I believe this discussion is important, just the other day I was sad about people colluding inference and training like if both were having the same carbon footprint. I am not defending Sam Altman or anyone else. But we need to look at the facts for what they're. Running a quantized local model, accepting latency and all, can make your use of LLMs as low as playing video games. I already said that in the past, and I only got downvoted. Maybe because everyone is hurt bout this technology, which is having detrimental consequences. I am not saying there aren't any problems... but I am accepting the world for what it is, as GPT4 has said ~3 years ago: confetti has left the cannon.

So, we need to deal with this reality. The training done on data like open source code, or even copyrighted materials, will always be a debt.

Just like colonialism. And yet, we are not hating each other for "simple" stuff like eating chocolate or drinking coffee which is most often than not unethically sourced. OR the minerals that make our electronic devices..

Again, it's not that I am vouching to ignore the issues. But I rather have us take the conversation as a debate or dilemma. Not just start attacking one another.

This all seems like that caricature of a vegan that hates the vegetarians for eating some cheese.

Look at the common ground. Here, on lemmy, more often than anywhere else, we have a big chunk of values and ethics in common.