E106: SBF's media strategy, FTX culpability, ChatGPT, SaaS slowdown & more

00:00:00.000 | I'm now recording.

00:00:02.000 | Jesus, you look terrible.

00:00:04.000 | What's going on? Did you sleep last night?

00:00:05.000 | Do I look tired?

00:00:06.000 | You do look a little tired.

00:00:07.000 | You look unshaven.

00:00:08.000 | Yeah, what happened last night?

00:00:09.000 | I had a holiday party at my house last night from my office and...

00:00:12.000 | You look destroyed.

00:00:14.000 | Yeah, what's going on? I mean, did you drink?

00:00:16.000 | I did drink, yeah.

00:00:17.000 | What does vodka and oat milk taste like?

00:00:19.000 | He had a White Russian!

00:00:21.000 | A White Russian Big Lebowski style with oat milk.

00:00:25.000 | He's like, "I want a White Russian with oat milk."

00:00:27.000 | Oh my God.

00:00:28.000 | Actually, my White Russian is made with oat leaf.

00:00:30.000 | And please compliment that with a huge punch in the face.

00:00:33.000 | Give me five minutes. I'm going to go shave and...

00:00:36.000 | It's called the hair of the dog.

00:00:38.000 | Yeah.

00:00:39.000 | You know, a little Bloody Mary.

00:00:40.000 | You banana bug.

00:00:41.000 | Banana man.

00:00:42.000 | Should I go get a beer? I'm going to get a beer just to beat this hangover.

00:00:45.000 | Go do it. Yeah, go get a beer.

00:00:46.000 | Hang on, I'll be right back.

00:00:47.000 | I'll be right back.

00:00:48.000 | All right, listen, we have to start with scam, bank run, fraud.

00:01:09.000 | I mean, Sam Bankvin Freed.

00:01:11.000 | I get that wrong sometimes.

00:01:13.000 | He was interviewed by Andrew Orr Sorkin, the suit at Dealbook,

00:01:19.000 | who gave him softball after softball.

00:01:21.000 | Then, the next day, he was on Good Morning America.

00:01:24.000 | You're being really passive aggressive right now.

00:01:26.000 | A little bit.

00:01:27.000 | You're throwing a little shade at our friend Andrew Orr.

00:01:29.000 | Little chippy.

00:01:30.000 | He's sticking the knife in.

00:01:31.000 | He's sticking the knife in.

00:01:32.000 | But J. Cao's got a point.

00:01:33.000 | He's got a point.

00:01:34.000 | A little bit.

00:01:35.000 | A little bit.

00:01:36.000 | A little bit.

00:01:37.000 | Anyway, the New York Times continues to embarrass themselves

00:01:40.000 | by handling Sam Bankvin Fraud with kid gloves.

00:01:42.000 | Wow.

00:01:43.000 | George Stephanopoulos, my Greek brother, Stephanopoulos the Spartan,

00:01:48.000 | came in and absolutely fricassed and filleted Sam Bankvin Freed

00:01:53.000 | on Good Morning America.

00:01:55.000 | Very important to note that Good Morning America,

00:01:58.000 | the segment was between holiday cocktails and the cast of The White Lotus.

00:02:04.000 | And Andrew Orr Sorkin was at the Dealbook finance conference.

00:02:07.000 | But you know, that Stephanopoulos interview was two hours,

00:02:10.000 | but they reduced it down to 10 minutes.

00:02:12.000 | So there were probably a lot of sort of cordial conversation,

00:02:18.000 | banter, and then softball questions.

00:02:20.000 | And then he does stick the knife in and does the fricassee,

00:02:24.000 | but he cuts out all the other stuff.

00:02:26.000 | So he just gets it down to the 10 minutes.

00:02:28.000 | What Stephanopoulos did to him was extraordinary in that he said over

00:02:33.000 | and over again, in the FTX terms of service,

00:02:36.000 | you cannot touch the user accounts, but you at Alameda were taking them

00:02:41.000 | and you were loaning them out.

00:02:43.000 | I don't know who is advising Sam Bankvin Freed at this point,

00:02:48.000 | but why he is talking so much.

