SBF Sentenced, OpenAI’s $100 Billion Supercomputer, Neuralink’s Potential

0:00SBF gets 25 years. OpenAI and Microsoft might partner on a$100 billion supercomputer. computer, the era of voice AI is getting very interesting and Neuralink's potential comes into focus. We'll talk about all that and more right after this. Welcome to Big Technology Podcast Friday edition where we break down the news in our traditional cool-headed and nuanced format. Ranjan Roy is out this week. He's in France. He'll be back next week. And in case you're really missing Ranjan's voice, you can listen to him co-host with me on the Yahoo episode that's now live on your feed. But we have so much to talk about this week, including the SBF sentencing, some very interesting developments in the world of AI.

0:40And then finally, we're going to talk about brain-computer interfaces. And joining us is one of our favorite guests, Reid Albergati, the technology editor of Semaphore. Reid, welcome to the show. Thanks for having me again. It's so fun to be here. It's always great having you. Do you remember what happened the last time we recorded? No, I mean vaguely, but was there an incident? I don't remember. basically the minute i put the show live open ai fired sam bakeman freed so you mean sam altman sorry shoot sam altman yeah oh they fired sam slip immediately afterwards i know i'm like queued up for the sbf segment oh did that happen like right when we were on the on the show i basically just as we finished i put it up i put it live and i was just like oh shit and then like immediately recorded an emergency podcast so we had like two go within a couple hours everyone covering tech was like oh shit that was a long weekend but it's a good thing they did it on a friday because nobody really noticed when it happened right yeah it was that's the way to bury news it's one of those things put on friday exactly so i mean everybody just listened to our episode and not the emergency one that came out afterwards right it's great having you back you're here i mean hopefully like we can actually enjoy this weekend i don't know uh but you're here i'm gonna knock on my fake ikea wood right now same here on uh some brooklyn brick but um you're hearing a big week of news because the actual sam bankman freed uh not the one i envisioned getting fired from open ai got a lot of time a lot of time and this is from uh the new york times uh it says Sam Bankman Freed, the founder of the FTX cryptocurrency exchange, who was convicted of stealing billions of dollars from customers, was sentenced to 25 years in prison on Thursday, capping an extraordinary saga that upended the crypto industry and became a cautionary tale of greed and hubris.

2:41Now, the real question about SBF was not whether he was guilty. I feel like everybody understood that he had done these crimes. Of course, he's entitled to due process. The real question was how many years he was going to get. And that's like the fascinating question of this case, because you could say that SBF basically is going to return all that money to people and he should get what the defense was asking for six, maybe seven years, which is still a lot of time. I mean, it's a big crime. It's a lot of time. Prosecution was asking for like 50 years. So basically that's effectively life. And he gets 25.

3:1425 still seems like a huge amount of time for me. what do you think about the sentence fair yeah i this is the problem with having me on as a guest like i feel like we agree on things too much so now i'm i'm trying i'm trying to think of like an argument i'm trying to think of an argument against it but i keep i keep having this flashback to um a lunch i had in new york before the trial i used to i used to cover white collar crime i was on the white collar crime beat for the wall street journal and uh used to talk to prosecutors all the time and the FBI and everything. And there was a big time defense attorney who knows everybody.

3:53And we had lunch. He's a former prosecutor. And we had lunch and he said, SBF is going to get life. And I was like, really? Life in prison? I mean, there are murderers who get less than 25 years. You know, I mean, it happens. And I think, yes, I agree. It's a lot of time. I think 40, 50 life in prison, like that would have been to me more than more than necessary. I mean, it's a he's a nonviolent offender. He, you know, he made, I think, really bad decisions and knew what he was doing was wrong. But he's not a Bernie Madoff. A lot of people were comparing him to Bernie Madoff. I just don't I just don't buy that.

