Self-improving AI / STOP: Self-Taught Optimizer
STOP: Self-Taught Optimizer
“In this work, we use a language-model-infused scaffolding program to improve itself.”
STOP: Self-Taught Optimizer 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
- Tools, harness, its own code“we use a language-model-infused scaffolding program to improve itself”
- Does the change last
- Becomes the next version“This is iterated for a prespecified number of iterations T, per available resources.”
- Does it improve the improver
- improves the improver“the resulting improved improver generates programs with significantly better performance than its seed improver”
- Who sets the goals
- people set tasks and scoring“improves an input program according to a given utility function”
- How it was tested
- tested on held-out tasks“we find the improved improver, detailed in Appendix H, outperforms the seed improver on each new downstream task without further optimization”
The authors say this is not full recursive self-improvement because the language models are not altered. With GPT-3.5 and Mixtral, performance degraded.
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)): counts
- Durable (Ren et al., self-improvement survey (2026)): counts
- Strict (Roman Yampolskiy, From Seed AI to Technological Singularity (2015)): counts
- Autonomous (Anthropic, When AI builds itself): does not count
How it got here
Earlier work in this catalog that STOP cites, oldest first. Under each: where that work sits on the map, then where STOP sits (changes in bold), and what STOP's own paper says about it. Its whole family on the citation map
1965 Good, the intelligence explosion builds on it
A definition: the intelligence explosion
“The broader concept of RSI dates back at least half a century, formalized by Good (1966) and later by Schmidhuber (2003), but our work represents a more modest and specific application of these ideas.”
2003 Schmidhuber, the Gödel machine sets itself apart
A definition: the Gödel machine
“Unlike in some prior work (e.g., Schmidhuber, 2003), any improvement attempt may result in worse performance, which can lead to further deterioration.”
2007 Hall, self-improving AI sets itself apart
A definition: self-improving AI
“Regarding implementations, while efforts have been made for G ¨ odel machines (Hall, 2007; Steunebrink & Schmidhuber, 2012), our work is first to leverage LMs for recursively self-improving code generation.”
2023 STOP
Tools, own code · becomes the next version · improves the improver
Also cites Yampolskiy, recursive self-improvement 2015, STaR 2022, Large Language Models Can Self-Improve 2022, Reflexion 2023, Voyager 2023, Promptbreeder 2023
Cited by Gödel Agent 2024, SICA 2025, Darwin Gödel Machine 2025, Hyperagents 2026, Ren, self-improvement 2026, Weco AIDE² 2026, Recursive Harness Self-Improvement 2026, Fibona Agent 2026, Harness² 2026, ECG-RSI 2026, Self-HPO 2026, Organon 2026, RSI-Opt 2026, Pu, recursive self-improvement 2026, Skill∞ 2026, Infinity Machine 2026
In the same square
Darwin Gödel Machine, Dream-RSI, Hyperagents (DGM-H), Infinity Machine, RSI-Opt, SICA: A Self-Improving Coding Agent, Skill∞, VISTA, Weco AIDE²