This Week in Startups
00:40:19 · Jason Calacanis, Lon Harris
[00:40:19] S0: What else do we got here on the docket here? [00:40:21] S2: We follow the the sort of copyright infringement AI story. There's a new wrinkle today. Adweek reports that Anthropic has struck its first content licensing deal with the Associated Press specifically for election coverage. So per two people who were not named or familiar, the deal was signed just before the midterms. No financial terms have been, announced. But on election night, this will be presumably 2028, Claude users who asked for polling updates get up to the minute information provided by and attributed to the AP. Until Tuesday, Anthropic had not struck any content licensing deal with a digital publisher. OpenAI, of course, has more than a dozen Axel Springer, News Corp, The Atlantic, and others. Digital media executives told Adweek in August that Anthropic believes its use of publisher content is legally permissible under fair use, but they're about to go public. Do you think this is an IPO thing? Like, we gotta we gotta close down this line of attack against us before we go public. [00:41:21] S0: It's not fair use to use their information real time. And if they have a competing business in that space, [00:41:30] S2: and
00:41:30 · Lon Harris, Jason Calacanis
[00:41:30] S2: and [00:41:31] S0: you're monetizing it. So people... Pretty clearly Anthropic is monetizing Claude. So it is illegal, it is not fair use, and it's an unknown issue if using it as training data is illegal or not. They've paid huge settlements. They had a billion dollar settlement for using pirated copies of books. There's still a remaining question if you buy the book, if you're allowed to train on it. [00:41:54] S2: They get in trouble for bootlegging those books, not the training itself. The training itself is still merged.
00:42:00 · Jason Calacanis
[00:42:00] S0: That doesn't mean the training issue is resolved. It is not. No, it is not. There's a major case between OpenAI and The New York Times where, again, two different things you can get caught with your hand in the cookie jar here. One is steal... Well, it's actually three. One is stealing the content or using stolen content. Anthropic did that, they paid the penalty. The second one is breaking the terms of service. That's what OpenAI pretty clearly did based on the discovery where Greg Brockman, I believe, says something to the effect of, oh, I figured out how to get behind the paywall and get all the content with our scraper. Like, said it in a thing. That's another, illegal thing to do, and that's a civil action that you can do there. And then finally, is it allowed to be trained on? I believe that that's going to be a win by the content companies, specifically music companies, movie companies, TV companies, video content. They're gonna fight tooth and nail. They're gonna say, if you trained on our data, you need to pay us. And then there's a fourth issue, which is on the output. If you do output, and it's somebody else's IP, and that's where this AP news comes in, these final two things. I feel like the first two have been cleaned up by the industry. The last two are unsettled and not cleaned up. And so if you go search the web and they put it into a server on their side and they use all their data, You could be breaking the terms of service, and you could be breaking fair use laws. It's so much easier. If you have a trillion dollar company that's making a $100,000,000,000, I think they should take 30% of their money and give it to training data companies and media companies. That's a big number, $30,000,000,000. That would be bigger than the entire movie industry, obviously. So even if you don't think that, if you just think three percent and $3,000,000,000 a year, $3,000,000,000 a year is a lot of money splash around. If you were to give a billion of that to magazines, books, newspapers, the written words, Substack, they would all line up at the trial. They'd be like, yes. Take the entire Substack. Substack would make it default. If you use Substack, you get cut into it. You get 90% of it. Substack gets 10% of that revenue. Everybody would be like, okay. And you... They could make it opt out, and it's part of the platform. Yeah.
