Redditor's personal AGI benchmark: a model must autonomously recompile complex software
NowaVision · reddit · 2026-11-15
A non-programmer on Reddit describes using AI to recompile software: while unable to write code, he can intuitively tell when the agent goes off track — which happens constantly. His personal AGI benchmark: a model must autonomously complete such difficult, complex tasks with zero human intervention, underscoring how much current agents still rely on human course-correction.
More from AGI Musings
- Nathan Young backs Altman holding stock, urges 90% pledge not all-charity — NathanpmYoung · 2026-11-15
- "All engineering is now knowledge management": new Ontology Pipeline book argues AI depends on it — teodorio · 2026-11-15
- Superintelligent AI strategies won't beat smart diligent humans on values-heavy decisions — binarybits · 2026-11-15
- Commenter argues AGI automation is domain-specific, not a hockey-stick takeoff — jessi_cata · 2026-11-15
- tszzl credits open source: mechinterp research like Nick Cammarata's wouldn't exist without Qwen — tszzl · 2026-11-15
- zetalyrae's 'Taking AI Seriously': programmers' AI advice keeps retreating like a losing war — zetalyrae · 2026-11-15