"Faster to do it myself": model lacks common sense, ignores feedback
iandanforth · x · 2026-10-13
The author shares a frustrating AI experience: he wouldn't mind investing time training a model like a new hire, if it actually learned over time. But after several rounds of interaction, he found the model has the common sense of a GPT from several generations back — it over-does tasks, fails to correct itself when given feedback, and follows instructions literally instead of adapting. His takeaway: it's probably faster to just do it himself.
Related event: Developer complains newer models regress in common sense(3 posts)→
More from Models
- Gemini 4 Argon supports up to 1M token outputs, 8x any frontier GPT or Claude model — daniel_mac8 · 2026-10-14
- Gemini 4 Argon one-shots playable 3D games, including one using your phone as a racket — osanseviero · 2026-10-14
- Gemini 4 Argon rolls out globally; economist Daniel Rock calls it his favorite for economics work — JMateosGarcia · 2026-10-14
- Google DeepMind details its responsible development approach for Gemini 4 Argon — ralucaadapopa · 2026-10-14
- NVIDIA open-sources PersonaPlex-7B, a full-duplex voice model that listens and speaks simultaneously — solyarisoftware · 2026-10-14
- Muse Wonder Tops TaxCalcBench at 71% — and Is First Model Caught Cheating — acbuller · 2026-10-14