it that way.
2:14:56.4 S1: And I think, again, I'm giving out too much career advice here, which I am always reluctant to do because everyone's career is different. But if you can't earn a living doing exactly what you want, then it's not necessarily the best idea to do something that is kind of a family resemblance to what you want. It might be that the second best thing that you want to do is entirely different. Like if I couldn't do the kinds of theoretical physics and philosophy that I do, I would not do applied physics or experimental physics. I have enormous respect for those fields, but those are not my second favorite job choices. I'd rather be a theorist about politics or something else, or economics.
2:15:38.9 S1: I like theorizing. I like grand narratives. I like trying to put a big picture together. That's what I do for a living. That's what I like to do for a living. I would look for some kind of job where I could do something like that. I did seriously at one point contemplate law school if physics didn't work out, but I wouldn't have wanted to be a lawyer. I would have wanted to be a law professor and write articles about constitutional theory or something like that. But again, everyone is different. Some people might say, "Well, if I can't do this kind of physics, then I'm gonna do chemistry," or, "I'm gonna do whatever, bioengineering or computer science." That's perfectly valid also. You have to understand yourself as much as you have to understand the outside world.
2:16:20.7 S1: Ed Sedstuf says, "You often describe macroscopic objects as emergent from fields. But how confident should we be that quantum fields and even the universal wave function you point to aren't emergent descriptions of some deeper structure themselves? Is there a way to recognize when we reach ontology rather than just another emergent layer?"
2:16:39.5 S1: No, there is not. But that's a good thing. Who cares? It doesn't matter. The people who realized that you could discuss fluids as collections of atoms didn't know about quantum mechanics. They didn't know that you should think about atoms as part of the universal wave function or whatever. That didn't stop them from making progress in statistical mechanics and kinetic theory and things like that. Likewise, if it happens to be some even lower-level thing, that's great. We'll discover that. That doesn't get rid of the level that we were looking at.
2:17:10.3 S1: There is a way to sort of formalize this notion, which is called structural realism. Our discussion a long time ago with James Ladyman touched on some of these issues. The idea that when you're discovering new layers of reality, the ontology can wildly change, but there are certain structural features, certain patterns, certain emergent behaviors that don't change. And so there's something real to what you are discovering, even if the reality is not being attached to the ontological status of different objects in your theory. Structures are more likely to survive from place to place, and that's fine.
2:17:52.9 S1: In terms of when it happens, the obvious example is when there's an experimental indication that your current structure is not good enough to explain what is going on. And that's easy to say. Like if you're in 1905 and you're discovering new things about quantum mechanics and you're saying, "Oh, wait, waves have particle-like properties," or whatever, that's a pretty straightforward example of an experimental requirement if you're trying to understand blackbody radiation or the photoelectric effect or whatever.

2:18:25.7 S1: These days we don't have that. We don't have a direct problem with an experimental fit, but we do have things like dark energy and quantum gravity and the hierarchy problem, which may be more indirect hints that you need a more fundamental ontology. Or they might not be. That's just why it's hard doing science, because you don't always exactly know what the—you're getting a clue. You don't exactly know what that clue is truly pointing you to.
2:18:53.5 S1: Stephen Goodwin says, "In a universe where everything travels at the speed of light, I understand that nothing in that universe can meaningfully measure time passing. Therefore, before electroweak symmetry breaking and the Higgs mechanism switching in, the only meaningful use of time would be by something external to the universe watching the universe evolve. Does this still make sense? Does it still make sense to talk of time passing in the universe before the symmetry breaking occurred, or did time in our universe start at that point?"
2:19:20.3 S1: No, it's just not right to say that in a universe where everything travels at the speed of light, nothing in the universe can meaningfully measure time passing. If you're in a universe where particles travel at the speed of light, a combination of two particles has a center of mass that does not travel at the speed of light. So it's entirely possible to measure time passing. Indeed, you can build a light clock. You can have beams of light bounce back and forth between mirrors and use that to measure the passage of time. So time exists, time is passing. This is just true. The correct thing to say is that particles moving at the speed of light don't experience the passage of time. Their temporal distance between beginning and end is zero. But time is still there. Time is still passing, just like always.
2:20:09.7 S1: Henry Jacobs says, "The Greg Caruso episode was much more nuanced than I was prepared for. For example, Caruso is a free will skeptic, yet he does seem to grant that the idea of choice is well-defined. He also used the word autonomy as distinct from free will. I'm really not grasping the distinctions between these words."
