I’m guilty too and I let (mostly local or hydro/solar-powered) LLMs write utility scripts for me, but I’m fully aware of the illegal way they’ve been put together and I wouldn’t use them to make a product I sell, that would feel unethical.

I hope one day someone makes a model that is trained on e.g. only GPL, MIT or BSD-licensed code that gives full attribution to the generated output, and one that uses only Creative Commons art that is cleared for remixing, for example. It’s not that these things aren’t available, just that the commercial incentive (those billionaires, eh) is stronger to simply steal everything and ignore any licensing or laws.

All this can be done ethically and with small environmental impact, the fact that they choose not to do it in order to chase power and profit is the most offensive part.