Vibeleaderboard
← All Intel
Intel / post

Mollick: GPT-6.2 Astra is powerful but hard to steer

Source
x.com
Date
Ethan Mollick@emollick

I’ve been using GPT-6.2 Astra prior to launch & I found it very uneven & weird. On one hand, it was harder to steer & work with than other models. On the other, when given an ambitious project (make a space elevator ride from scratch with Blender & Unity) it could deliver well

Context

Ethan Mollick, a Wharton professor who writes about how people use AI, used GPT-6.2 Astra before OpenAI released it and came away with a split verdict: the model was harder to steer and work with than other models, yet it delivered well when he gave it an ambitious build (Mollick). His example was a space elevator ride made from scratch with Blender, the 3D modeling program, and Unity, the game engine. Steering here means getting the model to do what you intend while it works, so his complaint is about the process of working with it, while his praise is about the finished result.

This is a single first-hand impression in a short post. Mollick does not say which other models he compared it with or how he measured the difference, and the attached video was not analyzed beyond an automatic speech transcript that contains only two short spoken lines, so it adds no evidence about how the ride was built or how well it works. The strength he describes does line up with one independent measurement: Artificial Analysis found GPT-6.2 Astra leading its AA-3D test, where human graders rate interactive 3D experiences built by models, at 1318 against 1271 for Anthropic's Claude Fable 5.2. That scores the output only, so it says nothing about the steering problem, which remains Mollick's account.

Terms in this piece · Glossary
  • benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
More from Ethan Mollick
Recommended reads
Comments

Checking sign-in…

Loading comments…