[goodmythical]

Published today, leading with "Recently, I ran an AI model locally and reached a conclusion that is hard to ignore: the major LLM providers are probably operating with substantial subsidies to sustain their services at current prices." is wild.

How do you get all the way to running a local model without having heard that this is the case? Like...it's so well documented...I wonder if author knows that a portion of the subsidy is that the models are being trained on free, cheap, and chat models...

[lrsaturnino]

Thanks - yes, I agree, and already I knew this was a thing. I just haven't yet concretely made the actual math, and it left me baffled - Anthropic and OpenAI are charging 1/3 of the COST tops.

[goodmythical]

Ah, yes, it is always interesting to see loss leaders and just general revenue wizardry that can be accomplished at scale. Things like "It can be more cost effective to take on debt than to earn additional revenue". Imagine how much it costs Alphabet to run hundreds of thousands of terabytes of not only streaming but transcoding every single day for youtube and you get a similar sense of "holy shit where is the money coming from".