（预览）Apple 寻找 AI 合作伙伴、第二起合理使用裁决与直面现实、《F1》电影及相关事项

Hello, and welcome to a free preview of Sharp Tech. All right, elsewhere this week, on Stratechery, you wrote about a follow-up opinion—a separate, parallel opinion—regarding AI training and the fair-use doctrine. I'll read from Bloomberg Law:

“Meta escaped a first-of-its-kind copyright lawsuit from a group of authors who alleged the tech giant hoovered up millions of copyrighted books without permission to train its generative AI model called Llama. San Francisco federal judge Vince Chhabria ruled Wednesday that Meta's decision to use the books for training is protected under copyright's fair-use defense, but he cautioned that his opinion is more a reflection on the authors' failure to litigate the case effectively.

“This ruling does not stand for the proposition that Meta's use of copyrighted materials to train its language models is lawful,” Chhabria said. “It stands for only the proposition that these plaintiffs made the wrong arguments and failed to develop a record in support of the right one.”

So, Ben—

No, this rundown says you're going to ask me whether I have some takes. I've already written about this. I podcasted about it. I'm dithering. You go first. What's your take?

Well, I didn't love the podcast we did on this issue right before summer break.

Oh.

Well, no, I wasn't just agonizing the entire break, but maybe for an hour or two afterward. It just didn't sit right. I didn't have that much time to prep, and I hadn't fully formulated exactly what I think about this issue and exactly why I ultimately lean toward treating training as fair use.

A week and a half of baking. What's coming out of the oven here?

Oh, boy. Well, I will say I loved Judge Chhabria's opinion. I just thought his explanation of all of it was a much more intellectually honest approach to dealing with the realities in play here and the stakes for the future of creators. So do you want me to read from Judge Chhabria, or do you want me to run through my whole thought process on this?

You're in charge.

I actually wrote it down the other night.

I'm letting you cook. I'm going to restrain myself, and you tell me why Judge Chhabria got it right.

Am I in the catbird seat here?

You're in the catbird seat. Exactly.

1. AI Training Transfers Creator Wealth

All right. So let me run through the way I think about this issue. I'll wear my anti-monopolist hat here and say that I'm instinctively uncomfortable with treating LLM training as fair use. It does seem clear that LLMs will harm the long-term market for books, journalism, and other creative work, probably including human-generated music, movies, and TV shows.

Human-generated podcasts?

Well, maybe not podcasts. I don't know. How big is the market really, if we're being honest?

Look, it's not big enough to get you a private jet from Nantucket, so it's big enough to get to Nantucket—

I know.

—but you might get stuck there.

I'm waiting for the superintelligence jet to come through.

That's right.

God, call me, Zuck.

If we know that's the direction this is going, and that it's only possible for it to go in that direction if these models are allowed to train on the work of millions of creators who own their IP without compensation, then we're ultimately talking about a transfer of wealth and long-term profit incentives from one industry that thrives on the work of millions of creators to an industry dominated by $5 trillion companies.

So I'm instinctively uncomfortable on that basis. That transfer of market power from creators to model makers without any real compensation seems antithetical to the concept of fair use, even if it's not necessarily antithetical to the technical letter of fair-use law.

And so, as I was agreeing with you on the last podcast, I was thinking, “Eh, I don't know. I feel like I'm betraying my roots here.” Then, thinking it through further, in a copyright context, I think courts are really stretching the boundaries of what copyright law is supposed to protect and enforce if they're talking about LLM output that is not itself infringing, and then trying to prevent speculative market harms in the future rather than actual market harms in the present.

I don't know that the law can really go that direction on a consistent basis, and that's ultimately where Judge Chhabria netted out here, while also reckoning with the reality that this isn't a bunch of schoolchildren reading books, then ultimately learning to write books and competing in the same market years down the line.

The scale of this does render it a completely separate problem from anything that the law has tried to reckon with before. I appreciated him for stating that honestly and then also coming to the conclusion: “Well, there's no real record of harm for these authors in terms of their market opportunities, and I can't rule against Meta in this case.” Does that make sense?

I completely agree.

There you go.

2. Copyright Cannot Solve AI Harms

No, and a little bit of a stand. So I appreciate his intellectual honesty by the end because the first 3 quarters of that opinion were a little bit of living in a fantasy land.

Okay.

And this bit about their needing to make a better case is a fantasy. There's no case to be made. That's the issue.

That's the issue, yeah.

Right? And so I'm glad you mentioned the anti-monopolist roots, because this is actually a replay of our debates about aggregators and monopoly.

Yeah.

My whole point all along is I actually agree with you. I agree with all the concerns. But antitrust law cannot fix the issues you're worried about. It sounds like we fast-forwarded to the end here, which is that copyright law can't fix the issues you're worried about, either.

If you want to address these issues, it's going to require a new law and a new approach. Then we have a separate debate about the trade-offs with some sort of law that explicitly addresses this.

Mm-hmm.

But just like the monopoly thing, that's been my thing all along. I'm like, “I get the point of view, but you have to pass a new law.” You can't retrofit antitrust to this because it doesn't fit, and it's what you just said about copyright.

It's like, yeah, all these concerns are real and valid, but Judge Chhabria, to his credit, by the end, he's like, “Yeah, but it doesn't really apply here.” My critique of him is his suggestion that you could ever find evidence to apply.

What he needs evidence for is purchase decisions not made by individuals. How are you going to get that data? It's—

Right, not made by individuals, not only today—

At scale.

