It seems that people wanting to focus on AI-for-science instead of topping LLM benchmarks don't have a place anymore in DeepMind.