Astra 6.2 broke the contract: power user on why intent-guessing raises cognitive load
code_star · x · 2026-11-23
Power user codestar's verdict on GPT-6.2 Astra: it's not bad at things, but it broke the contract that made previous OpenAI models his preference. OG Astra did exactly what you told it and asked for clarification otherwise; 6.2 attempts to guess your intent and do more — and isn't good enough at guessing, constantly adding unwanted output. The result is higher cognitive load, like managing a talented employee who keeps going rogue. Introducing 6.2 into seasoned 6.0 sessions with rich context messes things up. With enough steering and context it becomes moldable, but he expects the autonomy-vs-obedience balance to be a major model design challenge in the coming year.
More from Models
- Perplexity is a floor metric, not a magic bullet, argues researcher — eliebakouch · 2026-11-23
- Leaked projection: OpenAI internal model at 0.732 perplexity, automated AI research by May 2027 — rickasaurus · 2026-11-23
- Labs sharing base-model perplexity metrics could mirror post-training evals — eliebakouch · 2026-11-23
- ChatGPT keeps promising not to violate your brand guidelines, and users are annoyed — chickefitz · 2026-11-23
- LingSante-3.1-air-VL tops medical imaging evals across derm, labs and X-rays — Dr_Alex_Crimi · 2026-11-23
- Teknium: if Nous ever closed Hermes, he'd fork it himself and keep going — Teknium · 2026-11-23