OpenAI does this for literally _every_ release. They constantly say "Available to everyone" or "Rolling out today" or "Rolling out over the next few days". As a paying Plus member it irks me to no end, they almost never hit their self-imposed deadlines.

The linked page says

> GPT-5 is here > Our smartest, fastest, and most useful model yet, with thinking built in. Available to everyone.

Lies. I don't care if they are "rolling it out" still, that's not an excuse to lie on their website. It drives me nuts. It also means that by the time I finally get access I don't notice for a few days up to a week because I'm not going to check for it every days. You'd think their engineers would be able to write a simple notification system to alert users when they get access (even just in the web UI), but no. One day it isn't there, one day it is.

I'll get off my soapbox now but this always annoys me greatly.

Are you a paid user? I haven't seen a model selector in years.

Not true. I've been a paid user forever and on the Android app they have definitely obscured the model selector. It's readily visible to me on desktop / desktop browser. But on the Android app the only place I can find it is if I click on an existing response already sent by chatGPT and then it gives me the option to re-generate the message with a different model.

And while I'm griping about their Android app, it's also very annoying to me that they got rid of the ability to do multiple, subsequent speech-to-text recordings within a single drafted message. You have to one-shot anything you want to say, which would be fine if their STT didn't sometimes failed after you've talked for two minutes. Awful UX. Most annoying is that it wasn't like that originally. They changed it to this antagonistic one-shot approach a several months ago, but then quickly switched back. But then they did it again a month or so ago and have been sticking with it. I just use the Android app less now.

Actually this trick have been proven to be useless in a lot of cases.

LLMs don’t inherently know what they are because "they" are not themselves part of the training data.

However, maybe it’s working because the information is somewhere into their pre-prompt but if it wasn’t, it wouldn’t say « I don’t know » but rather hallucinate something.

So maybe that’s true but you cannot be sure.