Anyone spoken to people at Deepmind recently, who are working on Gemini?

Combining multimodal LLM with a TreeSearch layer?