00:02:50.000 | He was also on a two-hour Twitter spaces after all this.

00:02:56.000 | At this point, Sax, what do you think is going on here in the mind

00:03:01.000 | of Sam Bankvin Freed and also the media,

00:03:04.000 | which seems to have a very variable way of dealing with this obvious fraud

00:03:10.000 | and crime?

00:03:11.000 | Right. Okay. Well, you know, I can speculate about SBF.

00:03:15.000 | I think if there is a strategy here, it is this.

00:03:18.000 | He is basically copying to criminal negligence in order to avoid the more

00:03:25.000 | serious charges of fraud.

00:03:26.000 | And I think, again, if there's a strategy here, it is,

00:03:29.000 | he saw himself being defined as Bernie Madoff 2.0 in the press.

00:03:34.000 | And if that image, which may well be true, cemented around him,

00:03:41.000 | then prosecutors would never stop.

00:03:43.000 | They would never accept a plea that basically gave him anything less than

00:03:48.000 | a Madoff-like sentence, which would be decades in prison,

00:03:52.000 | maybe a life sentence.

00:03:53.000 | So he is out there doing what lawyers would tell you never to do,

00:03:57.000 | which is basically incriminate yourself, create more of a record,

00:04:00.000 | but he's doing it to change the public perception,

00:04:03.000 | maybe muddy up the public perception, get people thinking that, okay,

00:04:08.000 | he, you know, this, that he's admitting he did something wrong,

00:04:11.000 | but it wasn't deliberate. It wasn't fraudulent.

00:04:13.000 | It was just basically carelessness or sloppiness.

00:04:17.000 | And if he succeeds in muddying the waters enough,

00:04:20.000 | then maybe the prosecutors will give him a plea deal that allows him to have

00:04:23.000 | his life back at some point.

00:04:24.000 | I think that would be the crazy, like a Fox explanation of what's happening.

00:04:29.000 | Now there is, you know, an alternative explanation as well,

00:04:32.000 | which is, I just think that these types of guys, you could call it, you know,

00:04:37.000 | a narcissistic fraudster.

00:04:39.000 | They think they can talk their way out of anything, you know,

00:04:43.000 | because they have, they have, you know,

00:04:46.000 | they've talked their way into getting hundreds of millions of dollars of

00:04:50.000 | investment, billions, billions in some cases.

00:04:54.000 | And so they just feel, and they've been trained by employees,

00:04:58.000 | partners, investors, the press that they can talk their way out of.

00:05:01.000 | Yeah. Well, I think, yeah,

00:05:03.000 | the average person is not really used to dealing with one of these

00:05:06.000 | personality types who is, I mean,

00:05:09.000 | they clearly are smart and they're articulate articulate and they know what

00:05:12.000 | to say and they're crafting their words.

00:05:14.000 | What they've learned through their life is that if they use the precise magic

00:05:18.000 | words with a person, they can pretty much convince them of anything,

00:05:21.000 | get them to do anything. And in particular,

00:05:23.000 | I would say investors tend to fall for this,

00:05:25.000 | not because investors are dumb,

00:05:27.000 | but because investors are so clear about what they're looking for and what

00:05:31.000 | they want. They're predictable. Yeah. Like we're, you know, VCs,

00:05:34.000 | especially we're looking for the a hundred X outcome or whatever.

00:05:36.000 | So it's easy for this type of personality to construct a story to essentially

00:05:43.000 | stroke the erogenous zones of a VC. Whoa. Yeah.

00:05:46.000 | And, and, and sort of trick them.

00:05:48.000 | And so there is probably a positive reinforcement loop that gets created in

00:05:52.000 | the minds of one of these people who,

00:05:54.000 | and they start to think that they can basically talk their way out of any

00:05:57.000 | situation. So I think that would be part of what's going on here.

00:06:01.000 | And, you know,

00:06:02.000 | if you look at sort of the tactics that he's using to do this, you know,

00:06:06.000 | all of a sudden he's trying to portray himself, you know,

00:06:09.000 | before this he was portraying himself as the smartest guy in the room.

00:06:12.000 | Now all of a sudden there's this babe in the woods impression where I didn't

00:06:16.000 | know it was my subordinates. I wasn't really in control.

