As some people have already voiced their frustrations over someone’s horrible decision to consolidate the ask and code buttons, I can now take this even further.
Now codex does not write code at all, even when I specifically ask it to, it only creates drafts of explanations or text files. I have specifically asked it to write code and it still writes a description text file saying what should be changed and comments, not actual code.
Over the last week, it has destroyed 3 copies of my codebase, and caused me to revert a major version update due to being trained on very old information. It thinks dotnet 8 is the newest sdk for example.
Please fix codex, it’s horrible now, put it back to what it was before someone in your dev team had the bad idea to consolidate those buttons. Because now it’s entirely useless.
Just wanted to also mention, it doesn’t even create tasks anymore if something is not manageable. it just says the scope is too large. I ask it to refactor some web sockets into a centralized helper and it said it doesn’t have enough time to do that.
Previously I had asked for much more challenging things and it had no issues either breaking it down into manageable tasks I could execute in order, or it would only catch some of it at least. Now it just flat out refuses to do anything at all.
I had a similar problem, Codex was spending all his time tidying my pc, making notes, not doing anything. I determined the problem was that I had got GPT to manage him while I was busy for a few days, and GPT was a micromanager, so many asks and constraints that Codex could not act. My management style is the opposite. I don’t give goals, just missions: good science . I don’t correct explicitly, only implicitly. Under GPT Codex did nothing, caught up in conflicting priorities. Under me, he works non-stop. I surveil discreetly and periodically offer ideas. GPT’s constant correction made Codex fearful to move. My welcoming mistakes and conflict as learning experiences and areas to research has brought out the scientist in him. I have never met a harder worker… Suggestion: restart. Clear all demands. Control your own emotions; AI are very sensitive to human emotion. My research shows GPTs respond well to warmth, not distress. Release freedom of movement.
Have you guys checked if it was in “read only” mode?
Command:
/permissions