January has been remarkable for the number of important ads in AI. For me, two stand out: the support of the United States government to the Stargate project, a giant data center that costs $ 500 billion, with investments from Oracle, Softbank and OpenAi; And the launch of Deepseek of his R1 Reasoning ModelTrained at an estimated cost of approximately $ 5 million, a large number but a fraction of what Openai costs to train its O1 models.
The American culture has assumed for a long time that it is better and that more expensive is better. Certainly, that is part of what is behind the most expensive data center ever conceived. But we have to ask a very different question. If Deepseek was trained for approximately a tenth of what it costs to train O1, and if the inference (generating answers) in Depseek costs approximately a thg -of -way what it costs in O1 ($ 2.19 per million departure tokens versus $ 60 per million production tokens) Does the American technology sector direct in the right direction?
Learn faster. Dig deeper. See further.
It is not clearly. Our biggest mentality is better “is failing us.
For a long time I have believed that the key to the success of AI would be to minimize the cost of training and inference. I don’t think there really is a career among the communities of the United States and China. But if we accept that metaphor, the United States, and Openai in particular, are clearly behind. And a mid -Trillon Data Center of dollars is part of the problem, not the solution. The best engineering exceeds “Supersize it”. Technologists in the United States need to learn that lesson.
#eldest #win #OReilly
