The models are braindead

What version of the Codex App are you using (From “About Codex” dialog)?

26.519

What subscription do you have?

Pro

What platform is your computer?

Darwin 24.3.0 arm64 arm

What issue are you seeing?

Sorry to put this bluntly, but the models are retarded today and this issue is me (once again) being sick of it. No harness is able to fix this.

5.5 just tried to have me build a porch with no door. On another project it forgot in the next message that we were using specific servers, despite my correcting it the literal message prior.

me: i don't know how i can do better than what i did

Codex: You couldn’t have done much better. You said the directory was wrong earlier, you told me the service was on the Ubuntu box, and you pushed back when I gave you a local bind address. The failure was on my side: I didn’t reset my assumptions when you corrected me.

What steps can reproduce the bug?

Probably when you guys use non-nvidia hardware. We would know more if you had a modicum of transparency.

What is the expected behavior?

Messages are read and responded to in the order they were sent. GPT lately has been responding to messages sent earlier in the chat. This degradation seems related.

Additional information

I'm really sick of having to find out if the models are having a bad day by using the models to see if they generate cortisol.