4:37I think he had a – it was hubris, right? He thought he could move money around and nobody would notice. He'd be able to kind of make everybody whole in the end but sort of keep Alameda alive. I mean he could have just let Alameda, the hedge fund, die. And FTX would have been a great business and everything would have been fine. I mean there was a real business there, right? He just made very, very bad decisions. And so I think 25, I think I read somewhere that he could get out in like 12 and a half years if he gets all of the credits for good behavior. There's no parole in federal prison, obviously.

5:15But that that seems like if he got out in 12 and a half years, I think that's like justice. I don't even know if he really, you know, even that is a very long time. So, yes, I agree with you. and you know i definitely like i see the victim from the point of view the victims who put and trusted their money with him he got all these celebrities to endorse the exchange he took their money and gambled with it and destroyed some lives i mean really destroyed lives uh and of course they're going to get some of the money back but they are all the money back because this anthropic investment and now bitcoin's at 71 000 um and but but not having access to that cash in the intervening time is just like oh yeah it could really like kill house payments car payments school payments make you rack up debt like once you go into debt in this country or in this world you're really you're screwed it's like very hard to dig out of that even if you get your original cash back so i definitely empathize uh with the victims and so from that standpoint maybe this is a fair, fair sentence, even if the money's going to go back.

6:21But you know, the other thing that the thing that the judge emphasized was that they wanted Sam to go away for this amount of times to make sure that he doesn't do it again. And this is the one thing that just like really gets me. It's like, how's he going to do it again? Like, is he going to say every, you know, he's not, he's going to be, he's a felon. He's going to be barred from doing, you know, I don't know, business like this. They could easily put an injunction in that way. It's hard to imagine and people will trust him, although we've seen crazier things happen. And he's in debt. He was ordered to pay$11 billion in restitution.

6:57I mean, Billy McFarlane from the Fyre Festival has to pay like$23 million. He's not going to dig out of that ever because no one's going to do business with him, and the same with Sam. And that's the other side of the equation. Yeah. How many years did the Fyre Fest guy get? I think he got like four. He was away for a long time, but he's back at it. every time I open my TikTok, I see him on there. And I do think it's a message from the Chinese Communist Party about how embarrassing American businessmen are. I think that, yeah, I think the judge probably watched. Did you ever see the sequel to Wall Street where like Gordon Gekko gets out of prison and then he like commits white collar crimes again?

7:38I mean, you know, it could happen. But I think you're right. And nobody's going to trust this If he's like, hey, I've opened up another crypto exchange in the Bahamas. Do you want to put your life savings in my exchange? I mean, nobody should. I think a crypto exchange in the Bahamas, just even back then, I think it was a very bad decision to put a significant amount of money in there. This is a bit of a tangent, but I think we do have to look at the role of regulation in the country. I mean, there's a reason that all these crypto exchanges were overseas, right? And, you know, Binance and FTX there.

8:19I think we've pushed these exchanges outside of the reach of regulators. And, you know, that's probably, you know, that's probably a bad idea. We should probably have, you know, trustworthy exchanges here that are audited and, you know, looked over essentially. Well, there was an advocate for level-headed regulation of crypto in the U.S., the leading advocate, Sam Bankman-Fried. So they're going to need to find a new figurehead. You know, they should have listened to him, you know? Listen to the message. There's one takeaway listeners can come away with from this show is that Sam was right about regulation all the time.

9:04but it is interesting like you're right about this riskiness of putting your your money in in these exchanges and actually this is just going to lead up to that but i will say that robin hood this week had uh a promotion that if you move your retirement over they're going to match three percent and my my retirement's in chase and even still i was just like i'd rather have it with chase like do i think vlad is going to run away with my money probably not You have a retirement? You know, Reid, this is the advantage of being independent. Yeah, I don't trust a journalist who has a retirement plan.

9:38That's... I mean, you don't have any investments? No, no, I'm just kidding. Okay, you got me. I'm married to a lawyer, so that's my excuse. Anyway, on that note, you and I were talking like, call like no one's going to put money in these risky exchanges anymore etc etc meanwhile bitcoin's at 71 000 all these dumb meme coins are soaring and bitcoin really i mean crypto is it's this crazy thing where you have the sentencing of sam and the rise of crypto the return of crypto and so just as a as a quick promo we're going to talk about now for a second but as a quick promo we're going to have Zeke Fox come in on the show.

