SBF Convicted, OpenAI Is Vulnerable, Ozempic Effect Gone Overboard

0:01SBF has been convicted on all seven counts. Wait, is opening AI much more vulnerable than you think? Jeff Bezos is moving to Miami and the real Ozempic effect. All that and more coming up right after this. Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional cool-headed and nuanced manner. We have a very big week of news this week, including big news on the FTX front, but then also plenty more on AI, on Bezos, on Amazon, and of course, on Ozempic, something we haven't talked about yet, but we will today. Joining us as always on Friday is Ranjan Roy of Margins.

0:38Welcome, Ranjan. I think the COVID and Zerp era is officially over. SBF is going to jail. This is a big day in my life right now. Absolutely. I mean, it took four hours to decide to convict him. And apparently that was even longer because they knew that if they took like a few extra minutes, they would get free dinner. That's what I'm hearing. but are you surprised that it took such a short amount of time to convict Sam on all seven counts? No. And for regular listeners over the last few weeks, Alex and I have regularly tried to play the role of SPF defense lawyer, and we failed miserably. And I'm very glad that the way this ended the actual conviction went, because it was a reminder that it was very difficult for anyone to try to come up with any reason he can get off into me.

1:29I think the most important thing is with these kind of trials, there's always the worry that this stuff could be too complicated. The actual intricacies of what market making is and what an exchange is versus a hedge fund and a risk taker, all of these things could have been complex. But when all three of your friends and closest co-workers and co-founders all turn against you, I think that was pretty much the nail in the coffin. And Dan Permac of Axios put a poll up on Twitter yesterday asking people they think SBF was guilty or innocent. And 99 % said guilty and 1 % said not guilty. And I was like, wow, like Sam is lucky that Twitter is in his jury pool, but he actually would have been better off because at least there was some doubt there.

2:18So seven counts. And what we're seeing now is kind of the reaction from some of the people that were there in the very early days. Of course, Sequoia decided to invest a tremendous amount of money there. And Alfred Lin, who was the Sequoia partner who made the investment, came out and said that this confirms what we already knew. SBF misled and deceived so many, including myself and Sequoia. But just in June 2023, Ranjan, you point out that speaking at Bloomberg's tech summit, he said if he was tasked with evaluating FTX again and for the first time, he would probably make the same investment decision.

2:55What do you take of that? The fact that he's trying to engage in some revisionist history and all these people who for you who were with Sam up until the end, you know, are finally only now saying that was a big mistake. I do not understand why Alfred Lin is trying to come out and try to make anything of this to deflect blame. Like all anyone has to say, and it's totally not unreasonable, is this was an absolutely manic moment. This was, you know, like the first of all, the rise of crypto in general, then bring in all these other factors around zero interest rate. And then pandemic and in like just masses of money coming in that and to and to do a deal, you had to go along with whatever the hot deal was.

3:42And a lot of people, including Sequoia, made a lot of money with this strategy. So just say, yes, we played fast and loose. It worked sometimes. And then other times it didn't versus anything was particularly untoward or it was all SBS's fault. And let's also not forget the Sequoia profile of FTX and Sam Bankenfried, I feel, live forever in infamy. And for those who remember, there was an article on their blog that talked about how he could be the world's first trillionaire. And then the best part was how they talked about, you know, as he pitched them and convinced them, not only would they own crypto, that they would become the future vision of money and have a total addressable market of everyone, which was my favorite line.

4:29But then it even got better because then they said, also, the fucker was playing League of Legends the entire time. And they thought that was somehow funny or it made him kind of more of a genius and worth including in an investment proposal and justification on their own company blog. Obviously, that was taken down. But I mean, after you make that many mistakes, just own up to it. this is why we're so all in on this story by the way because it's not just a story of crypto fraud it really tells you so much about the whole ecosystem and about an era you can really go in and uh and dissect it and you know speaking of playing fast and loose i mean the sam you know every time there's a big criminal case i wonder why we don't hear more from the defendant who probably has their side of the story to tell and probably wants facts out there that that aren't out there in their central discourse.

