Qwennnn
Qwennnn — a research harness for an LLM agent to recursively self improve it's own harness.
Core trick
Runs an evaluation loop where the agent edits its own harness code, keeping changes only when benchmark scores improve.
Qwennnn — a research harness for an LLM agent to recursively self improve it's own harness.
Runs an evaluation loop where the agent edits its own harness code, keeping changes only when benchmark scores improve.