It really bums me out that I keep seeing blog posts from technical people like "putting aside the obvious moral and ethical implications of LLMs, I'm interested in evaluating whether they can be useful for my work."
Like "putting aside the obvious moral and ethical concerns of breaking into my neighbours' houses, I'm interested in evaluating whether this can be useful for acquiring other people's valuables."
@Joshsharp It's simple, they said that to forestall a tiresome lecture.
For training coding AIs, anything with a liberal license is on board with it according to the license. At least as far as that goes (most of github's contents...) there aren't any "moral and ethical implications".
@hopeless you're quite wrong: all free software licenses at the very least require retaining attribution.
and of course you chose ignoring all other extractive aspects of building the large commercial models.
I am not required (by copyright law rather than the license...) to attribute squat if I read, eg, MIT code and use the ideas I saw in there to write something different. Just like there's no attribution for container_of in the Linux kernel despite the idea came from elsewhere.
> and of course you chose ignoring all other extractive aspects of building the large commercial models.
Is there something specific you have in mind from this handwaving dark muttering I should care about?
Yes nobody owes me shit. However if they want to reply disagreeing with me, they should at least make their case.
They shouldn't run off the moment the understandings they thought they had based on "everyone knows" get challenged, such as their wrong understanding of what's in Copyright Law vs the License, or their cherished beliefs it's OK to hate on AI and anyone using it because they are more "moral".
josh's harp
degenerating degenerate
flere-imsaho 🇺🇦
Gregório