We're talking a little bit about open weight models, open source models. Those two things are very similar. I don't know if our audience cares deeply about the difference between those two things, Jaden. But interestingly now with the Chinese models blowing up, and there is some, I've seen actually some things about Kimmy 3 kind of like going back and forth. Like, is it really that powerful? But the thing we want to talk about today is a little bit hilarious, which is I'm reading now from actually Jaden's AI Chat Daily, which is a daily newsletter that I get. But it's Anthropics. Dario Amadei says he never called for an open weight AI ban. And yet there's other articles like, you know, Business Insider and others saying like Anthropics gets heat for being the only major AI lab not supporting open models. And Jaden, Anthropics is a little bit of an enigma. Right. And Dario is too. Right. Because I totally get this. I mean, what Dario is obviously saying is like, hey, whoa, we got to be sort of like careful here, like these Chinese models, et cetera. And everybody else is like, no, open weights are good and all that kind of stuff. Anyway, do you have more insight into what he actually said? Because the interesting thing about this and where I want to kind of kick it off is Anthropic burst onto the scene, Jaden, after Dario and his sister left OpenAI. Obviously, they left a little while ago, but it really burst onto the scene in 2023 as two things, right? Number one, they were the safety company, in contrast to OpenAI, sort of like supposedly. And the second thing is they had a long context window, which now turns out to be a little bit irrelevant. But in the beginning, this is how Anthropik really defined themselves. They have this long-term benefit, this board, this sort of strange board that oversees their board. They're all about safety. So what is Dario talking about here? What is he thinking about? What do you think is going on with this? I think this is such a fascinating story, but I do also want to do a slight correction from a previous episode that we did where I think I might have conflated open source and open weight model. So just to give everyone a differentiation between the two, the main difference between open weight and open source is that open weight models give you downloadable parameters to run and fine tune a model, while open source models also include the complete training code, architectural receipts, and data transparency that you need to recreate the system from scratch. So if it's truly an open source model, you get everything you need to rebuild that model, where if it's open weight, you more get like a program that you can run. And there's all of the different files that you need to run this locally on your device, but you don't understand exactly how it was built and everything about it. So anyways, I wanted to say that a couple of cool things that are happening right now, I think really excited about all of the impressive work happening in open source and even open weight model. Obviously, this has been something that has been a really controversial topic in Silicon Valley. There's 25 different tech companies. So IBM, Microsoft, Meta, NVIDIA, Perplexity, Palantir, along with tons of others, all signed an open letter. And I think OpenAI actually jumped on the bandwagon with this. And they were all arguing basically to US policymakers, like don't do premature restrictions on open source, on open weight AI models. Now, the reason why this is kind of becoming a big thing is because Anthropic, they recently spent $20 million lobbying and they've spent a lot of money lobbying against open weight models, to Congress. And specifically what they don't like is the Chinese models using distillation attacks, quote unquote, distillation attacks against them where they are distilling their model. So in order to distill a model, you have to pay for a subscription to the model or a license to the model. And then you ask it a ton of questions. You take those questions and you train it into your model and you train your model to talk like the model. So anyways, there's a funny a funny tweet i saw where someone said um bro literally predicted it and it's a retweet of david sacks but it's a tweet of him saying the entire tech industry save for anthropic has come out in favor of open source ai so what happens next will anthropic change its lobbying efforts not likely now the gaslighting begins nobody is trying to ban open source we just want to limit who can use it we just want to limit who can contribute to it We just want to limit how powerful those models can be We just want to make sure the guardrails we lobbied for can be removed The net effect will be the same. They won't stop until they kneecap open source and the rest of the industry needs to watch these guys like a hawk. Okay, David Sachs said that. And then we come to just today, the letter coming from Dario Amadei and the first thing in it, which is funny because it's along the lines of that. Like maybe David Sachs is going overboard. I don't know, right? But like, I probably think he's probably over the money. But the first thing they said was, look, we never called for an open weight AI ban. Essentially what Dario says in this, they never advocated for an open weight ban. He said that they told the US Senate Committee on Banking, Housing, and Urban Affairs last month that Alibaba's Quinn carried the largest known distillation attack against it. and I think in the way that they phrase a lot of this they're like look we're not calling to ban them we just think regulate them which would fall into David Sachs's camp of like a quote-unquote slippery slope where like they start by saying look these things are dangerous and then they try to eventually get to a ban even if they're not calling for a ban today the one thing that I will say is when they use the word distillation attack instead of just distillation why do they have to put the word attack on there and I know this will probably be controversial to a lot of people but like distillation attack it sounds like you're being hacked when in fact you paid money