Gary MarcusCognitive scientist and generative-AI sceptic; makes public bets against near-term AGI claims.
A safe bet is that GPT-5 will be a lot like GPT-4, and do the same kinds of things and have the same kinds of flaws, but be somewhat better.
Admitted by the site's editor alone — Seeded by the site owner at launch; single-operator admission (ADR-0012).
Normalized claim
GPT-5 will be a lot like GPT-4, with the same kinds of flaws, only somewhat better.
Confidence
high (80%)
Deadline
Aug 7, 2025 · inferred
Resolution rule
Hit if GPT-5 at release was an incremental improvement over GPT-4-class models with hallucinations and reasoning failures of the same kind, per independent evaluations and reviews. Miss if it was broadly judged a qualitative leap.
Crowd vote
No crowd votes were recorded.
Voting closed at the deadline.
More from Gary Marcus
If there exist AI systems that can perform 8 of the 10 tasks below by the end of 2027, as determined by our panel of judges, Gary will donate $2,000 to a charity of Miles' choice; if AI can do fewer than 8, Miles will donate $20,000 to a charity of Gary's choice.
OPEN