My “Knowmads” podcast on science and AI

Or click here if the above doesn’t work.

Recorded in-person in my office at UT Austin, with a bulleted list containing “ARC,” “Scalable Oversight,” and “Models” behind me on my blackboard for some reason (I no longer remember who put those there or why). 90 minutes long. Sometimes you see my disembodied arm waving in midair because of the way the cameras are combined. As always, I strongly recommend 2x speed for the correct experience.

This might actually be one of my best podcasts ever, although I wasn’t planning on that! Thanks so much to Bhavay Tyagi and Prachi Garella for driving all the way from Houston to record it.

Here’s a strict subset of the topics we covered:

• The story of AI models solving the Navier-Stokes Millennium Problem, insofar as it’s known
• Can recent AI proofs be called “truly creative”?
• The history of AI before the LLM revolution
• What do we mean when we call LLMs “black boxes”?
• The achievements of the field of interpretability
• What exactly happened in the OpenAI/HuggingFace incident
• Must we avoid all “anthropomorphizing language” when discussing the HuggingFace incident? (spoiler alert: no)
• Examples of major open problems in quantum computing theory that I cared about for decades and that AI models have recently solved
• Effects of the current AI cataclysm on the math community, especially students
• What annoys me the most when I listen to AI talks
• My experiences at OpenAI, why they hired me, and the watermarking work that I did there

Enjoy!

More AI-related content coming soon, as this blog—like much of the rest of the world—continues its transition to “all AI, all the time” (except still 100% written by an aging, deteriorating biological brain)