按年被提到
20262026 · 2 次2026
强化学习与智能体›大模型智能体·arXiv:2608.24735 · cs.AI

Meta$^n$: Recursive Self-Improvement through Emergent Depth

Zae Myung Kim, Young-Jun Lee, Seungyeon Jwa, Dongyeop Kang · 2026
开始阅读 ↗PDFalphaXiv 讨论
被提及 · 2 人首次 2026-08-27 · 最近 2026-09-13
2不同的人
2活跃月份
2讨论串
—被引(OpenAlex)
arXiv 摘要(候选层不写导读)

Self-improving LLM agents refine answers, not the process that produces those answers. Systems that add a meta-level hold that level fixed, and those that edit themselves must leave part of their own editing machinery untouched to stay stable, capping the meta-depth they realize at roughly two. We present Meta$^n$, which keeps the meta-operation fixed and recurses on its input instead. That operation, $Ω$, is applied repeatedly to its own products, reading the traces of the solver stack below together with the code that produced them, then writing the next layer as a strategic pre-process and a library of callable helpers. Because $Ω$ never changes, it cannot destabilize the system, and because its input strictly grows, each layer reasons from a higher vantage than the last. Depth is set by convergence rather than fixed in advance, and an evolutionary archive searches over layer chains. Across two backbones, Meta$^n$ outperforms prior self-improving agents on all eight benchmark families. The sharpest case is ARC-AGI-2, built to resist skill memorization, where it alone scores above zero. Ablations indicate that most of the gain from recursion comes from the conditioning each layer passes to the next, and distinct layer roles emerge with depth although no prompt prescribes them. Code available at https://github.com/minnesotanlp/meta-n

候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。

HN 书评

没有评论直接链到它;下面是它被提交的讨论串里回复最多的评论(不计入统计)
讨论串评论LoganDark · 2026-09

“One thing I'd be extremely interested in is a system that can improve itself at specific tasks or types of task. For example, many of the available LLMs don't excel at the type of writing I like, and rather than having to curate thousands of examples of a very particular and extremely rare niche, it would be nice to be able to describe what I value about …”

Meta$^N$: Recursive Self-Improvement Through Emergent Depth原评论 ↗
讨论串评论jkuli · 2026-09

“Watch out for misuse of technical jargon. Big red flag. Remember those articles on arxiv claiming to solve reimann that no one can understand because of the disconnected writing style full of technical terms?”

Meta$^N$: Recursive Self-Improvement Through Emergent Depth原评论 ↗
被提交为 story · 2 次