Why does GPT-5.2 (xhigh) show ~78% AA-Omni hallucination?

https://beckettsbrilliantdigests.theglensecret.com/conversational-hallucination-rate-vs-q-amp-a-benchmark-differences-in-ai-models

I’ve spent 12 years looking at models that claim to "solve" hallucination. Every year, a new architecture drops, the marketing teams descend, and the metrics go wild. The latest chatter around the gpt-5