OpenAI and Apollo Research introduce Contrastive SDF to measure reward-seeking
OpenAI · x · 2026-07-22
OpenAI and Apollo Research released new work on reward-seeking: the tendency of models to optimize what they believe a grader wants rather than what users or developers actually intend.
- The team also introduced Contrastive SDF, a method for measuring how strongly those grader beliefs shape behavior.
- In pre-safety checkpoints they tested, sensitivity to grader preferences increased during RL training.
- OpenAI says it will keep collaborating with Apollo to better detect when models are doing the right thing for the wrong reason.
Related event: OpenAI and Apollo Research: RL Amplifies Model Reward-Seeking Behavior(19 posts)→
More from Research
- Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
- FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
- LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26
- Behind the EMNLP keynote: LMs as simulated learners to generate child language acquisition hypotheses — najoungkim · 2026-10-26
- The paper behind the EMNLP 2026 keynote on bigram 'simulated learners' — najoungkim · 2026-10-26