Yelp Inc.

Sacks oppose ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E237 · 2025-08-01 ▶ 40:12

you just can't have a situation where China can train on the entire internet, and our AI models are hamstrung by needing to contract, negotiate contracts with every single website.

Friedberg oppose ×3 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E237 · 2025-08-01 ▶ 35:49

I strongly disagree with the idea that AI getting trained is the same as AI replicating copyright material.

Chamath oppose ×3 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E237 · 2025-08-01 ▶ 42:13

I read that and I thought, this is the peak of these deals. These deals will only go down in terms of dollar value from here. And it actually brought me to this point where I was thinking to myself, is it even realistic to believe that patents and copyrights actually exist in five years?

Jason support ×3 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E237 · 2025-08-01 ▶ 43:43

I will guarantee that this will be the beginning of the deals and the deals will go up from here.

Chamath support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E233 · 2025-06-28 ▶ 1:44:56

I told you about this. We did a licensing deal with OpenAI, we meaning Beast. They paid us, and they're using it for pre-training. It's quite reasonable.

Jason support ×3 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E233 · 2025-06-28 ▶ 1:39:20

And then Chachipiti and the New York Times should do a deal. And I predict they will do this, or with another language model that's smart enough to do it

Sacks oppose ×3 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E233 · 2025-06-28 ▶ 1:35:32

if all they're doing is transforming the work, they're doing positional encoding, and then coming up with their own unique work product, this judge said that that is not a violation of copyright. I completely agree with that.

Friedberg oppose ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E215 · 2025-02-15 ▶ 1:29:08

I philosophically disagree with all of your nonsense on this topic. All three of you are wrong. And I'll tell you why. Number one, if information is out in the open internet, I believe it's accessible and it's viewable. And I view an LLM or a web crawler as basically being a human that's reading and can store information in its brain.

Chamath support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E215 · 2025-02-15 ▶ 1:27:20

This is lossy compression. It is absolutely... I'm now on your page. I hate to say this, Jason. I agree with you. You were right.

I'm going to go out even thinner limb and say I largely agree with you. I think it's a bit rich to crawl the open web, hoover up all the data, offer direct substitution for a lot of use cases, because now you start and end with the AI model. It's not even like you link out like Google did. And then you just close off the models for safety reasons. I think if you trained on the open web, your model should be open source.

Jason support ×3 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E215 · 2025-02-15 ▶ 1:24:40

I have a prediction here. I think this is all going to wind up like the Napster Spotify case. ... I think that there is a non-zero chance, like it might be 5 or 10 percent, that OpenAI is going to lose the New York Times lawsuit and they're going to lose it hard and there could be injunctions. And I think the settlement might be that these language models, especially the closed ones, are going to have to pay some percentage in a negotiated settlement of their revenue, half, two-thirds, to the content holders. And this could make the content industry have a massive, massive uplift and a massive resurgence.

Jason support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E213 · 2025-01-31 ▶ 1:10:12

I think it's for three years is the license and they just did this blanket license for every book. They didn't look at yourself. They didn't look at how desirable it was. It was just like a blanket deal. Everybody gets $2,500 bucks per book for three years. And I think I'm going to just do it just to support proper licensing so that people can start going down this path.

Jason support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E194 · 2024-08-30 ▶ 25:56

It's pretty crystal clear that their content's behind a paywall. And that's how they make money.

we can train these models, like training is like reading. And like reading things is, like when something's available to be read, and you've engaged in the right economic thing for reading it, I think that's a kind of a reasonable, fair use thing.

Friedberg oppose ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E180 · 2024-05-24 ▶ 8:03

I think what we will see over time is that rather than have the ability to sue for their likeness, a lot of AI that is like a celebrity, the value of it will arise from the celebrity's endorsement, not actually using the celebrity's features.

Jason support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E180 · 2024-05-24 ▶ 21:34

What happens in these content cases is they get settled almost every single time. So, the case law doesn't get codified, they just get settled out of court. If you go look at all the Fair Use cases, they almost never go to the mat. And so, this one will just be settled. It'll just be a question at what price.

Chamath support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E180 · 2024-05-24 ▶ 13:02

Using a computer to probabilistically copy something is still copying something.

Sacks support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E180 · 2024-05-24 ▶ 20:16

And by the way, I mean, because it is such an industry-leading company, I think we could end up with some very bad fair use precedents or laws because Scarlett Johansson is so sympathetic as a plaintiff compared to OpenAI. And unless they show us some discovery that proves that they really did hire the voice actor and all the rest of it, I mean, this could lead to some very bad precedents for the industry around fair use.

Sacks support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E179 · 2024-05-17 ▶ 1:30:53

Well, look, this is the same conversation we've had two or three times where we're gonna need the courts to figure out what fair use is. And depending on what they come up with, it may be the case that Google has to cut them in by doing licensing deals.

Jason support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E179 · 2024-05-17 ▶ 1:22:21

And I think that we're going to see a major lawsuit here. Those people who are in those boxes are going to look at the answer here and realize maybe they don't get the click through and that this answer was built on that. And now we're going to have to have a new framework. There's going to need to be, Sacks, a new company that clears this content so that Google can do answers like this.

Jason support ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E178 · 2024-05-10 ▶ 41:25

The music industry is going to consider this opportunity to make Taylor Swift songs their opportunity. It's part of the four part fair use test is, you know, these who gets to capitalize on new innovations for existing art.

Friedberg oppose ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E178 · 2024-05-10 ▶ 36:56

What's the difference here and why are you making the case that perhaps artists should be uniquely paid? This is not a sampling situation. The AI is not outputting and it's not storing in the model the actual original song. It's learning structure.

So I think there's an opt in, opt out in that case, first of all, and then there's an economic model.

Chamath oppose ×2 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E174 · 2024-04-12 ▶ 56:18

The difference is if Google doesn't sue OpenAI, it means copyright is worthless

Friedberg oppose ×3 ▲ on 🏛️ AI's training data gets a copyright bill - citations, links and licences E174 · 2024-04-12 ▶ 50:37

the models don't actually hold the data that they're trained on. They develop a bunch of synthesized predictors... I'd rather focus on is there copyright infringement in the output? Does the output mimic an original copyrighted source? And if it does, then that's where the infringement should lie, not on the how did you do it