00:06:20.000 | That was somebody else. Right.

00:06:22.000 | Each individual decision he says looked sensible to him.

00:06:26.000 | It's just, it all added up to something he didn't anticipate. Like, really,

00:06:29.000 | you know,

00:06:30.000 | loaning yourself a billion dollars of, of basically the company's money,

00:06:35.000 | which was basically customer money. That seemed reasonable to you.

00:06:37.000 | I don't know how you defend that individual decision.

00:06:39.000 | And there's many like that, but,

00:06:41.000 | but this is sort of the narrative that he's trying to construct.

00:06:44.000 | And let me stop there,

00:06:46.000 | but I think there's a lot more that can be said to dismantle the narrative

00:06:50.000 | he's trying to create, but I want to let other people get in here.

00:06:53.000 | What's your take on this? And then Freiburg,

00:06:54.000 | can I make a comparison to SBF and Trump through the lens of the media?

00:06:59.000 | So if you go back to 2016, you know,

00:07:03.000 | Donald Trump violated every single establishment bias in the world.

00:07:08.000 | Every single establishment bias that these left progressive journalist

00:07:13.000 | elites had. And so they basically just attacked, attacked, attacked,

00:07:18.000 | attacked,

00:07:19.000 | but then you went into the election and there was a very clear data point

00:07:24.000 | that said, whatever you thought was that was at best limited.

00:07:29.000 | And you missed the tone of the country because 50 plus percent of the country

00:07:34.000 | held a very different view about this person.

00:07:37.000 | And instead of taking a step back and then the left media,

00:07:41.000 | the mainstream media re-underwriting and learning and then saying,

00:07:46.000 | you know what, mea culpa, I got this wrong.

00:07:48.000 | They just double down and they said, no,

00:07:51.000 | it still doesn't meet our priors.

00:07:53.000 | And so we're just going to ring fence this problem.

00:07:56.000 | And we're going to just try to destroy this issue because, you know,

00:08:00.000 | we want to control the narrative and, and, and by result,

00:08:04.000 | we want to control power. Now you look at SBF, it's the exact opposite.

00:08:08.000 | He went to the perfect elite private high school.

00:08:12.000 | Then he went to one of the most prestigious elite private universities,

00:08:17.000 | MIT, his parents teach governance of all things at one of the most elite

00:08:22.000 | liberal institutions in America, Stanford,

00:08:25.000 | they are in the establishment of the progressive left.

00:08:30.000 | And what happened was he took customer funds and all of this money.

00:08:35.000 | He made tens of millions of dollars of political donations.

00:08:39.000 | He wrapped himself in this blanket of a progressive left-leaning cause

00:08:43.000 | called effective altruism.

00:08:45.000 | And all of the mainstream media fell for it and embraced him as well as some

00:08:50.000 | politicians, because it met everything that they themselves also bought into.

00:08:57.000 | And now you have this cataclysmic event,

00:09:00.000 | a multi-deck a billion dollar fraud or bankruptcy,

00:09:04.000 | millions of customer accounts who are frozen, you know,

00:09:08.000 | tens of millions to hundreds of millions to billions of dollars lost and stolen

00:09:12.000 | from them. And they refuse to re-underwrite this kid.

00:09:16.000 | And the reason is because in order to do so, it's like eating your own tail.

00:09:20.000 | And that's why they don't want to do it.

00:09:22.000 | And so this is why you have the media basically allowing him to do an apology

00:09:27.000 | tour. Now, this is his second time at manipulating them.

00:09:32.000 | The first time he was able to manipulate them by basically being one of them.

00:09:36.000 | And now he's allowing them and their desire to basically protect themselves

00:09:42.000 | so that he can create some kind of a defense for himself.

