Most of these figures are guesses along a spectrum of “educated” since many models, like ChatGPT, are effectively opaque to everyone and we have no idea what the current iteration architecture actually looks like. But MIT did do a very solid study not too long ago that looked at the energy cost for various queries for various architectures. Text queries for very large GPT models actually had a higher energy cost than image gen using a normal number of iterations for Stable Diffusion models actually, which is pretty crazy. Anyhow, you’re looking at per-query energy usage of like 15 seconds microwaving at full power to riding a bike a few blocks. When tallied over the immense number of queries being serviced, it does add up.

That all said, I think energy consumption is a silly thing to attack AI over. Modernize, modularize, and decentralize the grids and convert to non-GHG sources and it doesn’t matter–there are other concerns with AI that are far more pressing (like deskilling effects and inability to control mis- and disinformation).

There are many open sourced locally executable free generative models available.

You are agreeing with the post you responded to. This ruling is only about training a model on legally obtained training data. It does not say it is ok to pirate works–if you pirate a work, no matter what you do with the infringing copy you’ve made, you’ve committed copyright infringement. It does not talk about model outputs, which is a very nuanced issue and likely to fall along similar analyses as music copyright imo. It only talks about whether training a model is intrinsically an infringement of copyright. And it isn’t because anything else is insane and be functionally impossible to differentiate from learning a writing technique by reading a book you bought from an author. Even a model that has overfit training data, it is in no way recognizable to any particular training datum. It’s hyperdimensioned matrix of numbers defining relationships between features and relationships between relationships.