to anthropic to generate those results and now you're training your model off of the generated results and it just think it's so ironic and we already talked about this a hundred times but it's so ironic that in the face of anthropic having to pay over one billion dollar fine for pirating every book in the world and then them going and getting mad at people for quote-unquote stealing the output of what they stole it there's a meme i saw on twitter and i'll finish on this and kick it over to you Connor but basically it's the scene from Princess Bride and it's um it's like the dude and he's just kidnapped the princess and uh you know the the prince or the you know the hero guy is going to try to save her and and uh it says like how can you be trying to steal what I've rightfully stolen or something like that is the quote anyways and it's like anthropic right now so anyways um it's interesting right now anthropic is not I guess the the big thing here is anthropic is not backing down. They're the only people that are not signing this. They're not on board with it. A lot of their kind of like head people at Anthropica, I've been seeing them tweet on X about, you know, this is they're misunderstood and whatever, but they're doubling down on it. They do not want these quote unquote open weight models out there. And I guess the last thing I'll put on this is like, why do we care about open weight models? Like, yeah, we don't really want China to become the leader in AI. I think that's probably clear from like a national security perspective but like why do we care about the open weight models companies care about the open weight models and open ai and anthropic both feel threatened due to the open weight models because you save so much money if you could legitimately run a model on your device or on prem without having to pay for token costs i mean it's a it's a massive cost saving but it's also a massive like cost losing right because there's sort of a there's sort of like a finite amount of intelligence that the world is going to need to consume. And right now, most of that coming, like a lot of that comes from humans and we're shifting a lot of the quote unquote intelligence maybe to robot, like to AI. Anyway, so it's kind of an interesting thought experiment of like how much intelligence do we need? There's the, of course, the argument that like the cheaper it becomes, the more you will use it like with anything, right? With economics. And so anyways, as these things become cheaper, Anthropica is worried that they'll get priced out of the equation if there's something just as good. So anyways, that's kind of the crux of the whole issue. There's so many directions to take this. I know, there's 100 directions, sorry. No, no, no, but one of them is that, first of all, do we think that... Daria was in a weird position, right? I mean, I don't think... I mean I don know Daria Amadei but I can imagine that Anthropic thought that they would be leading the pack this early Maybe they did But I mean with the head start that OpenAI had on them the headstart that Google obviously had on them heck the headstart that Meta had on them back in call it 2023 when, I mean, obviously Anthropics been around for longer, but like, it's just a funny position when the kind of the sort of the chaser in all this all of a sudden becomes the leader, right? All of a sudden, all bets are off, right? It's like now all of a sudden you're trying to protect your own whatever. Also in the face of all these articles and all these just facts where companies went from these idiotic leaderboards on like how much, you know, like how much, how many tokens can you spend, all that kind of stuff, which was, Jaden, you and I have been saying this for a long time. It's, even when we saw it happening, we're like, this is the stupidest incentivization program of all time. Like how much money of ours can you spend on AI to prove you're using AI. It's insane. It's just utterly insane. It made no sense for a million different reasons, which I won't go into now. Now companies are like, whoa, actually, let's do the opposite. Let's see how efficient we can be, obviously, because these giant companies ended up spending $10 million on AI in four months or something like that. And all of a sudden, Anthropics' value was really built on that. I'm not saying that you and I are huge Anthropics users, right? So I'm a, and I have been forever. I mean, in 2024, Jayden on this podcast says, I think Anthropic is the company of the year. So like we go back with Anthropic. The thing about this is, who do you believe? You know, because I think that there's such cult of personality in the AI world. So Sam is the ultimate cult of personality. And then Google like a little less so, but obviously Elon is maybe more of ultimate. But Dario is the same thing, at least, and this is kind of like a weird place to take this. but OpenAI at least has other voices in the room, right? They had Susquevara in the room. They had Greg Brockman in the room. They had Mira Morati in the room. Now a lot of these people are like left or leaving or anything like that. So Sam at least had sort of like other voices. Who are the other voices at Anthropic that are like we know Dario Amadei and what Dario Amadei says, Dario Amadei does. Now the U.S. government stopped working with Dario Amadei because they didn't like him because he like has such strong positions on things. He won't like sort of like hold Sam Altman's hand when they're raising in triumph as sort of like, you know, they assign these like Dario Amadei has opinions. Do you know what I mean? Like and this is it feels like it's one of them. So partially it feels like, well, yeah, Dario is, you know, can you can make the case where it's like Anthropic really is a safety company. If you really are the safety company, then, hey, all of a sudden, open weights and open models may be a little more risky at sort of like at the core of it. But there's a lot of it that isn't risky at all. Open source and open weights aren't risky things by definition. Do you know what I mean? Like it's just like how he's talking about people using