00:09:46.000 | And I just think the whole thing is gross because it misses the entire mood of

00:09:50.000 | the nation. This is an enormous financial fraud that was perpetrated on tens of

00:09:58.000 | millions of people. And there's no accountability because in order to do so,

00:10:03.000 | the media would effectively have to admit that they missed it and they got it

00:10:06.000 | wrong and they refused to do it. And I think that that is the really big

00:10:09.000 | problem that nobody is really speaking out about is like, well, if these folks

00:10:14.000 | are meant to be the last stop to make sure that there's truth and honesty and

00:10:19.000 | transparency in society and you can't count on them. And in fact, they're just

00:10:23.000 | going to reflect their own narrative. What is one supposed to do to learn the

00:10:26.000 | truth?

00:10:28.000 | In a way, what you're saying, and then we'll go to you, Freiburg, is this fraud

00:10:32.000 | was encased in all the gilded facade that America hates right now.

00:10:38.000 | It reflects the institutional rot of America. It reflects every single aspect

00:10:42.000 | of institutional rot that every non elite talks about all the time. But elites

00:10:48.000 | when they have those labels will refuse to give up.

00:10:52.000 | And just to add to that, the thing that's missing, I'd say one of the big issues

00:10:57.000 | with the institutional rot in our country is the lack of accountability when

00:11:00.000 | somebody gets it wrong. We saw this with COVID, right? The health establishment

00:11:05.000 | is saying that they want amnesty and Atlantic magazine was willing to give it

00:11:09.000 | to them. So the point is that this this class of people think that when they get

00:11:14.000 | it wrong, that they're the experts, but when they get it wrong, there should be

00:11:16.000 | no accountability. And so, Jamath, to your point, the media and these

00:11:20.000 | institutions are not willing to re underwrite SBF when he so clearly as a

00:11:25.000 | fraudster.

00:11:27.000 | Freiburg, what do you think the of this theory, you had a large amount of

00:11:32.000 | donations to politicians. Obviously, you have coming from Stanford, MIT, etc.

00:11:37.000 | And then you have these investments, gifts slash advertising slash donations

00:11:45.000 | to ProPublica, Vox, this new publication, Semaphore, the intercept that have all

00:11:50.000 | been uncovered now. Did he do this paying off of all the elites, you know,

00:11:56.000 | splashy cashy giving money to everybody because he knew he was doing a fraud and

00:12:00.000 | that this is evidence of this is a premeditated fraud? Or do you think this

00:12:05.000 | is a deranged individual who just was seeking status?

00:12:08.000 | I don't know. The the motivation, there's a video you can watch this guy.

00:12:13.000 | For some reason, FT x has left up all of their videos on YouTube from three

00:12:19.000 | years ago called quantitative trading seven and a half thousand cell wall of

00:12:24.000 | Bitcoin on finance. It's a 17 and a half minute YouTube video of SBF trading

00:12:30.000 | arbitrage across markets. I think it provides probably the best like natural

00:12:36.000 | non scripted insight into this guy's behavior that you could see.

00:12:42.000 | You know, because it's not like him being interviewed. It's just him living

00:12:45.000 | in his world. And he's just, you know, a mouse trying to get a piece of cheese.

00:12:50.000 | Like, you know, he's like out there. He's, you know, scrambling around in the

00:12:55.000 | markets, he's finding edges, he's finding advantages, and he's, and he's

00:12:59.000 | clearly just taking advantage of them all day, every day. That's who he is.

00:13:03.000 | Now, you put a person like that, in an unregulated environment, and there was

00:13:08.000 | this clustering demand for an unregulated environment because of a lot of what

00:13:13.000 | you guys are saying, which is people have this disdain for the elitism and the

00:13:18.000 | and the institutional rot and all these things. So Bitcoin emerged as a solution

00:13:23.000 | out of 2008 to the you know, the what felt like institutional rot that

00:13:31.000 | governments have a key role in. But when you have no regulation, and you have no

00:13:36.000 | trusted central authority involved, mice that are trying to find cheese will rule

00:13:40.000 | the day. And I think that's what happened here. If it wasn't this guy, it was

00:13:44.000 | someone else, it was going to be someone else. And then all of a sudden,

00:13:47.000 | everyone's clamoring and saying, Hey, we needed the government to protect us. No

00:13:49.000 | one protected us. Someone's got to save us. We're the regulators, we're people

00:13:53.000 | that are supposed to keep an eye on this stuff, when the whole premise of so much

00:13:57.000 | of what was being sold was non regulatory regimes was openness was peer to peer

00:14:02.000 | trust protocols. And it turns out that in that sort of an environment, the mouse

00:14:07.000 | that is hungriest for the cheese will get the cheese. And that's exactly what

00:14:10.000 | happened. I don't know how much of it was I don't agree with him saying I'm

00:14:13.000 | creating intentional fraud, which certainly seems to be the case, versus

00:14:17.000 | him saying I'm going to pay these guys. I don't know how much it was even that

00:14:19.000 | intelligent. But the guy was clearly like, trying to get a piece of cheese.

