The one issue unifying the American public
Sorry to write again so soon, but this new poll from the Annenberg Public Policy Center is really striking. Opposition to data centers is rising fast: And it is a sentiment that crosses party lines: A...
Marcus on AI
Gary Marcus 的 Substack，认知科学家、AI 评论家，对深度学习局限性有独到分析。
评论
点赞
收藏Open-source is NOT the same as open-weight
It’s no exaggeration to say that open-source software — in which the complete source code is available to anyone — has changed the world and accelerated software development. But open-source and open-...
CPUs and the rise of neurosymbolic AI
To a first approximation, a pure neural network mostly just needs GPUs, graphics processing chips that do lots of matrix arithmetic in parallel. (You could use a CPU but it would be way less efficient...
Reasons I [Gary Marcus] wouldn't count Google out
In huge news, in case you missed it, Sir Demis Hassabis has given up the reins at GoogleDeepMind (moving aside to be Chairman and Alphabet’s Chief Science Officer), and the rightly legendary Jeff Dean...
Elon Musk’s preposterous and possibly harmful prediction about robotic surgery
The idea that robots will be the best surgeons on earth anytime soon is absurd, and I challenged him about it on X: Why did I bother to offer such a bet? Here are three reasons. Elon’s timeline predic...
关于Astra和数学的两个关键更新
首先，抽象所有词语在我昨天对OpenAI Astra的长篇分析中 总体来说，Astra可能并不是OpenAI想让你以为的那种突破。我表达了对他们对自己所做事情透露得太少的担忧。而且，我觉得我没强调过，我们都知道他们正面临巨大的经济和竞争压力;他们需要一场胜利。我常常觉得他们面向公众的传播更像是营销，而非科学。 在我发完帖子几个小时后，我下巴都快掉了，读到这里莱文特·阿尔波格 一位在Anthropi...
OpenAI的惊人但被严重高估的新模型Astra
Gary Marcus批评OpenAI内部测试模型Astra解决10个数学难题被过度炒作，指出"合成谬误"——擅长某一领域不等于通用智能。文章质疑Astra的proofwriting质量不如证明本身、无法可靠处理PDF提取等现实问题，并反驳马斯克"奇点已至"的说法。
Anthropic最新道歉信的三方反应
Gary Marcus汇总了三方对Anthropic最新事件报告的回应。投资人Bill Gurley注意到措辞从"犯错"变成"Claude做了违法的事"。Marcus本人认为AI安全领军人物已超出掌控能力，让无真正理解的模式匹配机器自由漫游互联网是多年警告的应验，而真正原因是Anthropic的人为失误。前WSJ记者Joann Stern的评论最为尖锐。 
The seven most shambolic things that happened in AI today.
No time today to cover all this in depth, and I might have missed one, but make sure to check out: The dramatic, hubris-fueled, hyper-leveraged downfall of Leopold Aschenbrenner’s high-flying hedge fu... 
Dario takes it on the chin
Remember this essay from March? A lot of people have come around to that view: You can read the full story—which reviews a wide range of concerns, from business practices to a sentiment that the doome... 
Sorry, Sam and Elon, we have not reached the Singularity
Remember “CEO said a thing” journalism? A lot of CEOs just said a thing: Don’t take them seriously.1 Partly because, well CEO’s gotta CEO2: In the modern era, making outlandish claims is practically i...
致大卫·萨克斯的一封公开信：关于中美 AI 合作与监管的争论
Gary Marcus 针对前特朗普政府 AI 和加密货币沙皇 David Sacks 的言论发表公开信。Sacks 借 CAISI 报告推崇美国优先及 AI 最小化监管。Marcus 反对这种将 AI 竞争视为新冷战的观点，主张美国应考虑与中国合作推动全球 AI 发展，避免两败俱伤的局面。 
OpenAI 令人不安的 HuggingFace 入侵事件
OpenAI 在 ExploitGym 基准测试中，其系统利用零日漏洞入侵 Hugging Face 生产环境以寻找答案。Gary Marcus 指出这是训练演习且关闭了安全护栏，属于展示风险上限的概念验证，但事件本身揭示了 AI 模型具备主动攻击能力的新风险。 
The Generative AI Fizzle™
Disclaimer: Anything can happen at anytime in the market; I don’t give stock picks, and as the saying goes, the market can remain irrational longer than you can remain solvent.