it. But he's also talking about it from the position of somebody who's now essentially the industry leader. And the rate that they grew at, and especially with IPOs left and right, like you think that they want to end any of their sort of stranglehold on this now? I just I highly doubt it. The other thing that you're talking about is the other side of this equation, which is I wonder what we can learn. And I don't know because I'm not a historian on this stuff at all. But I wonder what we can learn from cheaper versions of things coming up and people saying, well, now there's a cheaper X. And so people aren't going to spend X on this brand. Do you know what I mean? Meaning like was there a moment with like iPhone where people like an iPhone is, you know, $900, $1,000. And now there's this new brand. So nobody's like but but no, people trust what they trust. And I get it. I get that sort of like a lot of companies are like, well, do we really want to use the open source? Jaden, I don't know about you. Like I talked to a lot of companies in the work that we do with AI Mindset and like they're very large companies. They're some of the biggest companies in the world. None of them are going open source. None of them. Like, I mean, because these companies, it's not worth it to them to risk it on anything other than Microsoft Copilot, OpenAI's ChadGBT or Anthropics Cloud and sometimes Google. The risk is insane to those companies They would never do it So I just wondering where the risk actually is and whether we believe Dario 100 agree I think the risk is a real issue You know I recently had I think I recently on a podcast said like, look, there's no risks with open source models, like they can't get your stuff. And I had someone messaged me with an interesting message on LinkedIn saying that the one risk vector that you do see with open source models is there's still integrations, right? All of these models aren't just siloed anymore maybe in the early days of chat gpt that was the case but today it's not you know like i log in with google to check my calendar i log in with all these other things and so with a lot of the third-party integrations there's ways to extract your data out of even open source model because the third-party integration right right right but that's kind of the risk you run everywhere so anyways there's uh that that's the risk vector but like as far as um the risk for enterprise i think that when it's a chinese company it's just gonna be a lot trickier i don't think that's gonna stop it because i did i do remember many people using um i'm using you know uh the the what was the emojis a whale what's the chinese company oh yeah deep uh deep seek but also and also but also llama right i mean like llama but they're not chinese right true true llama so like and i think right now depending on how good the models are if the models are like 50 as good people are gonna be like i need the best thinking But if it's almost as good and it's like, you know, 90% cheaper, I think economically we'll see a shift. Maybe not in the biggest companies that are the most risk averse, but there's plenty of SMBs and mid-market companies that they're like, look, we don't really care. That saves us a million bucks a year on tokens. Yeah, we're just going to do that, you know. So I think that's where we'll probably see this going into the future. And I do think perhaps it will eat into Anthropics profit a bit. But honestly, I think all of these businesses are growing so much and they're coming becoming more valuable. It's interesting because I remember in the early days of opening, I releasing their $200 a month subscription. I bought it. I tested it out. You had it for a long time, right? I used it for like you needed to get like Sora video generation and their agents tool. Anyways, at the time, wasn't worth it. It wasn't good enough. I canceled. I was like, that's not good enough. and today i've officially resubscribed to their 20 a month tier because they have uh they have a um equivalent of copilot that's or i'm not copilot of co-work of cloud work where you know it runs on your computer it takes control it looks at your files it generates stuff for you like it's you know it's it's like just as good and finally it feels like that 200 a month because i actually have two cloud max subscriptions as well and and like if you'd asked me like two years ago like are you going to be paying $200 a month for Claude or $400 a month for Claude? I'd be like, no way, that's ridiculous. And today I'm like, I don't blink. I'm like, that's the best $400 I've ever spent. Like the ROI compared to paying a human to do what it does. I'm like, yeah, it's insane. So I think at the end of the day, the cheaper these things get, the more they will be used. There is an economic incentive. So there is going to be adoption. I hope that we see more China out of necessity is making these open source models and open weight models. I hope we see more American companies following suit and not just opening eye, you know, doing throwaway bag of their last model. Even even XAI said that they were going to do it. I don't feel like they've really held on that promise as well. So I hope we see more of that. And I hope there's good ways for them to monetize as well without having to be. I mean, everyone wants to be a trillion dollar company. So that's where they're all racing. Right. But I hope that there's more of that. And regardless, China will disrupt if there isn't an American company and people will use it. Yeah, I was about to comment, but like we are way over time. We could go on and on about this forever. Listen, we'll let you guys go. Sorry, you guys have probably like missed your flight or something like that because you were like listening to this amazing podcast. Guys, if you are open to leaving a rating and review, you know that it would mean the world to us. We are so unbelievably close to 200. We would absolutely love that. And also, don't forget to check out Jayden's AI box.ai. It would be phenomenal. Ooh, Jayden, we're 11 away. 11. Come on, guys. Okay, listen, guys, we will see you all on the next episode.
