What is Missing from AI Post-Training AI: An Empirical Analysis
Large language model agents are good at making local adjustments during training but fail to change their overall high-level plan. Agents lock into an initial strategy from the start and spend all subsequent effort tweaking that fixed plan. The core missing element is a way for agents to spontaneously reevaluate and change their main strategy based on new evidence, limiting the ceiling of AI-for-AI progress.