RE: AI isn't as smart as you think
You are viewing a single comment's thread:
I do find it weird that not a lot of people seem to be too upset with the energy usage of a lot of LLMs, maybe it's not super common knowledge?
The energy usage is tiny, an AI model goes through two phases.
- Training
- Usage
Training is extremely time consuming and costly, but once that's done you can run many models on a Raspberry Pi to solve actual real problems. In the end, the final result is just a bunch of weights (numbers) representing paths to take in the neural network for specific inputs.
The amount of energy AI can save in the long run is astronomical. Imagine solving cancer, developing new energy solutions, making tedious task far more efficient, taking over the world with robots.
0
0
0.000
Thanks for that differentiation. I didn't realize the training only happened in the initial phase, I thought LLMs were continuously being trained with the latest available data, I'll have to look into it more. Thanks.
They do get updated, but it isn't something that happens in real time like sending updates and creates to a database. They need to be retrained, then deploy the new model. Training these models requires millions of dollars and can take months. So you won't see frequent updates until technology improves.