Evolutionary Context Search for Automated Skill Acquisition
New evolutionary method discovers non-obvious context pairings that transfer across models like Gemini, Claude, and DeepSeek.
Researchers Qi Sun, Stefan Nielsen, Rio Yokota, and Yujin Tang introduced Evolutionary Context Search (ECS), an evolutionary algorithm that searches context combinations using only inference calls. ECS improved BackendBench by 27% and τ-bench airline by 7% by discovering non-obvious, model-agnostic context pairings. Users can apply ECS to automatically find optimal context for tasks, providing an efficient alternative to manual prompt engineering or costly fine-tuning for skill acquisition.
Why It Matters
Enables AI models to acquire new skills post-deployment without expensive retraining, making knowledge updates faster and cheaper.