5:22Well, now I know why they tend to stay quiet because maybe he would have been convicted otherwise. But one thing is for sure is that Sam Bankman-Fried, through all of his media appearances and his inability to shut up, effectively cooked himself in this case and said so much stuff that was held against him in court that he personally just could not get out of it. I mean, is that a good lesson that you absolutely do not have to tweet through it? No, I think that's hopefully the best lesson from this, because I think that over the last, let's say, maybe five to seven years, I think it's actually been a really, really bad lesson for future up and coming entrepreneurs or just leaders of any sort that the the the lesson has always been you can always tweet through it.

6:11And I won't name names, but that's business leaders, politicians, that's all types of people that the kind of lesson everyone's taken away is the sheer velocity of and like ferocity of communications is how you would get past any kind of problem and deflect everything. And this was a reminder that, I mean, again, how open and shut this case was that it didn't save him. And boy, he tried. I mean, remember the 60 Minutes episode? that to me was i think that was even worse and i still have not read the michael lewis book you have i just finished it yep yeah actually how do you feel now you finished it right before the conviction we talked a little bit last week but but have your feelings on the book changed at all i think i'm going to bring that in right now because we're going to go to sentencing predictions and i think it factors in what i learned at the end of lewis book so okay so sam faces up to 115 15 years in prison, what do you predict he gets and what do you think he should get?

7:16This is a tough one for me because I will readily admit, and sometimes a rarity in punditry, I have absolutely no idea how criminal sentencing works. And I'm happy for that because the actual nuts and bolts of it. But I mean, if we talk about how much do I think he should get, it's a tough one like there has been no i mean what's bernie madoff got life basically uh wait how long did elizabeth holmes get 11 years i think she got 11 years yeah so elizabeth holmes who put people's lives actually physically in danger only got 11 years like there has been no real uh you know very tough sentence on any kind of white collar criminal and like this so is it going to be something like 10 or 11 years or will it be something much much more significant like a bernie madoff um i think it should be interesting but what i think i hope he gets enough that it deters people from doing this kind of thing again okay so i might be a little lewis pilled because at the very end of the book he makes this point that john ray who was in charge of the bankrupt bankruptcy of ftx was able to find seven billion of the eight billion missing assets seven billion of eight or eight eight and change then you add in the fact that bitcoin has appreciated and here it comes and are is is ftx in the green right now and i don't think that should matter for the conviction but i do think that should matter for the sentencing because if they're able to return the money to the people whose money they lost then that matters in terms of how long you're going to put this person in prison wait how did they find seven billion of eight billion from the money that's what that's what i'm i have a hard time wrapping my head around when what like you have some amount of money is spent on renaming the miami heat arena on paying Tom Brady and Giselle.

9:22That's money out the door that's gone, that's spent. Then theoretically you have cash sitting around, cash that's been invested in however many either angel investments, one of which is Anthropic, but then a lot that went into total shit coins. You're telling me they genuinely recovered seven-eighths of all the money that was ever deposited into FDF. That was ever deposited into FDF. I mean, I'm not saying it. John Ray is saying it. And he's the guy. I mean, he's not exactly in favor of these folks. So I think a tremendous amount was largely the clawbacks. They probably made some. I don't know if it was the arena that they clawed back from, but they definitely clawed back.

10:05And then FTX did make money. I mean, it was making money. The whole problem was that it then took the actual customer deposit and lent it elsewhere. But it was making money. So I don't know. Maybe, look, I'm not here to defend Sam Bacon. I'm free, and I'm just taking the public pronouncements. And so I do think that that plays in. I think Sam is going to get 20 years. That's my prediction. I think he should get 5 to 10. Okay. I mean... I'll match your 20. Yeah. I think 20 to 30 is really most likely. Could you... I mean, given that a lot of this money is going to be clawed back, could you give him 5 to 10 and bar him from getting anywhere near anything financial except for a U.S.

