Harness Engineering for Self-Improvement
- Source
- lilianweng.github.io
- Author
- Lilian Weng
- Date
Context
An can improve its working method by changing the software around its model. Lilian Weng's review examines this surrounding : the machinery that selects , runs tools, preserves work and evaluates results. Weng identifies it as a practical target for improvement without requiring the model to rewrite its own learned parameters.
Persistent files illustrate the mechanism. Experiment logs and error histories can outgrow the amount of text a model can consider at once. Saving them separately lets the agent retrieve relevant evidence and inspect previous failures. That record can inform changes to the workflow, whose usefulness must then be evaluated.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
“an AI uses its current intelligence to improve the cognitive machinery that produces its intelligence”
Checking sign-in…
Loading comments…