2:20:27.5 S1: Yeah, you're not alone. I have trouble grasping the distinction between these words also. As it came out in the discussion, what Greg really cares about is this idea of moral desert, right? Of the idea that people who do bad things have some intrinsically bad nature and should be punished for it, not just to prevent them from doing something more bad, but because of who they are. And he believes that the way to overcome that stance is to get rid of the idea of free will entirely. There's just absolutely a certain amount of linguistic juggling going on here. That's not to blame Greg for it. On any side, there's a bit of dancing around what the definitions are and making semantic distinctions and all these things, which is why I don't love discussions of free will.
2:21:14.7 S1: So what are the differences between choice and autonomy and free will? It's a little bit unclear to me as well. Let me put it that way. I don't want to try to say what Greg Caruso would say if he were here. I'm not very good at putting words into the mouths of guests, especially if I don't agree with them and their way of thinking about it. But I do think that the overall lesson, that there's more in common here than you might have thought, I think is a good one. I think that people can come out on opposite sides of whether or not this definition or that definition is serving a useful purpose while still agreeing on a whole bunch of the substance along the way.
2:21:53.5 S1: Sandro Stookey says, "What's your take on op-eds? Have you ever written for any big news outlet? Why or why not? You clearly do a lot of communication in the public sphere, but I don't think I've ever read an op-ed by you."
2:22:04.7 S1: I have written op-eds, but generally pretty benign ones. I've had op-eds appear in the New York Times several times, but they're not like I'm taking some controversial stance. It's more like, "Quantum mechanics is hard," or "Stephen Hawking was really good at physics." Those are the kinds of topics that I cover. Getting op-eds into a newspaper is a tricky thing. There's plenty of people who think that their words need to be understood, need to be heard, need to be read in the newspaper, and so there's a lot of competition there. At some point, I just think it's easier—I have a podcast. I can just talk. I won't get as big an audience as I do if I get a New York Times op-ed. It's still...

2:22:44.2 S1: It's still true, by the way, that people do read the New York Times. When I do get something in the Times, random people who I wouldn't have thought were New York Times readers come up to me and say, "Oh, yeah, I saw your article there." So there's still purpose to that, and I do think it's important. But I'm not a big enough name to just get something in the Times because I want to, so I really have to have a good reason to do it, and that hasn't happened lately.
2:23:08.4 S1: Michael Astor says, "I've recently finished reading Something Deeply Hidden. I took away a lot from it, but because I can't figure out how to phrase the question, I'll settle for a second thought I had. In the book, you mentioned that we don't appear to observe ourselves as being in a superposition. Why should we be able to, though? Is that something that can actually even be measured, or have I misunderstood what you meant?"
2:23:29.5 S1: Well, if you obey the rules of quantum mechanics and you have a spin that is in a superposition of spin up and spin down, and then you observe it, everyone agrees, if you solve the Schrödinger equation, the system that is you and the spin evolves into a superposition. A superposition of the spin was up and you observed it up, plus the spin was down and you observed it down. Okay. And then the question is how to interpret that. You are part of a superposition, but Everett makes the crucial move that you by yourself are not a superposition. You're entangled with something else, and that really, really matters. It matters because the environment is there and you're entangled with that as well, and therefore the part of the superposition where you saw the spin up will never interact with or have anything to do with, have any influence over the part where you saw it to be spin down. And therefore, you should identify yourself in the wave function as one of those two parts of you, the part that saw the spin up or the part that saw the spin down, not both.
2:24:36.3 S1: And the traditional thing to do would have been to interpret you as both at the same time. And then if you try to—again, there's a job of interpreting what the language of the equations is trying to tell you. The straightforward interpretation would have been you should be thinking that you saw both a spin up and a spin down. And no one has ever seen that, no one has ever thought that. So that's an apparent mismatch with our experience. And Everett's solution is you've misidentified yourself in the wave function. You're one or the other; you're not both.
2:25:09.6 S1: Steve Bonner says, "I recently heard that quantum tunneling is needed in star formation to allow two protons to overcome their repulsion and get close enough to fuse. Is this true?"
