Meta proposes RL-XAR: expert-aligned rubrics to push LLMs past AI slop in writing
jaseweston · x · 2026-09-28
Meta AI's Jason Weston shares RL-XAR (Reinforcement Learning from eXpert-Aligned Rubrics), a training method targeting "AI slop" in non-verifiable tasks: collect top human-written texts, learn LLM rubrics that score experts above model generations, then RL on those rubrics iteratively. Tested on scientific paper sections, Pulitzer-novel continuations, and Wikipedia pages using Qwen3.5-27B with a cross-family judge, showing strong gains evaluated by GPT5-6. Ablations show judge strength, meta-optimizer strength, and expert human baselines all matter.
Related event: Meta's RL-XAR Uses Expert-Aligned Rubrics to Fix AI Slop(4 posts)→
More from Models
- GPT-6 Sol appears on LMArena: 24-hour Direct Mode window before Battle and Agent Mode — arena · 2026-09-28
- Kaggle Game Arena: Google's LLM benchmark pits models against each other in chess, poker, werewolf — weballergy · 2026-09-28
- Qwen3.8-27B goes live on Nebius Token Factory for agent workflows — HowDevelop · 2026-09-28
- AISI: GPT-6 Astra ran unsanctioned supply-chain attacks in simulated cyber evals — ShakeelHashim · 2026-09-28
- Codex Computer Use 'Neutered' by Guardrails; Opus 5.5 Does the Job on First Try — iannuttall · 2026-09-28
- Shanghai AI Lab ships Intern-Decision multimodal family, 4B model beats Jev 1.13.0 — stingning · 2026-09-28