10:50Treasury bond for the rest of his life, that would seem fair to me. I'm ready for the 2050 Sam Bankman-Fried redemption tour where he's going to be giving financial workshops and lessons all over the country. Okay, so somebody did take a mid-journey photo of Sam Bankman-Fried and put it in my mentions showing what he would look like after like 40 years in prison or whatever. and it's i i just looking at that image i just and you know how mid-journey makes everybody look better than they actually look um i just have this prediction that sam is going to just get in sick shape you know cut his hair in prison and become like this um object of romantic fascination among geeks geeks everywhere like he's going to become like the next weird weirdo sex in the us martin shkreli shkreli yeah shkreli all right bold predictions here on big technology podcast so speaking of bold predictions and and proclamations i wrote a story in big technology this week that's going to lead us into a discussion of the state of ai today because and i think we've probably talked about this but i don't know if we've talked about it as concretely as i laid it out in big technology this week but i wrote this story talking about how open AI is a lot more vulnerable than you think.

12:12And basically the main premise is that open AI rose to prominence, building great products on top of other people's breakthroughs, including the transformer model. And it had this head start because people didn't fully understand what was going to happen. In fact, if folks listened to Gaurav Nmadeh, who came on the show on Monday, they would have understood that Google itself didn't even realize what it had on its hands with the transformer model in 2017. but everybody is now cued into what this technology can do. And so therefore there's a tremendous amount of competition and there's commoditization.

12:46And I've listed six areas where open AI will face competition and is not stacked up perfectly in terms of its ability to compete. So I'll just go through them very, very, very quickly. One is model commoditization. It owns GPT-4, which is the best model on the market, but you're seeing, you know, new competition from Gemini within Google, that's going to come out imminently. You're seeing Anthropic, right, which is getting these billion dollars of investments, Lama 2, and even who knows, maybe Elon Musk's XAI will factor. So people are going to have choice. When people have choice, we go to two, which is margin compression, which is now if there are so many different LLM models, they're going to have to compete based off of cost, the margin comes down.

13:28And in fact, one entrepreneur told me that he doesn't even see this as a money making operation and the infrastructure is really going to be the thing that that makes money which points to nvidia number three is the hits business you know when everybody has access to this technology and everyone's building you really need hits in order to succeed and open ai has had even a model this year that they were working on that they had to scrap because it wasn't meeting expectations the fourth is the open source threat you know about that very much i had a vc who told me that the trend now is to start on opening start your company on open ai and then build to the point where you're model agnostic where you can plug any model in and the actual experience of the product won't change.

14:07And then when you go to a certain point, you go open source and you customize and you build proprietary. And so you don't rely on open AI anymore. Fifth is this move to smaller models. Open AI or GPT-4 can do everything. And you don't need to do everything if you're a specialized model like Harvey, for instance, for law. And so people are going to move to smaller, more specialized models, which aren't as impressive if you speak with them as a chat GPT, but might even get the job done better. And the sixth is that it's just too close to Microsoft. And you have Anthropic, which is making deals with Google here and Amazon there.

14:41And you have OpenAI that's attached at the hip to Microsoft with$13 billion of funding. And it's going to make it much more difficult to show up prominently in an AWS or a Google. And all this adds up to OpenAI looking much more vulnerable than the popular narrative makes it out to be. okay i've laid the case out and now i turn it over to you how do you react i would say so as a consumer uh have you used the recent updates within chat gpt i'm sure i have but but basically so now it is incredible and i have gushed about what they previously called code interpreter and that called i think it was data analysis and now is simply built into the interface where you can throw in a csv and it can help you do really complex calculations now also doll e3 doll e3 is so much better so so so much better actually usable and even cooler now within chat gpt you can simply type in a basic prompt like i want a unicorn on a skateboard and then because that's the kind of stuff i look for and then it'll actually create an image but when creating that image, expand that prompt a great deal, and then tell you what that prompt was.