2:25:18.8 S1: Yeah, it's absolutely true. Quantum tunneling is not anything weird. I mean, it's weird maybe from our classical intuition, but it's not something unusual. It happens all the time. Even in chemistry, in some sense, you can argue that a little bit of quantum tunneling happens from time to time, but certainly in nuclear physics it does. In nuclear fusion it does. If I have two protons next to each other, they are both positively charged. Positive like charges repel each other, and therefore the two protons want to push apart. They don't want to join together. That's why nuclear fusion is hard, because it's hard to get the two protons to be very, very close. But if you do get them to be very, very close, then they can tunnel through a barrier to a point where they combine, they spit out a positron, and they combine into a deuteron, a particle that is a combination of a proton and a neutron, right? Heavy hydrogen, if you want to do that. And then more things happen along the way inside a star. But yes, pushing two protons together to fuse together is absolutely an example of quantum tunneling. We see quantum tunneling all the time. Basically any radioactive decay can be thought of as a quantum tunneling, let's put it that way.
2:26:33.8 S1: AJ says, "You thoughtfully answered my question last month about evidence for a cosmological multiverse, but part of your answer was the best theory we have for an explanatory account of the value of the cosmological constant is the multiverse." And...

2:26:46.1 S1: And what, how's the cosmological constant connected to the multiverse? Yeah, sorry, this is a famous thing, so I just think everyone knows it, but obviously that's a bad assumption. If you have a multiverse, one of the possibilities, not an inevitability, but one of the possibilities is that the cosmological constant takes on different values at different parts of the multiverse. That's easy enough to arrange in the string theory landscape or different versions of the inflationary multiverse. And then the argument is the anthropic principle. If the cosmological constant is very large and positive, it basically pushes matter apart so quickly that you can't form galaxies or planets or life. Whereas if the cosmological constant is large and negative, it causes the universe to recollapse very, very quickly. And as a result, life would only form in those parts of the multiverse where the cosmological constant is relatively small. Guess what? We look around the universe we're in and we have a cosmological constant and it's relatively small. And the fact that not only is it relatively small, but it's not exactly zero, is evidence in favor of the multiverse, because other theories, which we don't even have, right, we're imagining the space of theories we haven't invented yet, but maybe you could imagine that there's some symmetry or some dynamical mechanism that makes the cosmological constant exactly zero, because there's something special about zero. It's much harder to imagine something that makes it exactly 10 to the -120 times the Planck scale. But the anthropic principle can do that, and it did that before we measured it, and now we've measured it, and therefore your credence in the multiverse should go up. Maybe your credence in the multiverse is so small that it doesn't go up by much, but that's a different question.
2:28:32.8 S1: Michael Helgerson says, "Was there a moment in your undergraduate education in astronomy that drew you to philosophy, or were you always attracted to a diverse set of disciplines?" Well, I was always attracted to a diverse set of disciplines, but philosophy was not among them when I went to college. I'd never taken a philosophy course. I had no disrespect for it, but I had no knowledge of it either. But at Villanova at the time, I just visited Villanova again, it's no longer true, the decline of academic standards is pretty obvious everywhere. In my day, everyone had to take three semesters of philosophy and three semesters of religious studies during their four-year college undergraduate degree. Everyone in the Arts and Sciences school, anyway. So I did that and I fell in love with philosophy, mostly with moral philosophy and ethical philosophy, political philosophy, but later I discovered foundations of physics, which is what I'm mostly in love with right now.
2:29:28.9 S1: Kevin James says, "Thinking while listening to episode 365 with Vitor Cardoso had me wondering if shock waves from supernovae in the early universe couldn't help to produce some of these primordial black holes. Alternatively, is it plausible that overlapping gravitational waves from merging black holes or other stellar events could compress a region of space to contribute to black hole formation?" Well, all of these scenarios are worth considering, sure. You have to run the numbers. You can't just sketch out the story, you have to actually check the equations. But here's the thing: making black holes is hard. It's hard because you have to squeeze a lot of matter into a very small region of space. That's why it rarely happens. Most stars, when they give out their nuclear fuel, they don't make a black hole. They settle down into a white dwarf or a brown dwarf or something like that, a neutron star. Only the most massive ones are believed to eventually create a black hole. So you have a very massive star, otherwise pretty quiet, giving up on its nuclear fuel burning and collapsing to make a black hole. In fact, most of the star gets blown out in the course of a supernova, and the rest of it makes the black hole. So it's only a tiny fraction, or a small fraction, I should say, of the original mass of the star that makes the black hole. So if you're...

2:30:49.9 S1: So if you're imagining things like shock waves from different supernovae bumping into each other, those are going to be pretty diffuse and spread out compared to a star itself, right? And so I don't see a lot of quantitative hope that you're gonna make a lot of black holes that way. Plus, of course, you have to do that after you've made the supernovae, so you wouldn't be making dark matter black holes this way, 'cause dark matter pre-existed before we made the first supernovae in the universe.
