I suspect a large chunk of people in denial have either never used it and are basing their judgment on memes and cherry-picked failure modes that go viral, or have only used the earlier/weaker models that are far more prone to hallucination. I use ChatGPT Plus regularly, and find it occasionally imperfect but still very useful, but a few times I've tried using it while logged out and the free model it defaults to was far more flawed. Probably why OpenAI built GPT5 around a marginally better but far more efficient model that they can roll out to everyone (including the large cohort of free users).