This is exactly what I'm talking about. The immediate issue with GPT-5's launch was that some people were getting rerouted to 4o, and there was no control over which version of GPT-5 you're getting. Whatever he's talking to there looks nothing like a reasoning model.
This really hasn't been a thing since reasoning models showed up. Any recent example of such seems to come from non-reasoning variants.
>laughably sycophantic
Mainstream models are moving away from this, afaik. Part of the recent drama is that GPT-5 wasn't sycophantic enough for some users.