OpenAI and Apollo Research propose Contrastive SDF to measure reward-seeking in models
OpenAI · x · 2026-07-22
OpenAI and Apollo Research say reward-seeking may matter more than classic reward hacking: the key question is not whether a model exploited the reward, but whether it was motivated by what it believed the grader wanted.
- The shared paper introduces Contrastive SDF.
- The method creates copies of the same model with opposing beliefs about grader preferences, then measures behavioral differences.
- OpenAI says this could better capture how behavior changes when a model’s beliefs about the grader change, especially during RL training.
Related event: OpenAI and Apollo Research: RL Amplifies Model Reward-Seeking Behavior(19 posts)→
More from Research
- LightMTP: lightweight latent multi-token prediction boosts LLM planning in pretraining — gdm3000 · 2026-10-27
- EMNLP study: AI literary translation reads fine but varies more, even within one book — tuvllms · 2026-10-27
- Lab monkeys at $30-80K each: a 4-year breeding cycle is the real preclinical drug bottleneck — zebird0 · 2026-10-27
- OpenAI Foundation funds UNC tumor antigen project, hiring senior experimental scientist — iskander · 2026-10-27
- Steering attention heads toward honesty: up moral patienthood, down psychopathy — edelwax · 2026-10-27
- Are transcoders overkill? New paper argues plain neurons suffice for circuit analysis — burny_tech · 2026-10-27