AI Builder Pushes Back on AGI-2027: Jaggedness Persists, Models Still Ignore Simple Instructions
ChrisGPT · x · 2026-11-14
Responding to Leopold Aschenbrenner's claim that AGI by 2027 is plausible via straight-line extrapolation, the author argues jaggedness persists: even highly capable models like Fable 5.2 fail to follow simple directions in animation tasks. They want a real epistemic continual-learning harness or parametric learning stack, and concede these gaps could plausibly be solved by 2027.
More from AGI Musings
- Dario Amodei says models now tackle Fields Medal-level problems, biology next — QuanquanGu · 2026-11-14
- NY Fed survey: 87% of Wall Street pros now name AI the top financial risk — rohanpaul_ai · 2026-11-14
- Alignment debate: evolutionary psychology is a prerequisite for deconfusing AI motivation — jd_pressman · 2026-11-14
- Why Complex Systems Fail Suddenly: Mollick's Semi-Annual Reread of the 18 Laws — eldonredwards · 2026-11-14
- binarybits vs Chollet: 1960s-vs-today living standards suggest no wave of breakthroughs is due — binarybits · 2026-11-14
- Gary Marcus argues AI should get airplane-style regulation like aviation did — GaryMarcus · 2026-11-14