Copilot was trained on copylefted code while itself being closed. What was brought to attention by @ralC@lemmy.fmhy.ml isn’t efficacy, but Microsoft’s lack of ethics and social responsibility when it comes to their bottom line.

I honestly don’t have a problem with that. Everything that it was trained on is publicly-available/open-source code, and I’m not aware of any license that requires you to distribute your modifications if you don’t make modified binaries publicly available, not even GPL. And even then, you’re only required to make available the code that was modified, not related code. And I don’t even think that situation would apply in this case, since nothing was modified, it was just ingested as training data. Copilot read a book, it didn’t steal a book from the library and sell it with its name pasted over the original author’s.

This isn’t really any different of a situation than a closed-source Android app using openssl or libcurl or whatever. Just because those open-source libraries were employed in the making of the app doesn’t mean that the developer must release the source for that app, and it doesn’t make them a bad person for trying to make money from selling that app. Even Stallman is on board with selling software.

And even if you take all that off the table, you’re free to do the exact same thing and make a competitor. Microsoft didn’t make their own language model, they’re using a commercially-available model developed by OpenAI. There’s literally nothing stopping anyone else from doing this as well and making a competing service called “Programming Pal” and making their code open-source. In fact, it’s already been done with FauxPilot and CodeGeex and the like.

So yeah, I really don’t have a problem with it. This ended up a lot longer than I had originally thought it would, sorry for the novel.