00:44:08 · Jason Calacanis, Lon Harris
That would be a beautiful thing to do. You just rain some... I call it splashy cashies. And I'm winning in Vegas. I'm giving tips to the dealers. I put them in my blackjack hand. If I'm putting a $100 on hand of blackjack, I keep a stack of 10... 5, $10, $25 chips. I'll put a 25 next to it for the dealer if it's running nice or if I wanna change the mojo. Splashy cashy when you're winning. [00:44:29] S2: When I'm in... I I see people do that all the time. I've done it myself before where you where you bet... You put a bet on for the dealer, and then they get to play a look. Do you think they prefer that, or do you think they wish people would just give them the the tip? Because there's a chance they're gonna lose and not win the... That win anything. [00:44:44] S0: Okay. Let me explain something to you. Okay. 5% of people tip. 5%. [00:44:48] S2: That's it? At... In Vegas? [00:44:49] S0: That's I think it's, 5%. Wow. When you're gambling, I think it's, like, 5%. [00:44:53] S2: If you win [00:44:53] S0: in I I Because I think only 40% of people win. Right? So 60% of people leave a loser. 70... Let's say 70% leave a loser. Of the 30%, I think 20% tip, that's where I get the number from. [00:45:06] S2: But it's up the net. Like, I I often leave Vegas, and I lost aggregate over the whole weekend. But when you're up, it's always a nice thing to throw people [00:45:16] S0: People are cheap. [00:45:16] S2: Dollars. I... I'd be surprised. [00:45:18] S0: I think they like I think they like putting the money on the app.
00:45:18 · Jason Calacanis, Lon Harris
[00:45:18] S0: I think they like I think they like putting the money on the app. [00:45:20] S2: Right. I'm always very split because it's like, well, they're working in Vegas. They're a black jack dealer. They must love to gamble. Like, you would think that [00:45:26] S0: how boring it is to do that for eight hours. [00:45:28] S2: Yeah. Like, part of me is like they're... Now they're part of the game. It's fun for them. But then the other part of me is like, but if I was at work, I wouldn't want people to be gambling with buy tips. I'd wanna be like, give me the tips. [00:45:38] S0: One of my home games I play in, and there's a a waitress there. Mhmm. You know, getting people drinks and whatnot, you know, bring some food or whatever. Sure. And, when we play bomb pots, it's $50. Everybody puts $50 in to play this type of poker game called bomb pots. Mhmm. And they say, hey. Anybody wanna put... You know, I'll call her Susan here for privacy. If you wanna put Susan in and, you know, everybody throws, you know, or or, you know, somebody throws in $50, puts her in, and just for that main pot, just for... And so it's eight people, $50. She's the ninth, $4.50. If she wins, she's got a chance to win, you know, up to $450. And she's free roll. And it's capped at, like, I don't know, $1,500. But, you know, $1,500 as a waitress for the night is pretty good deal. [00:46:22] S2: That's not bad. Yeah. [00:46:22] S0: You can take [00:46:23] S2: it. [00:46:23] S0: That's a pretty great deal. Lon's like, I'll put on it. Yeah. [00:46:25] S2: I would... I love served drinks. Yeah.
00:46:25 · Lon Harris, Jason Calacanis
[00:46:25] S2: I would... I love served drinks. Yeah. [00:46:27] S0: That's fine. Absolutely. Alright. Keep going. [00:46:30] S2: Alright. Bloomberg reports that Apple is cutting the daily amount its engineering teams can spend on AI. In October, Apple lowered the daily AI spending cap for its engineering teams for $300 to $200. More cuts are in the works. Some allowances falling to $100 a day. Outside engineering, Apple is reportedly weighing a daily limit as low as $25. That I would blow through $25 a day pretty quickly, I think. It's a big change for a company that has worked to integrate AI into its engineering processes. I was fascinated by this change because those numbers sound very low to me. Do you think there is a type of work you could do at Apple in Cupertino where you would spend less than $20.25 bucks a day total on your on your AI? [00:47:17] S0: I mean, there's a big discussion right now going on in attack. If we're gonna have... If we're gonna be token constrained because we are data center and energy constrained [00:47:26] S2: Right. [00:47:27] S0: Where do the tokens come from? Well, people have been trained to have unlimited tokens. I think you're on a $40 plan for Grok, and our team's on a $40 a month plan that gets you all the stuff together. [00:47:38] S2: So individual is on the 40 a month plan, like, as part of the team. Yeah. [00:47:43] S0: So that's $500 a [00:47:45] S2: year. Yeah.