2 ms·
> same applies for AI which is very energy intensive too Training (building the model/AI) might be, inference (using the model to actually "run" the AI) is rel
by tgsovlerkhgsel 5y ago
> same applies for AI which is very energy intensive too
Training (building the model/AI) might be, inference (using the model to actually "run" the AI) is relatively cheap, I think. This may change with massive models, but I suspect power usage of the AI will not be a major concern for deciding where to put AI-operated factories.
- trhway 5y agotraining once vs. millions of inferences. And tomorrow inference may naturally include incremental training. Human brain consumes 20% of the body energy. I don't think AI will come to human brain energy efficiency soon, and thus i think AI (and its accessories, ie. like 3d/AR/VR model spaces where and what AI will be operating with) will be a major, if not main, energy consumer going forward. When Google self-driving car passes by the main sound coming from it isn't the engine nor the tires :)