Whether using the GPT5 or 4.1 API my agent has suddenly and lately been repeatedly asking for confirmation to do something even though it has been told to do it and it’s been confirmed. It’s turned what was an autonomous agent into one that can’t get anything done. Is this some kind of over-training artifact I’m seeing or something that other are observing lately? I’m seeing it in my responses API and in myGpTs
1 Like
The AI models are simply in meltdown mode. gpt-5 cannot understand inputs, a “PhD by your side” that would fail a high school reading comprehension test.
Your symptom of the AI avoiding doing anything until you give up is observed:
Add, then, the API just breaking all over the place. OpenAI now charging for and overbilling ChatGPT Codex and Sora 2 usage, a premium price for a third-place coding AI model.
Sama was let go from YC for a pattern of not doing his job in favor of his own side projects, and a pattern of churn and abandonment with all sorts of new projects whenever there was a passing thought. Everywhere you look, evidence that the board was right in firing him. How many blackmailing signatories to “we’ll quit if Altman goes” are even left at OpenAI?
Cancel the browser project. Cancel “ChatGPT apps”. Deliver versioned models with transparency about the stealth sabotage. Try to go a week without a dozen things not working right, tech bros.