Heavy AI user: marketed automations are fragile, exhausting, and dumber than we admit
mrks86 · reddit · 2026-10-24
A self-described heavy AI user on Reddit (top Anthropic and OpenAI models for university coding, hardware development, and automation) posted an "unpopular opinion": AI is far less intelligent than the hype suggests.
- He tried three times to automate an email-writing + Google Sheets workflow with Strawberry Browser; setup was fine but error rates were high, and long-term monitoring/maintenance proved more exhausting than doing it manually.
- AI helps a lot in large repos and hardware development, yet the models' lack of basic real-world understanding is still mind-blowing, and they tend to power through problems instead of solving them.
- His verdict: "powerful, yes; intelligent, absolutely not" — a PhD student acting like these models would be dismissed.
- On superintelligence doom: the real risk isn't AI being too smart, but too stupid to know when to stop.
More from AGI Musings
- Mathematician: AI is revealing the space of mathematical ideas as hyperbolic — CatAstro_Piyush · 2026-10-24
- Altman: models find hardware-level bugs 'outside the normal threat model', patching won't scale — soldierofcinema · 2026-10-24
- Reading《菌落万象》: intelligence is just an anthropocentric benchmark AI is about to top — oran_ge · 2026-10-24
- ICU-REACT paper: teaching LLMs how ICU physicians reason beats pure scaling — SoloGen · 2026-10-24
- CS grad: AI is killing my motivation to learn anything new — kubika7 · 2026-10-24
- When every VC uses the same AI for diligence, non-consensus investing gets harder — vaibhavbetter · 2026-10-24