Industry

Gary Marcus 回应质疑:2024年12月已预言数学因可验证性更易突破

Why we picked it

Gary Marcus 晒出 2024 年 12 月的推文截图,证明自己早于 o3 发布前就说过“数学因可验证性更易被攻克”。这条推文撕开了 AI 圈“预言归属权”之争的一角,也提醒我们判断谁真有先见,得看时间戳而不是转发热度。关注 o 系列模型和数学基准进展的人会感兴趣。

Gary Marcus 在 X 上回击 Arpit Gupta 等人关于“事后诸葛”的批评,坚称自己在 2024 年 12 月、o3 发布之前就已明确指出数学任务因结果可验证性而更容易被 AI 攻克。他展示了自己当时关于 o3 的推文作为证据,认为这一观点至今未获得应有的认可。

Body

Marcus 指出,批评者把他针对黎曼猜想的具体表述断章取义地套用到 o1 模型上。他强调自己的完整论述中包含“将定理证明器与纯 LLM 结合是有前景的方向”这一判断,呼吁大家阅读其原始工作而非互相转发推文。

这场争论反映了 AI 领域对“可验证领域(如数学、代码)进展更快”这一判断的归属之争。Gupta 的原帖认为该观点如今更像事后的合理化解释,而 Marcus 则以推文时间戳自证其先见性。

Original · Gary Marcus

Contra the below, I was fully explicit that math would be easier because of its verifiability in December 2024 (in my tweet re o3, before it was released, but so far zero people have credited me with that. Iinstead prefer to take something I said about Riemann specifically about o1 models without Leean out of context with the rest of my work where I said that combining theorem provers with pure LLMs would be promising. If people actually read my work instead of each other’s tweets, they would be a lot more impressed with my demonstrable. well-documented foresight. I have endless receipts, but as Jack Nicholson said, they can’t handle the truth. Instead, they see what they want to see. Arpit Gupta @arpitrage Few people now saying “math is easy because it can be verified” predicted ex ante that math would get solved quickly. There is obviously truth to the idea that verifiable domains will fall faster; but it’s also partially ex post cope 🔗 View Quoted Tweet📊 2 ⚡

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Key entities and technical topics under continuous tracking include Gary Marcus, o3, reasoning models, AI safety.

Source attribution: curated from Gary Marcus (@GaryMarcus).