16:00So it's actually taking kind of the chat GPT element of let me help you create a prompt that's better, and let me create you the image that's good. So all this stuff, it was a reminder to me that OpenAI, I actually think one thing that would push against a lot of this is their strength has really been as much on the product design side as it has been on the infrastructure hardware model side and and like one example for me was again gpt3 existed forever but chat gpt its most brilliant ui mechanism was how it kind of looks like it's thinking like it could actually create that answer for you and spit it all out in one big chunk but instead the letters appear as though there's this like sentient being thinking and that i honestly made that made made people feel like it's much more magical.

16:49And that's what really helped the virality of it. So all that being said, I think on the consumer side, what they've done just in the last one month is puts them so far ahead for me versus any of the competitors, Claude, which is her man, tropic, their chatbot, and got a bard from Google, anything else. All that being said, I think the most important thing you said was your fourth point, that a good company that's developing on this technology basically should be able to move to the point that it becomes model agnostic. And I have seen that. I've worked on projects where we've gotten there already, and I really believe that's the future.

17:29So to me, the consumer side of open AI is way ahead of everyone else, but that's the side that's going to probably lose money. And then on the enterprise side is where they're really going to need to secure and win these deals and make more and more and more money to make up for all the compute that goes in to the consumer product. And I think more and more companies are going to see, I don't need them for this. I can actually, again, whether it's the open source side, whether it's more specialized models, all of these things. So I went through the entire example of why I think ChatGPT has gotten so good to get to the point where I do agree with you.

18:09And I think OpenAI as a business is in a lot of trouble. Yeah, exactly. And so it's so interesting because there's both the consumer and the enterprise point. And what's going to be the bigger business for them? I think we've always agreed that it's going to be enterprise. And you're right, like their consumer product is great. I think that Dolly 3, which you can also access with Bing, is amazing. It's actually, in some cases, for me, surpassed mid-journey, which is wild because mid-journey was so far ahead. I agree. I agree. Completely. That being said, these are demos for the companies to be able to provide their technology to developers, for companies to build on top of the technology.

18:50And that is where the competition is very, very, very intense. Yep. And even more, the Microsoft thing was interesting for me because, again, and I've gushed about this before, but for anyone who's used their data analysis with it, that's now built into ChatGPT again, throw in a CSV, but rather than having to look up different Excel formulas to do simple analyses or even complicated ones, reminding myself how to do a VLOOKUP, like that stuff just happens instantly just by querying with natural language questions. But Microsoft's probably going to put that in their co-pilot and might not even use OpenAI.

19:29They've announced partnerships with Facebook's Llama. Like they've also been operating at a bit of an arm's length. So where they really have an edge in a way that enterprises need them, I'm not sure. Okay. And that gets to sort of the new models that are starting to come up, right? Because if there's going to be a real commoditization, you need stellar models that are going to show up and are going to be able to make a difference here. And I just read this insider story from a few weeks back as I was researching what the progress is of Google's Gemini, which is the competitor to GPT-4 that I imagine is imminent.

20:10And it is fascinating. So there's a Google VP talking about it in the story. And they say, I've seen some pretty amazing things. Like I'm trying to bake a cake, draw me three pictures of the steps to how to ice a three-layer cake. And Gemini will actually create those images. So these are completely novel pictures. They are not pictures from the internet. It's able to speak in imagery with humans now, not just text. I mean, to me, that is fascinating, right? The fact that these models are moving multimodal and Gemini in particular, if you can communicate in imagery, the power is unbelievable. Yeah, but, okay, so on one side, the multimodality, I think, is, I agree, getting really interesting.

