Seriously can’t agree more. Whatever small utility LLMs offer (I haven’t used them in my day-to-day work at all because a regular search works for me just fine, and I can create my own images), it’s not worth the incredible amount of resources used to perform a single query. Maybe if we had fusion energy and there were no environmental implications to further researching the theoretical limits of LLMs their use could be justified–but we don’t, there are, so it can’t.
I really think AI development should move to creating small and single-purpose models that can easily be trained with public domain data of your choice, so they can be ran locally and spend far less energy than models like Gippity.
Seriously can’t agree more. Whatever small utility LLMs offer (I haven’t used them in my day-to-day work at all because a regular search works for me just fine, and I can create my own images), it’s not worth the incredible amount of resources used to perform a single query. Maybe if we had fusion energy and there were no environmental implications to further researching the theoretical limits of LLMs their use could be justified–but we don’t, there are, so it can’t.
I really think AI development should move to creating small and single-purpose models that can easily be trained with public domain data of your choice, so they can be ran locally and spend far less energy than models like Gippity.