Definition
Plain language
A self-improving AI framework that decides at each step whether to rewrite its own code or retrain its own weights.
As stated in the literature
A self-improvement system (Hexo Labs and Oxford) where a Feedback-Agent reads scaffold source and trajectories to choose between harness edits and weight updates, dynamically selecting RL algorithms when retraining.
Also called: Self-Improving AI
Why it matters: Letting an AI system choose between editing its own code and updating its own weights is a step toward self-improvement that's both more general and more flexible than fixed pipelines.
For example, given poor performance on a benchmark, SIA's feedback agent might decide the problem is a bug in the scaffolding and rewrite the harness — or conclude the model itself needs retraining and pick an RL method to do it.
Heard on the show
“… The paper is "SIA: Self Improving AI with Harness and Weight Updates," from a team at Hexo Labs with a collaborator …”Episode 088 — Two Levers for Self-Improving AI: When Rewriting Code Isn't Enough