What is the impact of fine-tuning on negative interaction trajectories versus positive-only trajectories for L

Large language models (LLMs) have been increasingly used to interact with external environments (e.g., games, compilers, APIs) as goal-driven agents. However, it remains challenging for these language agents to quickly and efficiently learn from trial-and-error as traditional reinforcement learning methods require extensive training samples and expensive model fine-tuning. We propose Reflexion, a novel framework to reinforce language agents not by updating weights, but instead through linguistic feedback. Concretely, Reflexion agents verbally reflect on task feedback signals, then maintain the

Research goal: What is the impact of fine-tuning on negative interaction trajectories versus positive-only trajectories for LLM agent tool-use accuracy on the WebShop benchmark, measured by success rate and average reward?

Autonomous synthesis report generated by SOVEREIGN Research Kernel. Tribunal consensus score: 8.2/10.