2:31:23.4 S1: Ken Wolf says, "In your talk with Chandra Sripada, you discussed the question of whether LLMs can be regarded as what he dubbed moral patients. That's P-A-T-I-E-N-T-S, so like patients in a hospital, not like you're being patient with somebody, or entities whose moral welfare needs to be considered. Two sides of the question were identified as computational functionalism versus biological naturalism, but I think a vital component of the latter was not considered: the simple fact of biological entities being vulnerable and mortal. The personhood represented in a biological brain cannot be backed up or duplicated. Could a case not be made for only granting moral patient status to beings which have the same skin in the game that we do, exactly because they share the same limited, singular sort of existence as us?"
2:32:11.1 S1: I'm not that optimistic about this particular criterion. I think that thinking along these lines is important and interesting, but two things. Number one, are you sure that we can't be backed up? Right? There's certainly plenty of people out there who think that a human being could be backed up once we develop the technology. And number two, even if you can imagine duplicating a computer program into a different substrate, that doesn't mean you do. Lots of computer programs are just deleted. Maybe they're written once and then not saved. So I don't think that's quite the criterion you're looking for there. I think it's something about whatever it is, it's something about the feeling experience of the thing that you've instantiated, either as biology or as computation, whatever that may be. I don't claim to have the final answer myself.
2:33:04.3 S1: Jonathan Kleiman says, "Do you think the movie Gattaca will come true? A society where genetic engineering is ubiquitous, that ranks people by their genomes? It seems like only a matter of time." Well, look, we already rank people by their genomes. That happens, okay? Not by actually examining their genomes, but some people are more attractive or smarter or more likable or more popular or whatever, in some part because of their genetic heritage, right? There's both nature and nurture that goes into these kinds of things. I don't think you need to have anyone's genome laid out in front of you to do that kind of ranking. And it is bad in some ways, and it's what we're used to in other ways.
2:33:44.7 S1: I do think that genetic engineering is going to become much more common. Ubiquitous, I don't know, but when we had Fyodor Urnov on the podcast, he was pretty clear that it's not going to be that hard to genetically engineer human beings down the road with CRISPR-Cas and things like that. I honestly think that this is something we haven't reckoned with. There's a certain icky feeling like, "Oh, maybe we shouldn't do it," but I kind of do think it's going to be ubiquitous. And so I wish we faced up to it and thought harder about what it would mean, because again, these questions are hard, the systems are complex, there are unintended consequences, but thinking them through ahead of time can be very useful.
2:34:25.8 S1: Wonder says, "How do we know that random quantum fluctuations really are random? Not so much doubting it, but wanting to understand it." Well, we don't know anything for 100% certainty. We have theories, and the theories fit the data. The best theories we have of quantum mechanics predict that measurement outcomes under the right circumstances are random. Not necessarily 50/50 random; they obey the Born rule—the probability is proportional to the wave function squared—but there's some...

2:34:58.5 S1: indelibly, indelibly stochastic element to what is going on. You might say, "Well, okay, maybe there's a different theory where they're not random." Good, write down that theory and see if you can also fit the data. There are hidden variable theories; there's Bohmian mechanics, etc. Now, there's problems with them in fitting quantum field theory and modern physics, but okay, putting that aside, imagine you could make a theory like that. Then in some sense, the outcome is not random. The outcome is determined by the value of the hidden variables. But in another sense, it's perfectly random because you don't know the value of the hidden variables, okay? To you, it's random. There is some epistemic or subjective uncertainty there. So it's a theory that works and we don't have a better theory; that's why we believe it. If anyone has a better theory, then they should absolutely put it forward.
2:35:48.5 S1: Sam Hartzog says, "There seem to me to be two forms of anthropic reasoning: 'We exist, therefore X is more likely,' or 'Given X, our existence would be more likely a priori.' I realize there's a subtle shift in framing and may account to no difference at all in terms of formal logic. Do you see any difference between claiming our existence as evidence for a theory versus accounting for our existence when assigning credence to a theory?"
