3 ms·
> This includes making Gemini 1.5 more efficient to train and serve, with a new Mixture-of-Experts (MoE) architecture. Wait. They are abandoning PaLM 2, which
by nycdatasci 3y ago
> This includes making Gemini 1.5 more efficient to train and serve, with a new Mixture-of-Experts (MoE) architecture.
Wait. They are abandoning PaLM 2, which was just announced 9 months ago?