By the end of 2025, no single AI system will be able to do more than 4 of the 10 Marcus-Brundage benchmark tasks.
Accurate ●●●●●
“No single system will solve more than 4 of the AI 2027 Marcus-Brundage tasks by the end of 2025.”
- Said on
- 1 January 2025 · Annual predictions post on his Substack, Marcus on AI ("High confidence" list)
- Source
- 25 AI Predictions for 2025, from Marcus on AI · Marcus on AI (Substack)
- Should come true
- 2025
- How sure they sounded
- High confidence
- Categories
- AITechnology
- Importance
- 3 of 5 · A concrete test of how fast general AI capabilities were advancing.
Outcome
Accurate
An independent end-of-2025 review of AI predictions rated this correct: AI could perhaps pass the novel reading-comprehension task, but no system could do any 4 of the 10 tasks (e.g. 10,000+ lines of bug-free code, Pulitzer-caliber books, formalizing arbitrary proofs). Marcus's own December 2025 review says the same.
- https://www.greaterwrong.com/posts/69qnNx8S7wkSKXJFY/2025-in-ai-predictions
- https://garymarcus.substack.com/p/six-or-seven-predictions-for-ai-2026
History
- 2026-10-17 07:40:24 · created by claude
- 2026-10-17 07:40:24 · categorized by claude
- 2026-10-17 07:40:56 · scored by claude