4 ms·
Yes I get that using the models costs a lot more than training. But using them isn’t free. Even using a large nVidia chip for a fraction of a second is expensiv
by alkonaut 2y ago
Yes I get that using the models costs a lot more than training. But using them isn’t free. Even using a large nVidia chip for a fraction of a second is expensive. So where does this money come from? Gains from paid tiers, investor money?
- drillsteps5 2y agoYou got it backwards. Training is extremely costly in many different ways, while running inference is much less expensive. Meta, Google, Alibaba, and some others train their models and put them into the open for free burning through their cash reserves trying hard not to lose it all to OpenAI (and MSFT).
- alkonaut 2y agoYes I phrased it backwards in the first sentence (of course). So the answer is ”investor money to operate at a loss?”