Samsung Announces Its Big Foldable Unpacked - DTNSB 5046  ## Metadata - Author: [[Daily Tech News Show]] - Full Title: Samsung Announces Its Big Foldable Unpacked - DTNSB 5046 - Category: #podcasts - URL: https://share.snipd.com/episode/d76899db-0031-4e72-bca5-ed152d090205 ## Document Notes ## Highlights - **Legal Fair Use of AI Training Data** - A judge ruled that using legally purchased books to train AI models qualifies as fair use under U.S. law. - This is a first legal precedent likening AI training to humans learning by reading. Transcript: Tom Merritt And saying that was an infringement of copyright. Not a fair use. The judge, William Alsup, ruled that Anthropic's use of legally purchased books, which it also did, not everything was pirated, legally purchased books used to train its models qualifies As fair use under U.S. Law. This is the first time a judge has ruled that. Now, obviously, there are going to be appeals and all kinds of things, but this is significant. Alsup likened it to humans learning to write by reading, calling it exceedingly transformative. However, the company still faces a separate ruling over its use of millions of pirated books for training its models, books that it claims it just went out on the web and got stuff and Transformed it. And the copyright owners of the books are saying, you illegally accessed those books. This ruling also doesn't have any determination about the output of models and whether that infringes on copyrights, which is how a lot of these cases are now starting to break down Where judges are saying there might be a fair use on the input and the training, but if it can output a copyrighted work, maybe that's a violation. So in the absence of legislation, we're starting to see how these arguments are shaping up a little more. Jason Howell Yeah, and I'm sure we're going to see some real kind of similarities in how these things line up time after time. But qualifying the training as fair use and comparing it or likening it to learning to write by reading. I'm totally on board with that. I absolutely agree with that. I know a lot of people, you know, would disagree with ([Time 0:17:13](https://share.snipd.com/snip/2904cf1d-6e02-400e-8ec7-1f4249d1efaf)) - **Training vs Output Copyright Issues** - How data is accessed affects legality; downloading pirated books might break copyright. - AI models' output could infringe copyrights even if training is fair use, complicating the legal landscape. Transcript: Jason Howell Books probably constitutes breaking the law, then that's a real big mark on that process. And Anthropic's not alone in doing this. So if this ends up going negative for Anthropic in that regard, it has the potential to impact a lot of the other companies that have trained their data sets on exactly the same thing. Tom Merritt Yeah. I think it is interesting. First of all, yes, I agree with you that broadly speaking, it is like reading. I think a better argument and an argument I'd like to see a judge make is this isn't just like reading. This is like taking data and breaking it down into its letters and figuring out a word count. That's an even better analogy to me. It's like, you know, saying how many times the word for happened in a copyrighted work is not a breaking of copyright. And that's essentially what these models do. They create relationships amongst the tokens that they break them down into, you know. So this isn't a bad analogy, and I think it ends up in the right place, which is doing that to a work is very transformative, exceedingly transformative, as the judge wrote. So I'm glad that he came to that conclusion. It is interesting to rule how you got access to the work in the first place could be trouble. I hadn't really considered that. If I just go on the web and I download something, am I breaking copyright? Or is it when I do something with it that I'm breaking copyright? How is it no longer exceedingly transformative if I got it from the web and I just didn't check the provenance? There's some tricky implications to ruling that somebody who just went on the web and downloaded something is now no longer making fair use of something. So I'm curious how that gets worded. And I think it's great that they're focusing on the output. A world where models have to be careful that they don't output copyrighted works does sound exceedingly complicated, but on principle, I don't have an objection with it. Yeah. Jason Howell Yeah. Agreed. ([Time 0:19:05](https://share.snipd.com/snip/23c22ac6-6349-49fe-b5cd-2480b3b645a1)) - **AI Training as Data Analysis** - AI data processing likened to analyzing word counts, not just reading. - This token breakdown nature makes the training process highly transformative and less copyright-infringing. Transcript: Tom Merritt Breaking of copyright. And that's essentially what these models do. They create relationships amongst the tokens that they break them down into, you know. So this isn't a bad analogy, and I think it ends up in the right place, which is doing that to a work is very transformative, exceedingly transformative, as the judge wrote. So I'm glad that he came to that conclusion. It is interesting to rule how you got access to the work in the first place could be trouble. I hadn't really considered that. If I just go on the web and I download something, am I breaking copyright? Or is it when I do something with it that I'm breaking copyright? How is it no longer exceedingly transformative if I got it from the web and I just didn't check the provenance? There's ([Time 0:19:54](https://share.snipd.com/snip/c2044be3-c15a-48cd-b77f-809b071d91c8)) - **AI Empowers Individual Creators** - CEO of Runway states AI empowers individual creators more than replacing studios. - Runway aims to be the 'Adobe of AI,' providing tools, not producing content like studios. Transcript: Andy Beach Yeah. So Chris Valenzuela is the CEO of Runway, and he was recently talking to Nilay Patel at The Verge on their Decoder podcast around these very topics. And you're absolutely right. I think everybody that's involved in media technology right now is acknowledging this. And so he really made the argument with Nilay that AI was going to empower individual creators far sooner than it was going to replace the studio world. And this is a person who makes the tools, right? Tom Merritt Explain what Runway does. Andy Beach Go make synthetic content out there at scale. Tom Merritt Yeah, this is less risky than someone who's involved in union negotiations, perhaps, coming out and saying it. It's also self-interested. The person saying this is the person that's going to benefit from people believing that that is true. That doesn't mean that he's lying, but why should I believe the toolmaker when he says, gosh, you're really going to need my tools? Andy Beach Well, I think what he's saying, though, is that this doesn't just go out and replace the Hollywood system right now. But what it does is allow individuals to go take tools and make Hollywood like productions that they put out themselves on a YouTube or any other independent platform that they may choose To to publish to. So I think what he was really circling around is that and he said this in the talk multiple times, he likened Runway to basically being the Adobe of the AI era, not the Pixar. So they're not the studio, they're the tool that someone's going to go use to make content. Tom Merritt There is a lot of backlash against AI right now. And I have heard from many people that there ([Time 0:33:44](https://share.snipd.com/snip/d8803a7b-ce5f-4491-9485-acece7cf39e4)) - **Protect Your Rights Using AI** - Read terms of service thoroughly before using AI creation tools to understand content rights. - Opt out of training data inclusion if possible to protect your created works. Transcript: Andy Beach It's a great question, and I don't think that's 100% clear today. I think the best thing anybody could do if they were a creator and any model that they're looking at, this is a place where you don't want to skip past the terms of service. You want to be reading those EULAs in some detail and understanding what is the trade-off? What am I giving away by using this product and paying for it? Am I giving them access to my content to use for training? Am I going to have things that show up in their models that look like my work at a later date? I'm not saying that that is what it's doing. I'm just saying you want to make sure that you educate and arm yourself for the information. And often with these model makers, there are ways to protect yourself. In other words, there are ways to stop it from training against your content. But unfortunately, today, a lot of it is still opt-out versus opt-in, meaning that by default, you are part of the training set unless you go in and explicitly turn it off in some settings That are buried down in the privacy area. So you just want to make sure you understand the tool that you're using and the terms and conditions that, that are being applied to it. Tom Merritt And then on the other side, making sure you're making something ([Time 0:37:50](https://share.snipd.com/snip/cf1e02dd-7bb7-473a-8179-581ea69a2099))