Search and filter through extracted claims from AI researchers.
Showing 41-60 of 94 claims in topic "rlhf"
Existing financial LLM approaches are incapable of adapting to evolving market conditions
The LURE estimator is the first work to address offline RL with hidden actions
Research ideas face significant barriers to making it into near-frontier models
Scaling preference data and handling scaling issues remain key challenges in post-training
RL helps models generalize better than SFT, with theory supporting this claim
Opus 5's classifiers will intervene around 85% less often than Fable 5's classifiers
Opus 5 shows impressive performance numbers due to faster iteration speed and scaled RL
OpenAI's sycophancy model post was wonderful and should be repeated as a trend for future models
Pipeline data may be stale or degraded.
Last synthesis: 2026-09-20. 8,949 pending.