I mean, I think he's more broadly concerned with the long-term implications of this sort of behavior.

Right, which are very real. Those are totally real, right? But if the standard of the law is that you have to find explicit market harm, it's all in the noise, right? The harm's real.

Very fuzzy.

Yeah, exactly.

And I like that he said, “Meta makes the mistake the Supreme Court instructs parties and courts to avoid: robotically applying concepts from previous cases without stepping back to consider context. Fair use is meant to be a flexible doctrine that takes account of significant changes in technology.”

That's how I would like law to work: to be elastic in some respects and adaptive.

Oh, living Constitution guy.

Yeah.

This takes me back to my constitutional law debates in college, but yes.

Look, we're not going to turn this into a law school seminar.

I'm down with it. They're my favorite classes. I loved them.

I think his version of understanding the law, what it can do, and what it can adapt to makes more sense and resonates with me, as opposed to robotically applying past precedents to try to grapple with some of these technology challenges.

But you mentioned that if it were to go to Congress, you would still probably come down on the side of the AI companies here. So explain your logic.

3. Opt In Rules Reward Pirates

Well, this whole question runs in multiple directions. I think if you were to pass a law—which I think the UK either passed or is proposing—where, basically, by default, you have to get affirmative permission to include copyrighted works. I might be wrong on this, but you can imagine a law like that.

Good luck. How is that going to happen? Meanwhile, all those works that are available can be pirated, can be online, or whatever it might be. Unscrupulous actors will have access to them. It puts us in a situation where you're punishing the people trying to follow the law.

And so I don't like this, but it's one of those situations where you have to look at reality and have a realist—in foreign-policy terms—sort of view of the world.

Mm-hmm. Realpolitik.

Which is—

Let's do it.

If you pass this sort of law, the only companies that are going to suffer are the ones that are trying to follow the law, and you're actually helping and empowering the companies that don't care. That doesn't seem ideal to me.

Yeah. Well, that makes sense. For me, as I was thinking about it, I do think there's room to step back and say, “What sort of incentive structure do we want to create and protect for writers or musicians or film studios?”

But the reason I'm less passionate about pushing for statutory protection for that sort of class versus, say, pushing for laws that limit the take rate for Google or Meta or Amazon as they deal with businesses all over the world is that, in the latter case, we're talking about transfers of wealth from hundreds of different industries that are all beholden to tech platforms that essentially operate as utility companies.

In that case, I think the entire economy could be healthier and more dynamic if we're not limiting the upside in hundreds of different industries that have to pay their pound of flesh to Google or Facebook. But in AI, I think we're talking about erecting barriers to protect an industry of creators that is already facing so many structural threats.

Right, they're already doomed, so who cares?

Just because of the mere existence of technology.

That's right.

And so, as far as realpolitik is concerned—

No, the internet has already decimated all these folks. Exactly.

Yeah. And so, what are you actually trying to—

Yeah.

…protect by imposing—

The horse is halfway across the country at this point.

…huge burdens? Yeah. And so I think you just have to be real with it.

4. Copyright Is A Government Monopoly

Well, I think the other angle, too, in appealing to your anti-monopolist side, is that copyright is a government-granted monopoly, right?

Yeah.

That's the tension with that. People act like it's this God-given gift. It's like, no, this is an anti-competition thing that we do because we value the incentives of doing creative things. But all intellectual property law is government-granted copyright. It's by definition market-distorting.

I mean, you can get really abstract and say that that's also true of property law generally, but we don't have to go full socialist here.

Well, we have a Second Amendment discussion here. Step on my lawn, Andrew.

That's right. The law of the land out—

That's right.

…there in Wisconsin.

So, yeah, that's a factor for me. I think that a lot of copyright and intellectual property laws act like they're laws of nature, but they're not. There are very real challenges and issues that come from government-granted monopolies like patents and copyrights, which lead to abuses and high prices and can actually be counterproductive.

Given that, I'm hesitant to layer on all these potential extra problems in service of a government-granted monopoly for folks who, to your point, have made it this far and are probably going to be fine.

5. Human Broadcasting Still Matters

The other thing is that all the AI is individual. It's all custom to you. I continue to believe—and again, I'm totally talking my personal book here, so take it for what it's worth—that there are tens of thousands of people listening to this podcast. They can talk to other people. It's a common experience. It's something that has value by virtue of being one thing broadcast to many people, right?

Yeah.

And having the human component—and I think I mispronounced the name of someone earlier—we're doing our part to stay unique. To that point, the horse is gone.

6. Regulation Favors AI Incumbents

Right. Well, imposing huge burdens on AI startups around the world isn't going to change—

Oh, that's right.

…the structural forces—

Totally right.

…that are ravaging publishing.

All you're doing is favoring the incumbents even more—

Yeah.

…because they can jump through all the hoops and figure it out. It's like YouTube figured out a way to let you have copyrighted music in your videos. They just pay everyone. This whole convoluted system is great. I'm glad they got it to work. It sucked when people had birthday videos taken down because there was copyrighted music in the background.

There will never be another competitor to YouTube. All right, and that is the end of the free preview. If you'd like to hear more from Ben and I, there are links to subscribe in the show notes, or you can also go to sharptech.fm. Either option will get you access to a personalized feed that has all the shows we do every week, plus lots more great content from Stratechery and the Stratechery Plus bundle. Check it out, and if you've got feedback, please email us at email@sharptech.fm.