Power user burns through Pro limits on Astra, verdict: not AGI, just a bigger code monkey
GaryMarcus · x · 2026-09-08
Gary Marcus amplified a hands-on assessment of GPT-6 Astra. After burning through Pro limits twice using Astra exclusively, the author's verdict: a good model, but not opinionated and lacks taste — it doesn't really know what to do, so it just does whatever, wasting your time and tokens.
The author also doubts the 10T+ parameter rumors, guessing 6-8T by vibe, and concludes it's still "just a code monkey, although a slightly larger one and the best we have."
More from AGI Musings
- OpenAI Chief Scientist: Models Could Get Much Better at Math, but RSI and Alignment Come First — Dr_Singularity · 2026-09-08
- Engineer Says He Automated Himself Out of One of the World's Most Technical Jobs — MickeySteamboat · 2026-09-08
- Was the Golden Gate Claude Era the Peak of AI Alignment? — sprooos · 2026-09-08
- Mathematician Andrew Snowden's new preprint used ChatGPT for many arguments — littmath · 2026-09-08
- OpenAI chief scientist Jakub Pachocki: models could be much better at math, but RSI and automated alignment come first — Dr_Singularity · 2026-09-08
- The Superintelligent Dog Analogy: Can High-Dimensional AI Thought Survive Lossy Compression — MajmudarAdam · 2026-09-08