Gary Marcus argues scaling-only AI has reached critical limits
Gary Marcus argued that persistent failures in hallucination, reasoning, generalization, and reliability show that more data and compute alone will not produce trustworthy general intelligence. He called for structured alternatives including world models and neurosymbolic methods.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why it moved the index
A credible ceiling on scaling-only systems would preserve human advantage and slow progress toward highly capable general agents. Magnitude is 48 because the claimed limit affects a central capability pathway; confidence is 52 because the essay supplies examples and reasoning but not an independent systematic demonstration, and the DoomBench effect requires one inference.
Assessment history
-
R1
Away 48 · confidence 52
New historical evidence from the Gary Marcus backfill; the scaling-limit mechanism was not in the durable catalogue.
16 Aug 2026