DARLING, a paper fixing repetitive LLM outputs after post-training, accepted at NeurIPS

DanielKhashabi · x · 2026-09-25

The DARLING paper from Daniel Khashabi's group has been accepted to NeurIPS. N8Programs calls it one of his favorite papers and the reason he got excited about joining JHU CLSP.

The problem it tackles: post-training tends to make language models repetitive. DARLING rewards responses that are both high quality and meaningfully different from one another, using a diversity-aware reward to counter mode collapse. Worth a look for anyone tracking post-training diversity.

Related event: DARLING: RL Method Balancing Quality and Diversity in LLM Responses Accepted by NeurIPS(2 posts)→

Original post →

More from Research

Research channel →