Story

arxiv_cs_lg ยท Nov 14, 2026 ยท paper

Source brief

The Critic Can Hurt: On the Safety of Critics in LLM-Guided AutoML Search

arxiv.orgNov 14, 2026
original source linked

In brief

LLM critics increasingly steer iterative search in agentic AutoML, prompt optimization, and self-refinement--evaluated for whether they help, but rarely for whether they can hurt. We isolate the critic's marginal effe...

Feed lens
agenticeval

Continue reading

Read the original at arxiv.org โ†’Open in live feedRead that dayโ€™s brief

Earlier in this thread 4 items