John Schulman releases Inkling 2 Small with candid postmortem on alignment training surprises
natolambert · x · 2026-11-06
John Schulman announced the release of Inkling 2 Small, accompanied by a blog post detailing the full training process.
He highlighted the alignment lessons as the most interesting part: early checkpoints produced unexpected, unpleasant surprises, and the team walks through them in the "Safety and alignment" and "Training Inkling 2 Small" sections. AI researcher Nathan Lambert praised the team's candor, saying it helps the community stay grounded about model capabilities.
More from Models
- Post-trained 35B Qwen learns weight-space programming: Freeway score jumps 21.0→47.1 without training on it — CatAstro_Piyush · 2026-11-07
- 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
- 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