Gary Marcus on GPT-6 Astra: symbolic world models are vindication, but no proof of AGI
GaryMarcus · x · 2026-09-04
Gary Marcus published a take on OpenAI's rumored GPT-6 Astra, challenging Greg Brockman's AGI claims:
- Multiple reports suggest a genuine, impressive advance.
- After nearly a decade of campaigning for (neuro)symbolic world models, he finds it "extraordinarily vindicating" that an OpenAI product explicitly creates and manipulates symbolic world models during computation.
- The key open question is how robust that capability actually is.
- Success on ARC-AGI is impressive but not proof of AGI despite the benchmark's name; he expects many failures to surface.
The retweeter adds that, as a scientist, it's disappointing how little we still know about how the system actually works.
More from AGI Musings
- DeepSeek as a 'Power Object': the wave of takes reveals more about us than it — dbreunig · 2026-09-04
- Ex-OpenAI safety lead Miles Brundage: if your primary emotion on AI isn't concern, you're misreading it — Miles_Brundage · 2026-09-04
- AI Job Market Talk: GenAI Engineers With 3-5 Years Experience Command ₹2-3 Lakh Monthly Pay — ashishllm · 2026-09-04
- Alignment researcher: defining the AI's value system isn't the real problem — Sauers_ · 2026-09-04
- Researcher: LLMs' hidden cost of wasting your time on useless work is underrated — lateinteraction · 2026-09-04
- Andrew Chen: agents as your C-suite works at work — what's the personal-life equivalent? — andrewchen · 2026-09-04