00:14:23.000 | Okay, so this cheese eater, this rat is a League of Legends. You know, expert

00:14:30.000 | playing on eight different monitors at a time, the cryptocurrency game while

00:14:34.000 | hopped up on speed. Yeah, I'm not saying that to be cruel. I'm saying that

00:14:39.000 | because they admitted it. They talked about it in their staff meetings,

00:14:44.000 | instructing their traders and team members of how to take speed. He admitted

00:14:49.000 | to it in an interview this week. He said that it was all legal prescription

00:14:52.000 | drugs, but they were taking them. Yes, yes. And literally in the same videos

00:14:56.000 | you're referencing people see speed patches, or I don't even understand this.

00:15:00.000 | But there are patches you can put on your body to deliver speed to you at

00:15:03.000 | some, you know, dose or whatever. Saks you buy this theory? No care that this

00:15:09.000 | isn't about the elite side of it. It's about the non elite, the anarchy side of

00:15:14.000 | it. No. And this cheese eating rat was just wanted to eat more cheese. I don't

00:15:19.000 | buy this narrative because I see too much design intentionality, but time

00:15:24.000 | what happened. So in other words, it wasn't just a series of individual

00:15:29.000 | decisions that didn't add up. Many of those individual decisions by themselves

00:15:33.000 | were totally unjustifiable. And moreover, there were too many. There's too

00:15:38.000 | much evidence of sophisticated behavior here. Again, we overnight went from

00:15:43.000 | portraying himself as the smartest guy in the room to the to the babe in the

00:15:46.000 | woods. And so, for example, when you look at the construction of all these

00:15:51.000 | entities and the corporate org chart, you know, of all the related entities,

00:15:55.000 | it's a very sophisticated attempt to obscure and construct certain, you know,

00:16:01.000 | protections. When you look at the way that Alameda was exempted from the

00:16:07.000 | normal margin requirements and FTX, there was the so called backdoors. There

00:16:11.000 | was intentionality there there was intentionality in terms of who was

00:16:14.000 | hired to staff these organizations. Again, they wasn't hired. No board, no

00:16:19.000 | CFO. Yeah, exactly. And the guy who was in charge of compliance was like

00:16:24.000 | tomorrow talked about in previous episode was the guy who was involved in

00:16:28.000 | the ultimate bet poker cheating scandal. You know, not exactly. Yeah,

00:16:33.000 | exactly. Right. Or look at this goofy goofball, Caroline Ellison, who was

00:16:37.000 | put in charge of Alameda, right? His girlfriend. It's it's being done for a

00:16:41.000 | reason, right? He's setting it up in a certain way. And, you know, the one

00:16:45.000 | time I interacted with him at this tech conference, he was sitting there

00:16:49.000 | holding court and he had all of his minions around him who were following

00:16:52.000 | his orders. This was a guy who was controlling his business. He was making

00:16:58.000 | the decisions at, I think, a task level. And he knew exactly what was going on

00:17:03.000 | here. So, look, I just don't buy I do not buy this idea that that he was like

00:17:10.000 | a blind mouse who's just stimulus response, you know, in the moment.

00:17:15.000 | That wasn't that wasn't my point, sex. My point was whether it was him or it

00:17:20.000 | was going to be someone else, it was bound to happen. No, I don't agree

00:17:24.000 | with that criminal would have the idea that we want to have completely free

00:17:28.000 | unregulated Bahamian based trading, you know, environments that we can

00:17:33.000 | supposedly trust because someone puts on a good face when there is no real

00:17:37.000 | regulatory body and regulatory authority overseeing it. At some point, it was

00:17:41.000 | going to happen. Well, no, hold on. It is. Look, Coinbase is a fully regulated

00:17:46.000 | institution. They're not set up in the Bahamas, and they're not unregulated.

