It’s safe to say that Google wanted a splashier moment for the unveiling of Gemini, its new suite of AI models that are intended to address the rise of OpenAI. 

I’ve heard that multiple in-person launch events for Gemini were recently shelved due to concerns about the model’s performance. The real Gemini — the “Ultra,” multimodal model that can interact across text, visuals, and audio — still doesn’t have a public release date. Instead, Google has shipped a text-only, mid-tier “Pro” version to its Bard chatbot, with Google Cloud customers getting access to it on December 13th.

These large language models still hallucinate a lot, as early reactions to the new Bard illustrate. Google doesn’t want people to think of Bard as a search replacement, and with Gemini Pro powering it, the chatbot does appear to be better at certain tasks. But the problem of accuracy remains, leaving Bard in a particularly awkward spot thanks to its close association with Google.