@PyTorch this is good to see. I was never very confident about using TLX since I didn't know what I might be missing from the main Triton branch, or how stable it is. The tests and metrics are nice to see.
@deepanshusharmx@AnthropicAI If you replaced open weights with closed weights in the sentence you highlighted, it would still be true (other than the first part of the sentence which is false in both cases)
@ai_for_success the correct analogy here is if you could only use windows or GPU compute through the cloud through some sort of thin client. which kind of is the exact dystopia our trends are taking us to.
@ArielKwiat people are trying to take this to be some bold claim that american models distill from Chinese ones, but like is there a world where they don't? like it seems ludicrous to not distill the teeth out of your competitors. some companies get fussy about it, some don't.
@giffmana The problem is with datacenter build out happening deliberately in water scarce areas, reducing the water supply for residents there and polluting the water they do have available because of all the construction. Not to mention the light and noise pollution.
@mattjay Not sure why people think this isn't a big deal. Like your project being infected is bad, but that's only made a real problem by cursor executing a malicious binary. If it didn't do that, you could inspect your repo and remove the binary without incident.
@BrendanFalk It's possible but with open source models you lower your risk of this happening because millions of people are testing them and the inference provider can be you or whoever else you trust. This is much more likely and easily done by a black box lab.
@Lina_Hoshino Maybe people should create some standard licenses like we have in CS? Some variants mixing concepts of what modifications are allowed, what kinds of uses are allowed, which files will be sent, what happens if the artist quits etc. That way you COULD have detailed contracts.
@Lina_Hoshino You have been so outspoken and proactive about this whole situation 😭🙏 It's like the moderating voice of reason preventing all hell from breaking loose.
@vsbuffalo I had this with chatgpt too. It wanted to say intelligent but like it spelled the first 4 letters of the word in cyrillic script (I'm not sure if it was actual Russian, but it was 4 letters long and Google translate told me it was "intel")