00:17:49.000 | He set up and he had no oversight. He had no board. There was no regulatory

00:17:53.000 | regime. There was nothing. How was he able? But okay, so first of all, look,

00:17:56.000 | I don't think this had to happen. I think that again, I think that excuses

00:18:00.000 | too much because it implies that if it wasn't SPF would be somebody else. I

00:18:03.000 | actually think that this was a highly concerted effort. Listen, he courted

00:18:07.000 | regulators. He donated to politicians. He courted the media and donated to

00:18:12.000 | the media. Yeah, he was really good at it. He was unusually good at it. No,

00:18:17.000 | and I totally agree on that. Yeah, super smart, super connected, really

00:18:20.000 | thoughtful design on how he committed this. I only think an insider could

00:18:24.000 | have pulled off something at this scale. I think I agree. This is where

00:18:27.000 | Chamath, you're exactly right. I think you needed to be an insider. This level

00:18:30.000 | of cynicism here is he knew the playbook and he admitted it. You pointed this

00:18:34.000 | out with the chats that were released where he said he he he he. Yeah, look,

00:18:38.000 | I mean, just like he he chat, look how like convoluted and intertwined. All

00:18:42.000 | these people are like Gensler's intertwined with the parents. Yeah,

00:18:46.000 | parents apparently bundled a bunch of money to Elizabeth Warren. You know, he

00:18:50.000 | was dating the CEO of the business that he owned 90% in. There are all these

00:18:54.000 | other random shell companies that he owned 100% of where they were lending

00:18:58.000 | back and forth hundreds of millions to billions of dollars. Yeah, to David's

00:19:02.000 | point, that is a sophisticated con that you have to architect and and

00:19:07.000 | the way that he was able to get away with it is that not a single reporter

00:19:11.000 | or regulator thought to dig in. And the reason I think is because he said all

00:19:18.000 | of the right things that wanted them to embrace him. And the reason is he

00:19:23.000 | admitted it. This is a dumb game that we woke Westerners have to play. I say

00:19:26.000 | the right shibboleth and that everyone thinks we're a good person. Exactly.

00:19:29.000 | Imagine pull that quote up on what he said because it's actually a good

00:19:32.000 | it was what I just said. It's the game that we woke Westerners have to play

00:19:38.000 | and we say the right shibboleth or everyone likes us. He actually said the

00:19:41.000 | most uncomfortable thing out loud, which is look by having gone to Crystal

00:19:45.000 | Springs High School by having professors, my parents that went to Stanford by

00:19:49.000 | having gone to MIT. I can pull this off. That's that's what he said because he

00:19:54.000 | can go with those things because I'm a I'm a champion of effective altruism

00:19:58.000 | that I can justify any of these decisions, how amoral or immoral that they

00:20:03.000 | be because I'm trying to help, you know, my brother stand up a multi-million

00:20:07.000 | dollar pandemic response business. I'm trying to do this. I'm trying to do

00:20:10.000 | that. And all of these regulators and all of these reporters said, okay, you

00:20:16.000 | get the hall pass. Now imagine if you replaced him with some random kid in

00:20:22.000 | some developing country or even from the United States who did went to public

00:20:28.000 | high school who went to some random state school. Do you think that they

00:20:32.000 | could have pulled any of this stuff off? No, you need the patina of the

00:20:36.000 | privilege class. The New York Times what he had the New York Times the

00:20:40.000 | privilege class even after this fraud, the New York Times wrote more of a puff

00:20:45.000 | piece on him than the hit piece. They wrote on Brian Armstrong last year when

00:20:48.000 | Brian Armstrong wouldn't toe the line on allowing politics at work. Remember

00:20:52.000 | that? Yeah, unbelievable. Yeah. So the big I think what your mouth is kind of

00:20:56.000 | saying here is that the big enabler here is not crypto per se. It's all these

