Prompt EngineeringShiyan Liu et al.

Liu et al. publish Reflection in the Dark

Published Reflection in the Dark — demonstrated conditions under which reflective prompt optimization worsens accuracy from 23.81% to 13.50%

Why it matters

The first to isolate a failure mode of GEPA-style automatic prompt optimization

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2603.18388

Added in Week 35, 2026 · See that week

Same week

Hugging Face archives the text-generation-inference repositoryOfficialMistral AI releases LeanstralOfficialMistral AI releases Mistral Small 4Official

Nearby on the same axis

Reasoning Models Struggle to Control their Chains of Thought publishedPaperMechanistic Evidence for Faithfulness Decay in Chain-of-Thought ReasoningPaperConformal ThinkingPaperIwase et al. publish Reliable Chain-of-Thought via Prefix ConsistencyPaper