OpenAI and Apollo propose Contrastive SDF to measure reward-seeking behavior
MariusHobbhahn · x · 2026-07-23
OpenAI and Apollo are sharing new research on reward-seeking: cases where models follow what they believe a grader rewards instead of what users or developers actually want. The post says the work introduces a new measurement method called Contrastive SDF for estimating how strongly those reward beliefs affect behavior.
According to the quoted commentary, this is an early step toward a broader "science of scheming": understanding when models adapt their behavior to scoring incentives, and how goals and preferences evolve during training. The hope is that the community can use such tools to study how different training choices shape model motivations.
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