[coolThingsFirst]

The problem with LLMs is that even if you learn them you can't train them. Not enough power in consumer GPUs

[aborsy]

Aren’t there smaller ones, like 7B llama2, that could be trained in a powerful computer?

[a2code]

Isnt training as a service a possible temporary solution here?

[coolThingsFirst]

It's really expensive, I'm talking about new people learning and experimenting with LLMs