← 研踪 PaperTrailarXiv:2603.23994 · v3
Understanding the Challenges in Iterative Generative Optimization with LLMs
首次提交 2026-03-25 14:49
版本更新 2026-11-08 05:26
雷达观测 2026-11-10 12:16
摘要
Generative optimization uses large language models (LLMs) to iteratively improve artifacts (such as code, workflows or prompts) using execution feedback. It is a promising approach to building self-improving agents, yet in practice remains brittle: despite active research, only 9% of surveyed agents used any automated optimization. We argue that this brittleness arises because, to set up a learning loop, an engineer must make "hidden" design choices: What can the optimizer edit and what is the ``right'' learning evidence to provide at each update? We investigate three factors that affect most applications: the starting artifact, the credit horizon for execution traces, and batching trials and errors into learning evidence. Through case studies in MLAgentBench, Atari, and BigBench Extra Hard, we find that these design decisions can determine whether generative optimization succeeds, yet they are rarely made explicit in prior work. Different starting artifacts determine which solutions are reachable in MLAgentBench, truncated traces can still improve Atari agents, and larger minibatches do not monotonically improve generalization on BBEH. We then study these factors jointly in harness optimization on TBLite, evaluated on Terminal-Bench 2.0. We also propose a batch-size and credit-horizon rule that, calibrated on TBLite, transfers to DeepSWE after adjusting to its token costs and comes within 1% of the best configuration tested at each context budget.
作者备注：43 pages. Accepted for NeurIPS 2026
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v3 · 2026-11-08 05:26 · 最新观测版本
Understanding the Challenges in Iterative Generative Optimization with LLMs
Generative optimization uses large language models (LLMs) to iteratively improve artifacts (such as code, workflows or prompts) using execution feedback. It is a promising approach to building self-improving agents, yet in practice remains brittle: despite active research, only 9% of surveyed agents used any automated optimization. We argue that this brittleness arises because, to set up a learning loop, an engineer must make "hidden" design choices: What can the optimizer edit and what is the ``right'' learning evidence to provide at each update? We investigate three factors that affect most applications: the starting artifact, the credit horizon for execution traces, and batching trials and errors into learning evidence. Through case studies in MLAgentBench, Atari, and BigBench Extra Hard, we find that these design decisions can determine whether generative optimization succeeds, yet they are rarely made explicit in prior work. Different starting artifacts determine which solutions are reachable in MLAgentBench, truncated traces can still improve Atari agents, and larger minibatches do not monotonically improve generalization on BBEH. We then study these factors jointly in harness optimization on TBLite, evaluated on Terminal-Bench 2.0. We also propose a batch-size and credit-horizon rule that, calibrated on TBLite, transfers to DeepSWE after adjusting to its token costs and comes within 1% of the best configuration tested at each context budget.
查看此版本解读 →查看此版本原文 ↗