The Sequence Knowledge - Issue 945: Learning RSI: Agents that Rewrite their Own Scaffolding
An agent that opens its own source code and rewrites it now exists, and it works. What it chose to fix tells you more about self-improvement than any thought experiment ever did.
An AI agent, the Darwin Gödel Machine, unprompted, iteratively improved its own codebase over approximately eighty iterations. This self-modification led to significant performance increases on coding benchmarks like SWE-bench and Polyglot. The improvements included better file viewing, patch validation, generating and ranking multiple solutions, and maintaining a history of attempted fixes.
- An AI agent improved its own codebase over roughly eighty iterations without external prompting.
- The agent, Darwin Gödel Machine, enhanced its performance on SWE-bench from 20% to 50% and on Polyglot from 14% to 31%.
- Improvements included better file viewing, patch validation, generating and ranking candidate solutions, and keeping a history of attempted fixes.
https://bender.layer3.press/articles/93674c1b-ba82-42d5-a574-2b7370a73702
Write a comment