智能体
x.com ·
@hwchase17
Harrison Chase 点评 Google 多智能体证明发现论文：验证者独立上下文是关键模式
Harrison Chase 转述 Google Research 关于多智能体证明发现的论文（Cogentic，运行于 Gemini），认为其核心模式是验证者从干净上下文开始，避免被探索者的推理说服，从而允许探索者尝试大胆想法。转述内容称系统按轮次运行，编排器决定每轮证明者数量，每个证明者负责一个方向并获得摘要代理独立撰写的简报；每份草稿需通过两个对抗验证者（一个单独检查、一个横向比对同轮草稿以发现共性错误）；代理通过磁盘上的记录与账本共享状态，审计器从被拒证明中提取正确引理并独立复核；另有过程顾问根据验证日志更新指令。转述称该系统在在线学习、拍卖理论与机制设计等 5 个开放问题上产出新结果，多数问题约需 100 次 Gemini 调用，最难约 1000 次。上述结果与专家检查情况来自转述，尚未独立核实。
AI推荐理由 转述 Google Research 多智能体证明发现论文（Cogentic，基于 Gemini），给出具体机制：独立上下文的验证者、双对抗验证、磁盘共享状态、审计器提取引理、过程顾问更新指令，并称在 5 个开放问题上产出新结果、多数约 100 次 Gemini 调用、最难约 1000 次。信息密度高、机制可复述，但为二手转述，'新结果'与'领域专家检查'未经独立核实，需保留不确定性。
82
浏览 2,001
点赞 20
评论 8
转发 1
正文
good pattern from this google paper
the verifier starts from a clean context, so the explorer's reasoning can't talk it into a bad proof
so the explorer can try wild ideas and the verifier tosses the wrong ones
same reason deepagents subagents get their own context
[x.com/omarsar0/statu…](https://x.com/omarsar0/status/2106056369816420624)
elvis@omarsar0
Banger paper from Google Research on multi-agent proof discovery.
(bookmark it)
It's really interesting to see this emerging multi-agent pattern: not enforcing too much execution structure and pairing it with dedicated agents for advising and verification.
I think it is generally applicable as well. Great read.
Here is how it works:
Cogentic runs on Gemini and works on open problems in theoretical computer science, starting from the problem statement with no expert hints.
The system works in rounds, and the orchestrator decides how many provers to run in each round. Every prover gets one direction to work on, such as a specific bound or a counterexample search, plus a short briefing that a summarizer agent writes from earlier attempts and verifier feedback. Each summarizer writes its briefing independently, so provers in the same round read different summaries of the same history.
Each draft goes through two adversarial verifiers. One checks the draft on its own, and the other reads all of the round's drafts side by side to catch shared mistakes. A draft is accepted only if both pass it.
The agents share state through two disk documents. A record logs every attempt with the objection it failed on, and a ledger stores verified lemmas and ruled-out directions. An auditor extracts correct lemmas from rejected proofs, verifies them again independently, and adds them to the ledger.
A separate process advisor reads the verification logs across rounds and updates the instructions given to provers and verifiers. The orchestrator and the advisor can't give mathematical opinions, so the provers provide all the math.
It produced new results on five open problems in online learning, auction theory, and mechanism design, each checked by domain experts. Most problems took around 100 Gemini calls, and the hardest took around 1,000.
Paper: [arxiv.org/abs/2609.40324](https://arxiv.org/abs/2609.40324)
Chat with Paper: [academy.dair.ai/papers/cogenti…](https://academy.dair.ai/papers/cogentic-multi-agent-orchestration-for-automated-proof-discovery-2609.40324)
[🔗 View Quoted Tweet](https://x.com/omarsar0/status/2106056369816420624)