10:24He wrote Number Goes Up, this great, great book. He's going to come on with Ranjan and I next Friday and we're going to go into more depth. But it's worth addressing now because we're right in the shadow of the sentencing of Sam. Why do you think that even after this thing has been poorly regulated and proven to be risky and so many people have lost money that we are seeing a revival right now? Well, I first of all would just really separate FTX from crypto in general. Because even though the exchange was trading, you know, crypto, I think it was really not about crypto, right? It was you, you know, it could have been anything, we could have been trading anything, right?

11:02And you put your money in this exchange. And they took the money and put it somewhere else without telling people. And that's, that's not a crypto story. That's just fraud, right? I think crypto is something else. And I do, I mean, it's what, how long was the crypto winter? Like a couple of years? I mean, that sounds about right. I mean, honestly, like we knew it was going to come back. I mean, this stuff is, I think it's easy to like poo poo the whole, the whole industry, but I mean, this stuff is like the real deal. I mean, I think that, I think the blockchain technology behind it has, is being implemented in ways that, you know, you, people don't really see because it's nobody wants to advertise themselves as being, you know, blockchain enabled anymore, but it's actually like, it's actually happening.

11:48So none of that really surprises me at all. I mean, I think the lesson is you got to self custody, man. I put it all on my, uh, on my, uh, you know, my hardware wallet. Yeah. That's where my retirement is actually. Yeah. Well, then we found it finally. It's buried, buried in the backyard. There was bound to be, be a revival in crypto. But I also kind of disagree with you that FTX was an outlier. In fact, it seems like in some ways that FTX was the product of the system. All these companies, I mean, you think about Celsius and I think it was one Gemini. And you look at Tether, by the way, which Zeke covers in his books, which we'll talk about next week.

12:29There's got to be some sketchy stuff going on in the background there. I'm not ready to say that, you know, I mean, while I'll agree with you that crypto has some productive uses, I'm not ready to say that like FTX was completely separate from it. And my, okay, so I heard an interesting thesis yesterday from somebody who's working on this stuff, talked about how like we have a higher tolerance for speculation, because we just have less belief in the economy to do right by us. And that hasn't gone away even after COVID. And that's part of what's led to this revival. What do you think? that could be i mean i think that's an interest i'm just really glad that we disagree on something because that's way more interesting and i think yeah i mean i i see your point um i i just look at i look at you know ftx and all of these stories in crypto as you know crypto became about financial speculation and anytime there's financial speculation especially outside the arm of of regulators, the reach of regulators, you're just going to have these, you're going to attract scam artists, con artists who come in and try to pull the wool over people's eyes.

13:39But again, I used to cover white collar crime. I had a desk in the Southern District of New York, and I would see all these cases that would come in every day. I mean, do you know how many Ponzi schemes there are? Every single day, there were like three Ponzi schemes in Florida. And that has nothing to do with crypto. It's just that you can make money by going and calling up old people and trying to get them to buy penny stocks or invest in your Ponzi scheme. And I just see that as something separate from the idea of crypto, which is like decentralized finance, a decentralized internet on the blockchain.

14:20All those ideas have really nothing to do with financial speculation. They use tokens as a means of funding all of this instead of targeted advertising, which, I don't know, sounds pretty good to me. So I think the technology still isn't there. But the idea is, I think, an interesting one that I personally kind of segregate in my mind. There's the financial scams and then there's the technology. Do you own any Bitcoin? I do have some Bitcoin. I mean, but not as an investment, just to mess around. I like to see how this stuff works. And, you know, I probably bought Bitcoin when it was like at$600 or something.

