People are blind to the llm gpt 3.5/r1 paradigm, and fail to see the other domains nvda is quietly setting up