Evan Zimmerman on X: "On 20VC this week, @rodriscoll couldn't stop himself from noting the delicious irony of Anthropic bemoaning distillation due to its own fair use disputes. You can have whatever opinion you want about whether you think Anthropic did something wrong, but saying that there was "IP" / X<br>Post

Log inSign up

Post

Evan Zimmerman

@ejzim

On 20VC this week, @rodriscoll couldn't stop himself from noting the delicious irony of Anthropic bemoaning distillation due to its own fair use disputes. You can have whatever opinion you want about whether you think Anthropic did something wrong, but saying that there was "IP theft" like Rory did repeatedly is just wrong on the law, and this has been wrong since the first lawsuit by Sarah Silverman against OpenAI. I called it that very week back in July 2023.

For context: I have a JD from Berkeley Law with a certificate in technology law and have worked with IP for over a decade. Plus my startup focuses on IP lawyers.

LLMs are trained on large corpuses of text, including books, articles, videos, and other content created by third parties. Many of those creators felt that this training constituted mass theft and have sued all of the major LLM providers since the dawn of ChatGPT. There are dozens of these lawsuits; you can track them via Hogan Lovell LLP or Chicago-Kent Law School.

These creators, who all know what fair use is, are all trying to argue that LLM providers engaged in copyright infringement. And yes, to many people, it feels icky. I can't debate with that because everyone is entitled to their own moral intuitions.

But that is not what the law is. The law means something, even if you don't like it, and LLM training is clearly fair use. The reason is its "transformative" nature, ie, the fact that the new creative work (the large language model) is very different than the input. This is true even despite the other three fair use factors.

You can see this playing out in the actual courts. In case after case, the substantive claims are summarily dismissed by judges, including in the original Sarah Silverman case. When this has actually gone on past the pleading stage, in two cases (Meta and Anthropic), fair use came down on the side of AI. As it should; you might not like what the law is, and you might want to change it, but it is what it is. The only example of an AI case that didn't find fair use is ROSS vs. Thompson Reuters, where ROSS used a thin LLM veneer to steal WestLaw case headers, and even that loss is currently pending on appeal.

This is why the distillation case is different. The Chinese distillers are trying to create substantially the same product. There are other fair use cases that point in this direction, such as Ninth Circuit cases on reverse engineering (Sega and Sony) and the cases on collage (like Koons and Cariou). The fact that they are violating the EULA and acting in bad faith are further factors that make this different just on the law, let alone on moral grounds.

So go ahead and laugh at the schadenfreude, but don't mislead your listeners on fair use.<br>span:not(:empty)~span:not(:empty)]:before:content-['·'] [&>span:not(:empty)~span:not(:empty)]:before:px-1 [&>span:not(:empty)~span:not(:empty)]:before:shrink-0 min-w-0 overflow-hidden">Harry Stebbings

@HarryStebbings

12h

Dario just declared war on open-source.

Anthropic's message is clear: open source could destroy the entire AI business model, and Chinese open-source models are the cause.

I sat down with @jasonlk & @rodriscoll to discuss it, along with the biggest news in tech this week:

- Show more

span:not(:empty)~span:not(:empty)]:before:content-['·'] [&>span:not(:empty)~span:not(:empty)]:before:px-1 [&>span:not(:empty)~span:not(:empty)]:before:shrink-0">2:43 PM · Jul 2, 20261.9KViews

:host{display:inline-block;direction:ltr;white-space:nowrap;line-height:1}span{display:inline-block}:host([data-will-change]) span{will-change:transform}.number,.digit{padding:round(nearest, calc(var(--number-flow-mask-height, 0.25em) / 2), 1px) 0}.symbol{white-space:pre}2:where(number-flow-react){line-height:1}number-flow-react > span{font-kerning:none;display:inline-block;padding:calc(round(nearest, calc(var(--number-flow-mask-height, 0.25em) / 2), 1px) * 2) 0}2<br>:host{display:inline-block;direction:ltr;white-space:nowrap;line-height:1}span{display:inline-block}:host([data-will-change]) span{will-change:transform}.number,.digit{padding:round(nearest, calc(var(--number-flow-mask-height, 0.25em) / 2), 1px) 0}.symbol{white-space:pre}1:where(number-flow-react){line-height:1}number-flow-react > span{font-kerning:none;display:inline-block;padding:calc(round(nearest, calc(var(--number-flow-mask-height, 0.25em) / 2), 1px) * 2) 0}1<br>:host{display:inline-block;direction:ltr;white-space:nowrap;line-height:1}span{display:inline-block}:host([data-will-change]) span{will-change:transform}.number,.digit{padding:round(nearest, calc(var(--number-flow-mask-height, 0.25em) / 2), 1px)...