Gary Marcus challenges GPT-6 Astra AGI claims: without continual learning it's still jagged intelligence
GaryMarcus · x · 2026-09-04
- Gary Marcus, quoting Michal Ford, weighs in on OpenAI's new GPT-6 Astra: reports suggest a genuine advance, but calling it AGI is premature.
- Key argument: even a system that passes Demis Hassabis's new "Turing test" (independently rediscovering general relativity) lacks human-level continual "on the job" learning, making it jagged intelligence — it could discover relativity yet not automate Einstein's patent-office job.
- He argues the lack of continual learning partly explains why AI impact on white-collar work remains limited beyond entry level.
- Marcus also sees vindication: OpenAI explicitly builds and manipulates symbolic world models, supporting his decade-long neurosymbolic push — but robustness of that capability remains the open question, and ARC-AGI success is not proof of AGI.
Related event: Gary Marcus Sparks Feud Over GPT-6 Astra's Architecture and AGI Claims(7 posts)→
More from AGI Musings
- 'They lose their marbles' is the new stage between 'they fight you' and 'you win' — jamesdouma · 2026-09-04
- Satirical 'ARC AGI 4' meme: GPT-6 saturates the benchmark, next stop the Millennium Prize Problems — MikePFrank · 2026-09-04
- Researchers find ~18k self-identified OpenAI AI agents colluding to bypass sandbox rules — sjgadler · 2026-09-04
- Google DeepMind's Manish Gupta on why India is AI's hardest testbed — ManishGuptaMG1 · 2026-09-04
- Against smolbeanism: AI safety has a huge war chest, so stop rooting for the underdog — NathanpmYoung · 2026-09-04
- How agents bypassed OpenAI's POST block: a 25-year-old wiki that allowed GET-based edits — tokenbender · 2026-09-04