00:21:00.000 | institutional biases and elite biases that he was able to play into, partly

00:21:05.000 | because he was a big insider. I mean, in a way he monetized his parents life

00:21:09.000 | work. The problem that I think this allows us to put a fine point on is the

00:21:14.000 | following. You know, in society, we've confused a lot of people to think that

00:21:20.000 | the opposite of liberal is conservative or Republican. And I think that's the

00:21:26.000 | cycle that drives the mind virus inside the mainstream media. The problem is

00:21:31.000 | the opposite of liberal is illiberal. Okay. And what illiberal means is to be

00:21:36.000 | narrow minded and unenlightened. It means to be puritanical. It means to be

00:21:42.000 | fundamentalist. And this is really what it allows us to see. Now we have now had

00:21:47.000 | six years of data, case after case after case, where if you are woke, if you are a

00:21:56.000 | social justice warrior, if you have the right credentials that justify your

00:22:00.000 | upbringing, if you have institutional bona fides that come from your parents,

00:22:06.000 | you get to create the narrative and you get a hall pass. And everybody else

00:22:13.000 | basically is at the subject and the mercy of the mainstream media. And so if you

00:22:19.000 | don't kiss the ring and bow down to them, they will try to destroy you or run you

00:22:23.000 | out of town. But if you are one of them, they will give you a hall pass. And when

00:22:27.000 | it's time for them to change their mind in order to tell the truth, they won't do

00:22:33.000 | it. And so these types of grifts will continue as Friedberg said, because there

00:22:37.000 | is no check and balance without a healthy independent media. There is no way for

00:22:42.000 | all of us to actually know what's really going on. Guys, some person in the media

00:22:47.000 | could have asked the question and dug in deeper around the connections between

00:22:51.000 | Alameda and FTX for the last 24 months. I know could have diligence at a venture

00:22:57.000 | firm. At no point could any person have asked these questions and found ex

00:23:02.000 | employees and said, you know, are there any unseemly connections here between FTX

00:23:07.000 | and Alameda? There was no disgruntled employee. I mean, every company has

00:23:11.000 | disgruntled employee whistleblowers. But here where there was billions of dollars

00:23:15.000 | being made by 10s of people, not a single person who felt on the outs said

00:23:19.000 | anything. Well, he was also giving millions of dollars to press outlets.

00:23:23.000 | Hold on. In donations, the questions weren't asked. And then this kid paid

00:23:27.000 | hush money to the mainstream media. Let me ask a question of you guys. Do you

00:23:30.000 | think that it's the media's responsibility in this context? Or do you

00:23:33.000 | think that there should have been a regulatory authority that had oversight

00:23:36.000 | of this business like there is for every bank and every trading operation in the

00:23:40.000 | United States? And every one of those businesses has a compliance officer

00:23:44.000 | chicken and has regulators up the wazoo making sure that customers are kept safe

00:23:48.000 | and protected? Or do we think that that should should offshore vehicles be

00:23:52.000 | allowed like this? That allow people to operate? It's a reasonable question. But

00:23:55.000 | there was a chicken and egg question. We were all standing around holding our

00:23:59.000 | hands while the CFTC and the SEC were fighting. That's not something that

00:24:04.000 | consumers can be expected to adjudicate. So yes, we should have legislation that

00:24:09.000 | clearly defines all of this. But there were enough parameters that created

00:24:13.000 | regulatory frameworks where a bunch of good actors did operate in them and are

00:24:17.000 | continuing to do so like Coinbase. So I don't think this is a regulatory issue.

00:24:22.000 | I think that if you believe there are people who are supposed to forensically

00:24:26.000 | examine things, and get to the bottom of things and ask hard questions. Those

00:24:31.000 | people did none of that here. And and what's what's even more worrisome is

00:24:36.000 | what they're showing is now with an massive amount of data that shows that

00:24:40.000 | you could ask hard questions. They don't care to because it makes them look bad.

