Chain-of-Experience for Continual LLM Improvement
The paper tests whether LLMs can improve during inference by accumulating feedback across repeated attempts.
The authors call the setup Chain-of-Experience, using self-feedback and external signals such as correctness or coding test pass rates. Across math, coding, and knowledge tasks on eight models, iterative experience beat feedback-free baselines. They report a 5.6% overall improvement and 19% lower API cost across tasks and models, with most gains appearing early. ArXiv · AI/CL/LG's note
The authors call the setup Chain-of-Experience, using self-feedback and external signals such as correctness or coding test pass rates. Across math, coding, and knowledge tasks on eight models, iterative experience beat feedback-free baselines. They report a 5.6% overall improvement and 19% lower API cost across tasks and models, with most gains appearing early. ArXiv · AI/CL/LG's note
score 5