RLAF: Nace AI trains decision models with a reasoning model as judge
nischay_twt · x · 2026-11-07
In a new paper on Foundation Decision Models, Nace AI researchers propose Reinforcement Learning from Adjusted Feedback (RLAF): a reasoning model that spends tokens on thinking trains another model to make decisions without generating answer tokens. RLAF combines verified outcomes with probabilities assigned by a generative judge — for Drex Max, a Qwen3.8-27B base model run at high reasoning effort — instead of training against a single correct label.
More from Models
- Claude now casually generates design concept visuals while working, user delighted — RileyRalmuto · 2026-11-07
- New AI lab reportedly squeezes more from 12T tokens than DeepSeek gets from 45T — teortaxesTex · 2026-11-07
- Qwen 4 Preview surfaces in YouTube video, hinting at Alibaba's next-gen model — opoot_ · 2026-11-07
- User shares approval email for Anthropic's Cyber Verification Program Defense Access — AIFlow_ML · 2026-11-07
- As ChatGPT ends unlimited chat, Plus subscribers weigh canceling and moving to Claude — ShinyPlasm · 2026-11-07
- Qwen 4 Preview Appears on Arena, Qwen 4 Max Rumored With 1M-Token Context — WorldofAI · 2026-11-07