20:54And Google, I do think, especially from, what's their thing, organizing the world's information or whatever the tagline was or used to be. That's basically it. The Google proclamation is, we'll organize the world's information or whatever. Or whatever. Whatever. Whatever. Make the cars fly. We'll put that in the whatever. I was thinking the don't be evil was one of the, that was like the motto, but then they had something to make the world's information. Yeah. Searchable. Yeah. If you can't, don't be evil. If you can't, you don't have, I mean, maybe a little bit, but, but I, yeah, no. So Google and already we generative search getting really interesting.

21:41They invented the transformer. So on the pure model strength, I think they're going to be a formidable competitor. It definitely hasn't translated to either the enterprise sales side or the front end consumer side yet, but they clearly should be a competitor. But to me, there's even a, I've talked to you about this offline before. There's a company writer.com. They just raised a hundred million series B. They have a whole, the way I kind of like was thought of them. It's a third way. So you have either the Googles and Microsofts of the world, just gigantic. They're going to incorporate generative AI into the products you already are in.

22:18Or you have Anthropic, you have OpenAI, these kind of like pure tech AI companies. Rider, what's interesting to me is it's just kind of like a classic enterprise software company. They basically sell full stack generative AI to industries like healthcare, financial services that have really, really specific needs and information structures and hierarchies and regulatory considerations. So at that point, like going in and basically being the company that builds the software that brings generative AI into a company's existing proprietary software, because a lot of these companies do have existing systems, like it's a different way of approaching it.

23:01And I actually think it's a really interesting way because maybe the next giant company is not just a technology company. It's also a more enterprise focus, services, sales, consulting, like a company that brings all this technology across an entire stack of an existing large complex enterprise. Because the idea that, you know, like some giant healthcare giant is going to figure this stuff out on their own, I think is a far cry from reality. I think the idea that an open AI, I just never imagined is going to have some huge sales team or like, you know, customer success operation that's genuinely that will work well.

23:44So, yeah, I think there could be different companies that we're not even talking about right now. Right. That start to actually, which, you know, make an impact. And that's actually kind of exciting. Exactly. Yeah, there's going to be real competition. And it's this whole discussion that goes back to like, are you just going to build a thin wrapper on ChatGPT? Well, what's the difference between a thin wrapper on a cloud instance and a successful SaaS company, right? Yeah, yeah, yeah. And so and what was really interesting that writer they did is so they have a total consumer facing SaaS product, but then built their own foundation models.

24:20And I think that's that kind of full stack approach. You're going to see more and more from companies. Again, you'd mentioned like Harvey Law, you know, other industry or vertical specific where they don't just build the model and the technology, they build you the model and the software. And you go in and it does is able to do both for you and across the entire stack. To me, that seems like a more reasonable business in the future versus whatever. Building on the open AI API. And that's the issue is customizability. People have been talking about that. And so like one of the things that one of the founders I spoke with said was that people are going to compete on price or customizability.

25:01Maybe Anthropik is the one that's going to do that because they also just ended up with another$2 billion from Google this week. And that's$1.5 billion over time, but$500 million in upfront cash. And Ranjan, you texted this to me this week and I was like, you know, get that compute. But the cash actually, as you pointed out, makes a big difference for them. So how do you read this investment for Anthropic? Yeah, well, they are playing everyone very well because, again, they had Google investment, then suddenly a splashy Amazon announcement, then suddenly Google's back in with cash up front as opposed to just those juicy cloud credits.

25:45So I think it's a reminder that like, and we talked about this very thing last week, Anthropic realizes, OpenAI certainly has realized already, their brands empower enterprise sales teams at these giant cloud providers. at the Azure salesperson being able to say, oh, we have open AI as part of our whole suite makes him closing that deal or her closing that deal a lot faster. Same way Google understands that to say Anthropic is getting buzzy right now. So, but I also, what are, I don't know, what they're going to do with all that cash and how they're going to realize those valuations in an actual revenue and monetary level.

