A resurfaced 2029 AI prediction list reads like a benchmark in human skepticism
Peanut_Extreme_8208 · reddit · 2026-07-25
An old screenshot of “Marcus on AI” lays out five 2029 predictions about AI capability limits.
The list says AI still won’t be able to:
- understand movies deeply enough to explain plot, characters, and motivations,
- reliably answer questions about novels,
- work as a competent cook in an arbitrary kitchen,
- build more than 10,000 lines of bug-free code from natural-language specs,
- or convert mathematical proofs from prose into symbolic form for verification.
The image reads like a benchmark of human-era skepticism, and it is being shared as a dated prediction set worth revisiting.
Related event: Gary Marcus Makes Heavy Bets Against AI Capabilities by 2029(3 posts)→
More from AGI Musings
- A handful of foreign tech firms could hold a kill switch over entire enterprises — yacineMTB · 2026-07-25
- Compute Density Sets Intelligence Ceiling: Exploring AI's Physical Limits — jachiam0 · 2026-07-25
- Gary Marcus and Grady Booch Debate: How Many Generations Until AGI? — GaryMarcus · 2026-07-25
- ‘AI psychosis’ turns into a self-owning AI-world meme — ctjlewis · 2026-07-25
- A long argument says AGI is arriving gradually as models and products co-evolve — dotey · 2026-07-25
- Vivek Haldar backs KYC for unrestricted model access and post-hoc enforcement — vivekhaldar · 2026-07-25