Given Elon's track record of optimizing for cost and efficiency, I'm sure his conversation with his xAI engineers will be very colorful given Deepseek's achievement with their R1 model. At the end of the day, it seems obvious that having both massive compute and algorithmic efficiency will be a winning strategy but it's kind of weird to see Elon throw more money at the infra problem before trying harder to optimizing the process. Granted, he has said multiple times that massive efficiency improvements should be possible given that the human brain operates of far fewer kW but I haven't heard much happening on that front from xAI.
You are viewing a single comment's thread from: