Anthropic Lance Martin：1000 个 Claude 智能体协作形式化费马大定理

他对长时程智能体的核心判断是：成败主要取决于奖励机制和评测设计，而不是提示词本身。这类大规模多智能体协作需要把任务拆成可验证的子目标，靠 evals 来衡量每一步是否正确，而不是靠一句精心打磨的 prompt 让模型一次跑完。

这个观点对开发者有直接参考价值：与其在提示词上反复调优，不如把精力放在可验证的评测和奖励信号上，这在数学证明、代码生成等可自动验证的场景里尤其适用。

Great chat with @RLanceMartin (Anthropic) at #AIECode. He just scaled 1,000 Claude agents to formal...

Great chat with @RLanceMartin (Anthropic) at #AIECode . He just scaled 1,000 Claude agents to formalize Fermat’s Last Theorem. His take on long-horizon agents: it mostly comes down to rewards and evals, not the prompt.📊 2 ⚡

This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Claude, Anthropic, 智能体, 提示词工程.

Source attribution: curated from AI Will (@FinanceYF5). Cross-source corroboration: 11 independent sources have covered this development (including TestingCatalog, The Rundown AI, rohanpaul_ai), confirming industry-wide signal strength.