New paper: value training in LLMs causes unintended side effects on other values
IAugenstein · x · 2026-11-18
Researchers from the University of Copenhagen and University of Cambridge (Siddhesh Pawar, Chen Cecilia Liu, Anna Korhonen, Isabelle Augenstein) released the paper "Unintended Side Effects of Value Training in LLMs".
- Context: aligning LLMs with human values is now a common post-training goal, but values are interconnected
- Gap: current work only checks whether the target value moved, not what else changed
- The paper examines unintended side effects of value training on non-target values
It adds a measurement lens for "train one value, others drift too" in alignment research.
More from Research
- Google-Backed Kaggle Agent Contest Offers $400K to Crack Engineering Benchmark PLE — amankhan · 2026-11-18
- Benchmark hacking? Labs' own SWE-bench scores beat independent runs by 4-5 points on average — RexDouglass · 2026-11-18
- NeurIPS paper: hidden measurement error in LLM pipelines distorts evals — and grows with more data — RexDouglass · 2026-11-18
- MLCB and MLSB merge into MLBio, first joint conference set for 2027 — anshulkundaje · 2026-11-18
- bf16 breaks VLA fine-tuning: one of the month's most interesting robotics papers — m_wulfmeier · 2026-11-18
- Europe launches ELAI, a virtual frontier AI institute spanning 17 countries and 33 institutions — IAugenstein · 2026-11-18