This could be an interesting new paper by Maosong Sun and his team.
From the abstract:
"Large language model (LLM) agents can now post-train an LLM end-to-end. They can write code, launch training, evaluate checkpoints, and improve downstream performance, raising the prospect of AI-for-AI.
"Large language model (LLM) agents can now post-train an LLM end-to-end. They can write code, launch training, evaluate checkpoints, and improve downstream performance, raising the prospect of AI-for-AI.
We argue that this picture conflates two distinct capabilities:
execution-level capability, iterating within a selected training strategy; and strategy-level capability, revising the high-level judgment as experimental evidence accumulates.
Analyzing a large corpus of publicly released post-training trajectories, we find that across different tasks, the agent's training strategy is locked in at the very beginning, and the entire remaining budget is spent on local adjustments within the selected strategy.
We then examine three natural explanations -- missing experience, missing guidance, and insufficient reasoning--with escalating interventions.
Extensive experiments show that
(1) an experience-driven scaffold improves execution across the board (+12.6 points on GSM8K and +40.8 on HumanEval) but leaves the strategy static;
(2) human guidance effectively redirects the initial strategy, yet the agent falls back into local adjustment loops once training starts; and
(3) additional inference compute pays off on easier tasks but yields almost no gain on the hardest one.
In conclusion, what agents lack is neither experience, guidance, nor reasoning compute, but a mechanism for spontaneously reevaluating their strategy during execution."
