Jeremy Howard: Sol 6.2 feels like a real regression — fast answers even on xhigh, but shallow
jeremyphoward · x · 2026-11-19
Fast.ai founder Jeremy Howard reports that after only a few hours with Sol 6.2, it has been a real regression for him. His hypothesis: the model doesn't think enough — it responds very quickly to nearly every question even at the xhigh reasoning setting, but the answers come across as shallow.
More from Models
- Ethan Mollick re-runs his gothic city shader prompt on GPT-6.2 Astra — fully procedural this time — emollick · 2026-11-19
- Early evals: Gemini Neon reportedly beats Opus on long agentic tasks — bindureddy · 2026-11-19
- LocalLLaMA Users Compare Their Main + Backup Model Stacks for Reliability — Vusiwe · 2026-11-19
- Leaked benchmark screenshot hints new model 'Gestalt' beats everything, incl. human golden designs — garrytan · 2026-11-19
- Confused by Gemini's naming mess? This Codex-made diagram sorts it out — chongdashu · 2026-11-19
- Skeptic questions Anthropic's claim of bigger AI speedups than OpenAI's — ziv_ravid · 2026-11-19