Google论文：Cogentic多智能体协作发现5个未解数学问题新证明
New Google paper reveals how Gemini found new proofs for 5 unsolved math problem…
多智能体协作攻克数学难题是Agent领域的重要突破，展示了结构化验证在复杂推理中的价值，值得研究者关注其架构设计。
New Google paper reveals how Gemini found new proofs for 5 unsolved math problems.
Google 最新论文揭示了 Gemini 如何为 5 个未解数学问题找到新证明。
Organize AI like a research team with strict checkers and shared notes:
像研究团队一样组织 AI，配备严格的审查员和共享笔记：
A single prompt often isn't enough for hard research problems. They need many attempts, tough review, and a memory of what already worked.
对于困难的研究问题，单个提示往往不够。它们需要多次尝试、严格的审查，以及对已有成功经验的记忆。
Google's system, Cogentic, gives Gemini that structure. Several agents try different ideas at once, checkers assume every step is wrong until proven, and proven pieces are saved for the next round.
Google 的系统 Cogentic 为 Gemini 提供了这种结构。多个智能体同时尝试不同的想法，审查员在得到证实前假定每一步都是错误的，已验证的部分会被保存下来供下一轮使用。
Most problems took only about 100 model calls, and human experts confirmed every proof.
大多数问题仅用了约 100 次模型调用，且每个证明都经人类专家确认。
If your agents tackle long, hard tasks, give them a strict checker and a running record of proven work, not just a better prompt.
如果你的智能体处理漫长而艰巨的任务，请为它们配备严格的审查员和已验证工作的运行记录，而不仅仅是更好的提示词。
– arxiv. org/abs/2609.40324
– arxiv.org/abs/2609.40324
Title: "Cogentic: Multi-Agent Orchestration for Automated Proof Discovery"
标题：“Cogentic：用于自动证明发现的多智能体编排”
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力