logoalt Hacker News

amluto • today at 7:18 PM • 3 replies • view on HN

I am right now being reminded of a problem with GPT-6 Astra: it surprisingly regularly completes part of a (not very complex) task, gets a bit distracted, and declares itself done. The fact that the original task is not at all complete doesn't seem to bother it, and it kind of seems like this bad state is sticky in the context and it struggles to recover.

I wonder if the newer models are better about this.


Replies

valdiorn • today at 9:53 PM

This was the biggest shock when I moved to Anthropic models. I didn't realise that there were models that could do actual long-term work. I'd give chatgpt a 5 stage project and after doing step 1 it would just go "yeah I'm just quitting now, if you want to keep going here's what you need to do, but I'm done working". It would just never work more than 20-30 minutes.

clarus • today at 7:52 PM

Ask it to create and run a goal about the task. This will make it stick to it.

➕ show 1 reply