AI 自我进化会“背题”，这篇给它加了刹车
让 AI 自己改进自己的“操作手册”(提示词、工具、记忆这些外围配置)，已经能带来不小的提升。但研究者发现，这种自我进化容易“背题”:在训练用的任务上分数大涨，换个没见过的任务，提升就缩水甚至归零。这篇给进化过程加了“正则化”约束：限制每次改动的数量、鼓励探索没走过的路、并砍掉那些太小、太贵或没用的改动。结果在 8 个基准上，不仅训练任务上最多涨 14.1 分，5 个没见过的任务上也最多涨 4.7 分，而且跑起来还省了 30% 的 token。它不是你明天能用上的东西，但它指出了 AI 自我改进的一个关键陷阱：进化不等于变强，可能只是更会考试。
📄 原文摘要(英文)
An LLM agent's capability is largely magnified by its harness, namely the prompts, control flow, tooling, memory, and context management surrounding the frozen backbone model. Recent methods increasingly automate this process by iteratively proposing and selecting component-wise edits of an agent harness, practically establishing a form of recursive self-improvement (RSI) at the agent-system level. However, such recursive evolution may overfit by memorizing the training tasks, showing large in-distribution gains that shrink or even vanish on out-of-distribution benchmarks. We introduce Regularized Recursive Self-Improvement of Agent Harnesses (RRSI), which incorporates the principles of regularizations into harness self-improvement by constraining the evolution candidate proposal and selection. The proposer operates with a temporally annealed budget, limiting how many edits a candidate can bundle, and it encourages unexplored trajectories based on evolution history. The selector is equipped with a critic and a pruner: the critic screens benchmark-specific proposals, while the pruner, removes changes that are too small, too expensive, or no longer useful. Together these constraints favor reusable agent mechanisms over benchmark-specific ones or even noises. Across eight benchmarks spanning coding, agentic workspace and engineering design tasks, RRSI gains up to 14.1 points on the split it evolves against and up to 4.7 points on the five out-of-distribution benchmarks, while producing a harness that runs on 30% fewer policy tokens than the unregularized evolution. Code is available at https://github.com/google-research/rrsi and project page is https://regularized-rsi.com/.