For a long time, you could fool models with questions like "Which is greater, 4.10 or 4.5?" Maybe they're still struggling with that at OpenAI.

At this point, I'm just assuming most AI models — not just OpenAI's — name themselves. And that they write their own press releases.