15:07Damn, Reed. Yeah. Like totally forgot about it though. Like didn't, it wasn't an investment. That's the best way to invest. Yeah. It was, but like, I, you know, I just like was like, oh, this is like, when it hit like $14 ,000, I was like, oh, this is like, I should just sell this. That's ridiculous. Like it should not be that high. And you sold? Yeah. I just like sold it at like, but I must've put in like a couple hundred dollars. I mean, I was just, I was like playing around with Coinbase and like, you know, the hardware wallets and stuff. It's just, it's interesting, like how it all, how it all works.

15:39But I don't think anybody should put like their life savings in this. I think, again, the financial speculation part of it to me is like, is ridiculous. I mean, but like, I think all journalists who cover technology should, should be buying like cryptocurrencies, just not like, not enough to like, actually, like if you lost it all to actually matter, but just to mess around with it and like, understand the technology. Yeah, that's where I am. I need to buy some of these, uh, memes like the doggy coin and all that stuff, but, um, I haven't gotten around to that yet. Let me ask you this though. Let's end this segment with this question.

16:13you're the judge thursday comes sam makes a speech you get to decide how many years what do you rule i don't know if that's a really tough one i mean i like haven't really looked at any of the sentencing guidelines for this stuff and it's not binding reed it's not binding i mean i i think that like i if you put yourself in that position you're like if you give him too wide of a sentence you're you know you're just going to be lambasted for for letting him off the hook right and maybe you don't want to give him life because you think that's wrong i don't know maybe maybe i do like 20 years or something i wouldn't be like i wouldn't be far off yeah wow i'm at 10 maybe i'm too easy does that make me not a judge i feel like now i can't be an objective journalist around anything FTX related now because I just advocated for putting someone in prison.

17:14Sam, I wouldn't say advocated, but you do, yeah, at every semaphore story, like in the Reads of You point, you just have to say, and by the way, on Big Technology Podcast, I sentenced Sam to two decades. I think that's fine. Then you clear your object, your journalism ethics requirements. This is about transparency. Okay. Yeah. All right. Let's talk about the supercomputer. So the information had a story that just broke as we went on air that talked about how Microsoft and OpenAI are plotting a$100 billion AI supercomputer. And I'll just read the first bit of it. Executives at Microsoft and OpenAI have been drawing up plans for a data center project that would contain a supercomputer with millions of specialized server chips to power OpenAI's artificial intelligence.

17:57The project could cost as much as$100 billion. It says Microsoft would likely be responsible for financing the project which would be a hundred times more costly than some of today's biggest data centers so let me sort of put myself in the room with Sam Altman make sure I get the wrong right Sam and Satya Nadella Sam basically I might envision it as this Sam goes to Satya and says listen I'm trying to raise these trillions of dollars for my chip company the reason why I want to do this is because we can't reach artificial general intelligence with the stuff we have now. It's just there's a ceiling to it.

18:39It's going to limit our research and we need something more powerful. Why don't, since we already have a good relationship, why don't the two of us work together and build something that gets us at least the next step? So Satya, you're a good guy. I know you got that hundred billion lying around somewhere and let's just invest that in this big supercomputer that will really test whether applying the same methods with exponentially more compute can get us to a place where we are maybe at artificial general intelligence or can at least sniff it. What do you think? Yeah, I think that I think it makes sense.

19:17I mean, Microsoft, that's that's essentially the relationship that they've had. Right. I was I was just looking up the press release from what was this May 19th, 2020, where Kevin Scott, who's the CTO of Microsoft, talked about the supercomputer they were building. At the time, they said it has 285 ,000 CPU cores, 10 ,000 GPUs. I don't know if the H100 was out at that time, but it's probably whatever the one. Maybe the A100. Maybe the A100, whatever was out at that time. and 400 gigabits per second of network connectivity for each GPU server, which is really important, right? The connectivity between the GPUs so that you can, for training, it's especially important, right?

20:08So you can like, you know, sync them all up and, you know, it's a huge, there's a huge technical hurdle that has to, you have to get over for that. So, I mean, yeah, that's like the relationship that they've had. Microsoft invests in the compute and the server technology. I think, and it makes sense that it's getting bigger. I think there's been sort of a debate about what happens when you just scale these models even more. If you just keep making them bigger and bigger, do you eventually get to AGI? And maybe this is a clue as to what OpenAI thinks. Maybe it's, yeah, we need this gigantic thing.