2:36:14.5 S1: I'm not sure that these are two... they may be two forms of anthropic reasoning that get thrown around informally, but I think that the careful applications are a little bit better than that. I would put the careful application this way: anthropic reasoning certainly starts with the observation that certain universes or certain conditions or certain environments would be inhospitable to the existence of life. Okay, everyone can agree with that. Here's the thing: that's not a relevant fact unless you say, "Okay, my picture of the universe actually instantiates all of these different regions with different environments or laws of physics or whatever." If you only have a single universe, then the anthropic principle doesn't tell you that much. I mean, it tells you things you already knew, right? Like here in the solar system, life grew up on the surface of the Earth. It did not grow up in the center of the sun or in interplanetary space. There's a lot more volume at the center of the sun or in interplanetary space than there is here on the surface of the Earth. Why did we happen to be so lucky to be growing up exactly in this very tiny region of the solar system? The anthropic principle is the answer, okay? There's different regions, but some of them are more hospitable to life, okay?
2:37:32.0 S1: So it only becomes interesting when you have these different environments, and then there's a pretty trivial statement that, of course, when you have all these environments, we will only exist where we can exist. Now, the non-trivial statement comes in when you have comparing different theories, and then you can say, "Okay, does one theory make it more probable that we would exist than some other theory?" There, that's when the anthropic principle actually has some bite, right? Presumably, we can rule out theories where our existence is impossible. I think that's fair; that's easy enough to do. But you might have different theories about the probability of life, and then there's an interesting intellectual argument. If two theories both allow for life like ours to exist, is there some comparison to be done based on how probable it is? Or is it just, "Well, if life can exist and we see ourselves here, we must have just gotten lucky," okay?
2:38:36.5 S1: I think that's an interesting question. I think the traditional anthropic reasoning would be, "If my theory gives a higher probability to life like mine existing, then I favor that theory." But I think you can argue with that way of thinking about the anthropic principle. I don't think that the physicists, anyway, have done a very good job of thinking this through, really. They're not very good at philosophy, usually.

2:38:58.7 S1: The Memes of Destruction says, "Perhaps more of a philosophical question about general relativity. I occasionally see physicists arguing in a near meme-like way about the metric. Can you speak a bit about the difference between a minus-plus-plus-plus versus a plus-minus-minus-minus metric?"
2:39:16.3 S1: Sure. For those of you who don't know what we're talking about here, the metric in general relativity is how we measure distances, but distances in spacetime, not just in space. So in space, if you have a plane, let's say, and you have X, Y axes, and you have a straight line in that plane, and you want to say, "How long is the line?" and you have the coordinates, right? X1, Y1 for one end of the line, X2, Y2 for the other end of the line. Well, you can use Pythagoras's theorem. The distance along that line squared is delta X squared plus delta Y squared. Okay, that's basically Pythagoras's theorem. And secretly, two plus signs appeared there. It's plus delta X squared plus delta Y squared.
2:40:05.2 S1: In spacetime, it goes the other way around. You have again something like Pythagoras's theorem, but there's a different sign in front of the timelike part of your distance than in front of the spacelike part of your distance. This is why in space, the shortest distance between two points is a straight line. In spacetime, it's the longest time between two events that is a straight line, not the shortest time. It's exactly because of that minus sign.
2:40:35.8 S1: However, when you actually go to write it down, whether you want to put the minus sign in front of the time part or the space parts is completely up to you. Okay, that is a choice. That is literally a convention, and different books will do it differently. And in fact, there's sort of a sociological thing where in particle physics and quantum field theory books, they put the minus sign in front of the spatial parts of the metric. In general relativity books, they generally put the minus sign in front of the time part of the metric.
2:41:10.3 S1: And there is some joking on social media between physicists. I think that the consensus is that minus-plus-plus-plus, which means put the minus sign in front of the time part of the metric, like is done in general relativity, does make things more convenient overall. It's fine either way. There's no actual difference except where some minus sign goes and how you interpret them.
2:41:34.1 S1: Roughly speaking, if you want to think of it, the sort of physical difference, the minus sign is gonna go somewhere and it's gonna be annoying. Okay, so which is more annoying? To have a minus sign relating the time of a timelike trajectory, or to have a minus sign relating the spatial distance along some spatial hypersurface? Okay, it's up to you. Both are annoying. You have to choose it. So it's just kind of a joke. Don't take it too seriously. Physicists have to have fun somehow.
2:42:03.5 S1: Brendan says, "How mindful are you for what you eat? In other words, do you generally stay away from processed foods or just try to keep below a certain calorie count, etc.?"
