XO_THCa

there is a part of me which wants to side with the US-based AI companies but since most don't really care about people that much and only care about capturing as much GDP as they possibly can i have to ask: how is model distillation any different than these companies all training on copyrighted data (which we know they do)? also, if these chinese firms can trivially make a bunch of calls to these models and create a distilled version which is good enough for 90% of purposes, then why aren't the US-based AI companies just doing it themselves and saving everyone a lot in token costs? perhaps they are already with the flash models, but maybe they're not doing it as much as they could be and want to charge as much for tokens for as long as they think the market will bear it also, why don't any of them release some useful open source models on their own (besides Meta). why is the world relying on Chinese companies to do this? information and knowledge will be free we're not going to be serfs to 3 companies who maintain the models just because they want us to be

tfw you want to deploy your app and it runs great locally and builds but the final build in ci/cd is broken and every troubleshooting attempt takes 30 minutes