Post-training from positive examples alone: 235B LLM writes indistinguishable from humans
AllThingsApx · x · 2026-11-12
A team led by andrewwhite01 presents a method to bootstrap post-training using only positive examples, validated on arithmetic, molecular structures, and scientific writing. They trained a 235B LLM whose writing is claimed to be indistinguishable from human writing; the shared visualization shows the model's reasoning evolving from generic filler to specific, hypothesis-driven grant-proposal text during training.
More from Research
- CoRL 2026: papers say teleop is optional, but demo-floor humanoids still rely on it — jonstephens85 · 2026-11-12
- Probing Qwen3-8B: Action Embeddings Land on the Right Body Parts, Up to 100% — Sauers_ · 2026-11-12
- Parallel experimentation beats picking one path in AI-powered math formalization, argue researchers — xennygrimmato_ · 2026-11-12
- Structure prediction agent: an agent picks the best protein folding model and learns over time — nlarusstone · 2026-11-12
- Pretraining in the agentic era: inject multi-turn tool patterns into CPT, don't rely on NLL alone — rdesh26 · 2026-11-12
- Fleuret: Rethinking AI data acquisition and processing entirely would be a huge booster — francoisfleuret · 2026-11-12