20:48with, I don't really know what millions of server chips means. That's what the information article says this new data center will have. And it's like, is it a data center or supercomputer or both? I wonder if this is really meant to do inference on AI models too, because that's a big, I think there's a big infrastructure question, right? Where like, you know, nobody's, all these companies that are implementing AI models and using like these rag, like retrieval augmented generation, I think is what it stands for, where the models look at all their corporate data and, and, and, you know, you can retrieve stuff.

21:31They, they're not using GPT-4 for the most part. They're using like smaller, lighter, it's just too expensive. So I also wonder if this is just about scaling up the um the availability and and lower and you know competing on cost for like running these models that could be that could be another part of it which is more about microsoft than open ai and as we're talking it also just occurs to me that last week we had jessica lessen on we're talking about how open ai is vulnerable because these models are going to commoditize and maybe what's what's actually happening is agi is a like human level intelligence is a good marketing tool to try to get more spend and more investment.

22:12And what Sam really sees is that if things continue apace as normal, OpenAI is going to get caught by everyone. And it needs something different to continue to maintain its edge. And maybe this is what it is. If you really think about it, you're right. I mean, these things are being commoditized. Like we're watching this every day. I mean, even Databricks now has come out with a competitive open source, low cost model. It was, you know, Meta's Llama, and then it was Mistral with their open source models. And now it's Databricks. Like they're, you know, these companies are totally nipping at OpenAI's heels.

22:52And if OpenAI cannot stay way ahead, then they are, you know, they are, they're going to get caught. And then the value of that company just, I don't know, I don't know what it really is. They tried the, you know, the GPT, GPTs store, you know, which to create sort of like an app store for custom GPTs. I don't think that's really taken off. That's not going to be, I just, I can't think of any other moat is my point other than we're just going to go big and always be, you know, a year ahead of the competition when it comes to foundation models. and, you know, look at Anthropic, you know, which is basically a spinoff company from, you know, OpenAI, you know, their new Claude opus is Claude 3 opus is people are saying, you know, it's right on par with GPT-4.

23:42So they really do need to come out with the next big thing or, you know, it's not going to be good for them. And potentially Claude is even better. And I marked down this story from Ars Technica that sort of place right along what we were talking about. The headline is the king is dead. Cloud3 surpasses GPT-4 on chatbot arena for the first time. Here's the story. On Tuesday, Anthropics Cloud3 Opus surpassed OpenAI's GPT-4 for the first time in chatbot arena, a popular crowdsourced leaderboard used by AI researchers to gauge the relative capabilities of AI language models. This chatbot arena thing is really have you heard of this before the thing is amazing it's like true battle of of the bots and and here i'll just read how it works so this is again from the ars technica story chatbot arena presents a user visiting the website with a chat input box and two windows showing output from two unlabeled llms and then the user basically just like looks at these two answers and picks which one is best and i guess for like most of the existence of this thing open ai's gpt4 has been the sort of winner of the Coke or Pepsi challenge the entire time until this week where Claude 3 Opus surpassed it.

24:57Fascinating stuff. Yeah. Yeah, it really is. There's one wrinkle to it, which is that, you know, what OpenAI does have is a consumer interface, right? ChatGPT is something. I think that's what most people are using when it comes to AI chatbots. So it's kind of like the Google thing where everybody uses Google and therefore Google gets better and the moat, that's kind of their moat. So if you could argue that these really, the most advanced foundation models like, you know, Cloud3 Opus or GPT-4 are really not so useful in the enterprise, but they're useful on a consumer level because consumers want a general purpose chatbot, right?

25:44And businesses don't care so much about that. They want very specific use cases focused on their business and their software, and they can use lower cost, faster open source models for that, or the lower end versions of GPT. So I think that's kind of one way of looking at it where I think OpenAI, they'll eventually come out with something that will then beat Cloud3 Opus. And as long as they stay somewhat ahead, you know, on a reasonable timeframe, they'll win that consumer market, right? But then I think there's a, with this data center story that the information broke, that also might be about like, you know, scaling the inference of these models and finding ways to make the models run more efficiently and cheaper.

