30 Oct 2025 18:13:58 noodle@pastanoggin.com:
> i think they meant LLM model training which uses way more power than
> inference. and most people just use pre-trained stuff like
> chatgpt/etc which are huge models so i bet their training power cost
> is very high compared to a small (trained with non-enterprise
> hardware) locally trained model
MIT have stated that each query to e.g. chatgpt is like using the microwave for 8 seconds (not sure what wattage microwave).
This url is one way to avoid AI overviews and which I assume use much less energy but I dislike personally in any case.
https://www.google.com/search?q=%20&udm=14
No comments:
Post a Comment