Entropy doesn't track what matters in chain-of-thought, BlackBoxNLP paper finds
gsarti_ · x · 2026-09-18
A paper by Sara Candussio et al. (accepted to the BlackBoxNLP reproducibility track) tests whether entropy in a chain of thought actually tracks important information. The answer is mostly no: the apparent correlation is largely spurious, and the authors identify the confounding cause behind why entropy looked informative — implying prior entropy-based CoT analyses may have been systematically misleading.
More from Research
- Open benchmark and language models for AI in aging biology published in Cell — marinkazitnik · 2026-09-18
- Genomic language model gLM2 controllably generates new biosynthetic domain architectures — BrianHie · 2026-09-18
- TDiMS: fragment distances beat large pretrained models at chromophore property prediction — bravo_abad · 2026-09-18
- UVFaceFusion open-sourced: topology-consistent 3D face reconstruction from multi-view images in under 3s — rsasaki0109 · 2026-09-18
- ProgramAsWeights: compile English-described AI functions into tiny CPU-only neural nets — yuntiandeng · 2026-09-18
- TMLR dumps Medium, launches its own official blog with all posts migrated — thegautamkamath · 2026-09-18