Self-improving AI / LM2LM
LM2LM
“we propose LM2LM, which trains LLMs with RL to generate effective LM improvement proposals.”
LM2LM is one of the projects on the self-improving AI map. Below is what it changes about itself, whether the change lasts, and whether it improves the improver, each with the line from its own source.
Where it sits
Columns: what it changes, from the surface to the deepest change. Rows: whether the change outlasts the run. A number is how many projects sit in that square. See it on the 3D map
- What it changes
- Prompts, memory, context“where ht records the proposals executed in previous steps and their measured outcomes”
- Does the change last
- Gone when the run ends“An AutoResearch campaign consists of a sequence of research steps, where each step requires the proposer to decide what proposal to explore next based on the current task”
- Does it improve the improver
- does not improve how it improves“Each experiment trains the target model from scratch.”
- Who sets the goals
- people set tasks and scoring“Each task specifies a model family, model size, pretraining budget, and a reference configuration with validation loss”
- How it was tested
- tested on held-out tasks“No held-out experimental records are used for proposer training, RWM grounding, or calibration.”
Date is arXiv v1. A proposer model, Qwen3.8-27B, is trained with RL to suggest recipe changes for separate target language models; a target model is trained from scratch when a proposal is actually run. The proposer does not change itself or its successor, so on the changes axis only its working context changes. Its weights are updated by the authors' RL loop, with rewards predicted by a frozen GPT-6-sol research world model whose in-context records are topped up with real runs three times, for 200 real runs in all. The paper calls this a potential path toward recursive self-improvement. The 10 held-out tasks differ from training tasks in model size or training budget; a separate run holds out the whole Qwen3 family.
Does it count as RSI?
It depends on the definition. Under the four used on What counts as RSI, by what its own source says:
- Loose (Eliezer Yudkowsky, Intelligence Explosion Microeconomics (2013)): does not count
- Durable (Ren et al., self-improvement survey (2026)): does not count
- Strict (Roman Yampolskiy, From Seed AI to Technological Singularity (2015)): does not count
- Autonomous (Anthropic, When AI builds itself): does not count
How it got here
Earlier work in this catalog that LM2LM cites, oldest first. Under each: where that work sits on the map, then where LM2LM sits (changes in bold), and what LM2LM's own paper says about it. Its whole family on the citation map
Cites STaR 2022, Self-Rewarding Language Models 2024, The AI Scientist 2024, Gödel Agent 2024, SICA 2025
In the same square
ADAS: Automated Design of Agentic Systems, AlphaEvolve, AREX-2, MLEvolve, ParetoRSI-Gym, PostTrainBench, Promptbreeder, Reflexion, Self-Refine, The AI Scientist