A discussion I had with Anthropic Claude verbatim. (Bloomberg: circular support amongst small group of individual players and the inherent financial risk if current frontier models for AI cannot provide the expected value) Prompt: Re the financial risk referred to in this Bloomberg piece, is there any evidence of work on a better inference and reasoning model that doesn’t drive exponential requirements in data centre capacity to support the model. It seems the current model requires is highly inefficient requiring infinite capacity to support. # There’s substantial work underway — but with an important caveat that’s directly relevant to the … Continue reading LLM inefficiency and what could be the better model than next-token prediction