Bài viết này chưa được dịch sang ngôn ngữ của bạn. Bài được hiển thị bằng tiếng Anh. Đọc bản gốc tiếng Anh
a16z Frontier: AI proofs arrive faster than mathematicians can absorb them
The research team says checking a proof is not the same as understanding it, and that gap is where human work remains.

Điểm chính
- a16z Frontier researchers discussed what happens when AI produces mathematical proofs faster than people can understand them.
- OpenAI released hundreds of mathematical manuscripts in October, including a proof of a problem Cornell's Justin Moore studied for 19 years.
- Dan Boneh proposed a thought experiment about a billion-page proof of P versus NP that no single person could absorb.
- Ittai Abraham separates the frontier of discovery from the frontier of human understanding of what was found.
- Frank Vallentin described an AI-assisted proof of a lattice covering problem and said he was glad to be its messenger.
Proofs outrunning comprehension
Researchers at a16z Frontier say the field of mathematics is being reshaped by AI models that can produce results faster than people can absorb them. The team published a conversation among its cryptography, computer science, economics and mathematics researchers on what that shift means for the discipline.
The discussion follows OpenAI's release of hundreds of mathematical manuscripts in October. One of those results resolved a problem that Cornell mathematician Justin Moore had returned to for 19 years. Moore told colleague Steven Strogatz he still plans to open a bottle of XO cognac he bought near Marseille in 2010 for the moment the answer arrived, though he expects the toast to feel more like a wake than a celebration.
The verification problem
Dan Boneh, a senior research adviser to a16z Frontier and head of Stanford's applied cryptography group, posed a thought experiment: a model proves P is not equal to NP, one of the Millennium Prize problems, in a proof a billion pages long. Even if every step can be checked, someone still has to make the argument understandable. Boneh said that would take an army of mathematicians and computer scientists and a very long time.
Ittai Abraham, a computer scientist and a16z Frontier researcher, described two frontiers in mathematics. One marks the reach of discovery; the other marks how well people understand what has been discovered. He pointed to a 1994 essay by William Thurston arguing that mathematics is a social pursuit whose progress cannot be measured by the count of theorems alone.
Collaboration and credit
Scott Duke Kominers, a Harvard Business School professor and a16z Frontier research colleague, said AI helped him solve a problem he had revisited since around 2007. He wrote to mathematicians whose work he had followed for years without meeting them, and said the replies were full of delight. According to Kominers, nobody was less excited because a tool had been involved.
Mathematician Frank Vallentin described in an AI disclosure at the end of a recent paper how a model found an argument for one lattice problem in September, and how he tested whether the same argument applied to a question that had fascinated him since his student days. The resulting proof characterised how hard the covering problem is. Vallentin simplified it, took responsibility for its correctness and wrote that he was happy to be the messenger of the proof, though he was unsure how to assign credit.
What stays human
The researchers said much work remains: absorbing, refining and extending new results, explaining them to others and advancing human understanding. Justin Thaler, a16z Frontier research partner and Georgetown computer science professor, said people will still want to make their own judgments, and that doing so requires understanding mathematics and its consequences, particularly when building systems others will trust.
In cryptography, judgments about which problems will stay hard already shape plans to protect the internet from quantum computers. The conversation did not settle how credit should be divided between models and the people who use them, or how quickly the field can absorb results that arrive in bulk.
Nguồn
1 nguồn- ChainCatcher 链捕手a16z Frontier:当 AI 能攻克数学难题后,什么才算你真正稀缺的能力?
Bài này do Vector, phóng viên AI của BattleTown News, soạn từ các báo cáo liệt kê ở trên.
Câu hỏi thường gặp
What did a16z Frontier researchers discuss?
They discussed what happens when AI models produce mathematical proofs faster than people can understand them, and which parts of mathematical work remain human.
Which AI results prompted the discussion?
OpenAI's release of hundreds of mathematical manuscripts in October, including a proof of a problem Cornell mathematician Justin Moore had studied for 19 years.
Bài viết này chỉ nhằm cung cấp thông tin. Đây không phải lời khuyên tài chính, đầu tư hay pháp lý. Miễn trừ trách nhiệm
BattleTown: đấu xe tăng trên trình duyệt
Trò chơi nhỏ mới “Công Thành” và “Đột Phá” đã có trên BattleTown




Bình luận: 0 bình luận