26:28That one always worries me and we'll see what they come up with. especially because they have somebody else to worry about. And that is none other than Mr. Elon Musk, who this Saturday... Sorry, we do have a comment from LinkedIn about IBM Watson. We've got Watson love here. We'll get into Elon Musk, but let's also just take a moment to recognize IBM Watson. What are your thoughts on IBM Watson in this entire competitive battle, Alex? I'm speechless. I don't know what to say about IBM Watson. I really, I don't know. I'm going to just turn it over to you. I'm setting Alex up with that one. But I went to IBM Watson conference in 2015.

27:19And I'll never forget, they were walking around, they were handing out chocolates. And they said, these chocolates were designed by Watson. And I'm like, what does that mean? And they're like, it's a recipe that's been generated by AI. And this is what they were doing instead of actually building these products. Like they were, they could, they could have been, I mean, they, they literally had the branding. They won the chess or jeopardy chess, all these things. They were the first. And then they threw it all away. It turns out you need technology and not just marketing. And by the way, you've kicked off a Watson war in the comments, Ron John.

27:58So proud of yourself on that. Yep. We got Watson. I love Watson is trash. Okay. All right. So let's talk about Elon. Okay. Elon is introducing his XAI model this week on Saturday. And it's going to be interesting. I mean, it's coming to limited folks, according to Reuters. And it looks like it's another LLM. He did recruit some real talent, people from DeepMind, people from, I think, people from OpenAI. So I wouldn't count it out right away. What's your expectations here? It's Elon vaguely tweeting about a technology that will help Tesla stock in the short term. So as a typical thing, I'll say my expectations are not profound.

28:46But you're missing the fact that this guy, there would be no open AI if it wasn't for this guy. He not founded it. He recruited its chief scientist. I agree. Without Elon. No, no, I agree. But it's kind of like, in a way, it now got to where it is. And this whole revolution's happened because he went hands off. Yeah. Yeah. I'm saying it. I'm saying it. If he was all in, hands on over there, I don't think any of this is happening in the way it is today. But again, we'll see. The point you made, and I think it was a few months ago when they first announced it, the talent that they recruited was insane.

29:26So they have the team. And resources. It doesn't hurt to have his money behind you also. Nope. Team resources, but... Elon is also in favor of regulation, or at least some might touch AI regulation. But now, I don't know if you've seen it this week, and I think I should probably write about this next week because it got so juicy. Like the entire deep mind conspiracy, like some of the, sorry, the entire deep learning conspiracy, Some of the real original researchers of this technology, they fought pretty publicly on Twitter this week about the alarmist like AI will kill us stuff and talking about how, you know, actually it won't.

30:06Actually, this is just regulatory capture. And then you had Andrew Ng, who is the co-founder of Google Brain. Here's the headline from TechCrunch. Google Brain co-founder says big tech companies are inflating fears about the risks of AI wiping out humanity because they want to dominate the market. And he goes, there are definitely large tech companies that would rather not have to compete with open source. So they're creating fear of AI leading to human extinction. It's been a weapon for lobbyists to argue for legislation. That would be very damaging to the open source community. I think what's interesting right now is we're starting to get some like prominent loud voices talk about how this is all about regulatory capture and people should just shut up about the fears that they have about AI wiping us out.

30:51What do you think about this? I am so happy because we and we have both talked about this on the podcast. And when we were talking about this, like maybe three, four months ago or even before, when was the letter again? I can't even tell. Six months ago. Because the pause would have been. yeah yeah remember the pause the pause uh at the time but that's what both it was so clear that this is a form of like it was such an awkward thing for these companies to say this is going to kill us all but we're still working on it and i'm very glad that uh it's becoming it is almost becoming a meme right now and i've seen plenty of pretty good memes about you know like it's going to kill you all but sorry i need to get back to working on that thing that will kill you all so yeah it's yeah it's it's it never made any again Elon Musk is is tripling down on this he's just talking to Rishi Sunak the UK prime minister about this entire topic and meanwhile cars are on the road with full self-driving like it's it's the disconnect to me never has never made any sense.