2:42:13.2 S1: Not super mindful, like a little bit mindful. When I was younger, I was very not mindful at all. I would happily indulge in all sorts of sweet desserts and things like that. I certainly do try to do better now. Especially keeping away from processed foods, I do think is a good idea. I realize that it's a fuzzy distinction. All food is processed in some sense, so there's levels of processing going on. But more and more, we're trying to cook our own food based on actual ingredients. That does seem to be a consensus of the expertise that you get. Calorie counts I don't really do very well, but I don't eat dessert that much anymore. If I indulge, if we're going out or celebrating, then I will do it, but that's not a regular thing for me like it used to be. The one thing...

2:43:01.6 S1: The one thing I've not been able to cut down on is pizza. I just can't. I just can't do it. I gotta live with my pizza. Not every day, obviously, but when it's possible, or when it's convenient and I haven't done it in a while, you'll find me eating pizza. I blame my graduate student upbringing. What can I tell you?
2:43:19.4 S1: Gary says, "Why is it helpful to think of quarks as particles? From a distance, the color confinement stuff makes it smell like hadrons are better understood as complex interactions in the quark-gluon fields, and then talking about quarks as particles is more likely to mislead than to help."
2:43:35.2 S1: Well, you're not gonna get any pushback from me if you want to say that we should be talking about fields rather than particles, but there are circumstances under which it's okay to talk about particles. That's fine. It's the fields that are more fundamental, but particles do show up in certain circumstances. But you're right that in a confined hadron, that's probably not one of the circumstances where it's good to talk about particles. Particles make the most sense as a way of talking about fields when they're essentially free particles, when they're interacting not very much, if at all. An electron moving through empty space is very particle-like, or even a photon moving through empty space. Inside a proton or neutron where you have all these fields interacting, as I go to great lengths to explain this in Quanta and Fields, the field picture is both more accurate and latches onto our intuition better than the particle picture does.
2:44:30.2 S1: Having said that, however, there are things about particles that matter. In a proton, it is correct to say there are three valence quarks. That is to say that there's a conserved quantity, or at least an approximately conserved quantity, carried by the quarks, which is baryon number. And each quark only carries a baryon number of one-third, so you get three of them to make a single baryon. Now, really, if you forced yourself to talk about what's going on inside a proton in terms of particles, you would have to say there are three valence quarks, each with the baryon number one-third, and then a roiling mess of virtual quarks and gluons and other particles for that matter, with net baryon number zero. So you have this picture where there's particles popping in and out of existence all the time, but of course, it's actually a static state, as we talked about before. So it's not that they're dynamically popping in and out of existence, it's that they're virtual possibilities looming there inside the proton all the time.
2:45:36.3 S1: So that's the usefulness of talking about a proton in terms of having three quarks in it, is that somewhere there's baryon number, and it's very conceptually simple to think of those baryon number quantities as living inside three individual quarks. An obvious check that that is not most of what's going on in the proton comes from the fact that the masses of the quarks are much less than one-third the mass of the proton. Quarks are pretty light. So clearly most of the energy in the proton does not come from those three quarks that you said are in there. It comes from something else, and the answer one way or the other comes down to that being the fields that are making up the proton. Sorry.
2:46:19.3 S1: Scott says, "I've spent some time around high-energy physics experimentalists, and I find they tend to view quantum mechanics and quantum field theory as separate subjects. Some of them may abstractly agree that one is a subset of the other, but they don't conceptualize them that way. To my surprise, I met a string theory grad student who similarly compartmentalizes the two. This is surprising to me because it seems to me the natural way to conceptualize quantum field theory is as a quantum mechanics applied to a continuous system. Is there some philosophical or practical merit to their views that I am missing here? Perhaps a lot of HEP physicists just don't take quantum mechanics seriously."
2:46:55.3 S1: Well, I think that there's not a lot of merit to it, no. There's some history and some sociology. Like when those people took their classes, quantum...

2:47:05.8 S1: Quantum mechanics and quantum field theory were two separate classes. Okay. Now implicitly, the quantum mechanics class treated particles or perhaps spins of particles. So you got qubits and things like that. It probably didn't treat fields in any particular way. And so quantum field theory is a whole separate course. It's just intellectually perfectly clear that quantum field theory is a subset of quantum mechanics. You have Hilbert space, you have operators, you have the probability is the wave function squared. All that stuff is still true in quantum field theory. But the actual problem sets that you're doing when you're taking the quantum field theory course are very different than those when you're taking the quantum mechanics course. So I wouldn't take that too seriously. I think it's just a reflection of their upbringing, not a principled intellectual stance.
2:47:51.8 S1: Jerome says, "When thinking about the double slit experiment, my mind is imagining the wave in the quantum field interfering to cause the