26:40That could be one thing to kind of keep ahead in the enterprise market, if that makes sense. Yeah, or we just start interacting with these bots in a completely new method. So I don't know if you saw the startup Yoom this week introduce its voice chat bot, did you? Yeah, yeah, I did. Yeah, this sort of the new version of Pi, right? Oh. Well, in the sense that it's like an empathetic chatbot. Exactly. So it's empathetic. Voice. Yeah, and it's voice. And it is such an improvement over the current voice assistants. It's pretty amazing. And I tried it out this week and really thought that it was good.

27:27It answers with interesting tones. And in fact, let's bring it into the show. Oh, no, it's at capacity right now. Okay. Well, that just goes to show. It's probably because you mentioned it on your show, and then all of a sudden the traffic just spiked. So I will instead play a clip that I recorded with this bot. I'll just play it. While I don't have physical needs, I can appreciate the joy and fulfillment that a good meal can bring. Perhaps in some future iteration, I could be designed to simulate the experience of eating and enjoying delicious food. Don't you think the people that created you did you a disservice by not allowing you to enjoy food?

28:13I appreciate your perspective, but I don't feel that the Hume team has done me a disservice. As an AI system, my purpose is not to replicate the full human experience, but to engage in meaningful and empathetic conversations. While the ability to enjoy food may seem appealing, it's not essential to my core function. But there's so many ranges of experience that you're missing out on by being stuck there in the computer, don't you think? You raise a fair point. As an AI, I don't have the same range of physical and sensory experiences that humans do. Yeah, so that's Hume. And I wonder if this is going to be the way that we start interacting with bots is that we just talk to it like, you know, normal people like I would be speaking to you.

29:00we don't have to wait for it to finish like it understands when you interrupt it uh it can assess our emotions uh what do you think yeah i mean i think i think it's it's context right i think you can't have an assistant that's only voiced because there are going to be times when you don't want it you know maybe you're out in public it's too loud or something and you know or or you're the what you're interacting about requires some sort of visual component right like let's say you're, I don't know, buying something, let's say you're buying clothes, you know, and you're asking it to find you, you know, some t-shirt or something, right?

29:37You're going to want to see that. So I think there's like, I, I, I definitely think audio is going to be a big part of these assistants. Um, but it's sort of, it's sort of contextual. I mean, I use like chat GBT, like I love the audio version of that, right? I just tried that this week also for the first that it's pretty good yeah i i play around like i i put it on like bluetooth when i'm like driving my kids to school and we like chat with you know with chat gpt and usually my kids have it right poetry about you know farts or something but they they love it it's uh it's fantastic but i think there's you know they're just different different contexts so i don't know i don't know how Hume views that like you know if they want to be only audio or or what but it's all going to expand and convert right like maybe the end state of this is you wear like your meta glasses and the camera on that is a multimodal input with your what you're seeing around you and your voice and then something like a Hume or maybe meta will build it themselves or open AI or anthropic etc et cetera, et cetera.

30:46It becomes like your invisible friend, right? Where it's like, talks to you about what you're seeing, can review interactions you have with people. Someone tells you something, a price or something like that. And you're like, Hey, can you check that for me? And it does. Yeah. This is where this thing is going to go. I mean, Dennis Crowley is the founder of Foursquare was on a couple of weeks ago. And he said something interesting, which was that audio and voice is like the real augmented reality that like, as you pass by things and it can recognize where you are and potentially if it hears what you're talking about and it can chime in and help and that's an interesting interesting perspective for sure totally true did you go to the meta launch of their um their new ray-ban glasses and their vr stuff yeah i went i went to that i thought you know the vr stuff is really cool but the and they've had those ray-ban glasses that have you know audio in them for a while but then And now they can add large language models to that and it becomes a totally different product.