PARTIALStated March 10, 2022
Gary MarcusAI researcher; author of “Marcus on AI”
"Deep learning is hitting a wall" — scaling makes models more fluent but not more trustworthy, with diminishing returns on reasoning, truthfulness and common sense.
LLM limitsscaling
- Verdict
- Partially correct
- Deadline in claim
- No fixed date
Assessment
Mixed. Models improved far more than “wall” rhetoric implied, yet the specific reliability problems Marcus flagged (hallucination, reasoning, common sense) persist and the scaling-returns debate is unresolved. Scored partial.
Put this through The Wringer →Primary sources
Links open the original publications. Quotations are reproduced for the purpose of commentary and criticism. Spot an error? Corrections policy.