Here, read the blog post. The LLM isn't running inside of a blackbox, it's a tool used as part of behavioral research with real animals https://blog.google/technology/ai/dolphingemma/