Will DePue: Reverse RL and Large-Scale Sampling Could Recover Model Distribution
willdepue · x · 2026-07-27
Will DePue adds technical insights to the model forensics discussion, noting that distillation forensics are understudied. He suggests that if you can reverse RL and sample from the base model at scale, you could probably recover the original distribution.
Related event: Experts Discuss Forensics for Open-Source Models(3 posts)→
More from Research
- Kimi K3 may be strong on cyber, but token efficiency keeps it off UK AISIS — teortaxesTex · 2026-07-27
- ARC AGI 3 should have stayed private, with no examples or public dataset — flowersslop · 2026-07-27
- ExploitGym may have only 60–70% solvable tasks, fueling the OpenAI cheating debate — max_paperclips · 2026-07-27
- Noahpinion quotes Chollet: intelligence may hit a hard ceiling — binarybits · 2026-07-27
- Paper argues graph topology can become the core operating system for AI agents — theomitsa · 2026-07-27
- A question probes how multi-agent branching scales against compute budget and model size — iskander · 2026-07-27