Bold prediction. #Datacenters are for training. #AI inference will be distributed, highly local and open weights. Actual...

Bold prediction. #Datacenters are for training. #AI inference will be distributed, highly local and open weights. Actually, not that bold. MoE, weight-space programming and decision models mean that while we do need frontier models, we do not need to inference on them nearly as much as #OpenAI or #Anthropic claim.You do not need a subscription to #LLMs or AI.My employer the University of California is about to needlessly spend millions on subscriptions, in a deal with OpenAI. No confidence!