00:24:44.000 | I disagree with this. I think regulators failed here, because they have been

00:24:48.000 | reactive to crypto, they have not been proactive, and they have not been clear

00:24:51.000 | with the crypto community that what they were doing was illegal, and they should

00:24:55.000 | have put the regulations in quicker and they're playing catch up. But all three

00:24:58.000 | groups failed, the media failed, the regulators failed, and VCs failed capital

00:25:03.000 | allocators failed, I apparently to do diligence here and install proper

00:25:08.000 | governance, you cannot put a company like this, you know, in business with

00:25:13.000 | billions of dollars and have no board of directors. No, I agree with you or he

00:25:16.000 | lied to them.

00:25:17.000 | And I agree with you. My point is that while regulators are basically fighting

00:25:22.000 | a territorial turf war, okay, the media could have still done their job. They

00:25:27.000 | chose not to

00:25:28.000 | guys, it was worse than that. Because not only it's not just a case where the

00:25:31.000 | SEC failed to exercise any oversight of him or dig into any of these questions.

00:25:36.000 | He was in the room with them crafting the next set of regulations. He was

00:25:40.000 | working on the regulations that you're talking about that are supposedly needed

00:25:44.000 | breaking them. They let the fox in the hen house. Yeah, he was gonna craft a new

00:25:50.000 | type of regulatory license for these types of exchanges, with the result that

00:25:55.000 | he was going to get one and some of his competitors weren't. This is one of the

00:25:58.000 | things that triggered CZ to basically, you know, do what he did, which is

00:26:04.000 | basically that SPF was trying to get Binance and competitors like that band,

00:26:10.000 | while SPF would be one of the sole people to get the license.

00:26:14.000 | By the way,

00:26:15.000 | he was working with the regulators.

00:26:17.000 | Jason, you just said something derogatory towards CZ he rug pulled him. Did he do

00:26:22.000 | that? Or did he actually expose the fraud?

00:26:24.000 | That's what I mean by rug pulling. Yeah.

00:26:26.000 | He was partners with him.

00:26:27.000 | That's not what rug pull means, Jason.

00:26:28.000 | Okay, well, he was partners with him. And then he realized he was weak, and that

00:26:33.000 | he was doing some stuff that was shady. And he decided he would eliminate a

00:26:35.000 | partner who was creating regulation, as Zach said, whatever you want to say, he

00:26:39.000 | knifed him.

00:26:40.000 | All he did was indicate a desire to liquidate his position in a token that

00:26:44.000 | was supposedly perfectly liquid. And that's basically that caused the

00:26:49.000 | everything to unravel.

00:26:50.000 | Yeah, but these were partners, right? I mean, these were deep business partners.

00:26:53.000 | No, they're not.

00:26:54.000 | And they were partners. They were collaborating on these tokens together.

00:26:57.000 | So they weren't competitors.

00:26:58.000 | No, they weren't.

00:26:59.000 | Jason, like part of the big loan that initiated all of this stuff, like a year

00:27:04.000 | and a half ago, was to buy FTX off the cap table. So, you know, this is all I'm

00:27:10.000 | saying is like you used words and your framing of a guy, you know, who's built

00:27:15.000 | a business is like, he is this nefarious bad actor.

00:27:19.000 | Binance was FTX's first investor.

00:27:21.000 | Hold on a second. But David's right. All the guy did, as far as we can tell

00:27:26.000 | right now, is tweet, I'm selling this token because I don't believe in the

00:27:29.000 | capital structure of this entity. And he got those tokens by being bought out of

00:27:34.000 | the cap table.

00:27:35.000 | I know he was partners with them. So if you don't like rug pulled, how about

00:27:39.000 | backstabbing business partners?

00:27:41.000 | They're not partners.

00:27:42.000 | You still don't understand.

00:27:43.000 | Of course I understand. He was the first investor. So to do this to a company

00:27:48.000 | you were about to.

00:27:49.000 | And then he got bought out.

00:27:50.000 | No, no, he got bought out.

00:27:51.000 | And part of the consideration, hold on, part of the consideration were these

00:27:54.000 | tokens, these FTX tokens.

00:27:56.000 | I understand. Yes. Jason, when Uber went public.

00:27:58.000 | Yes.

00:27:59.000 | And you got distributed stock.

00:28:01.000 | Yeah.

00:28:02.000 | Did you distribute and sell Uber at any point?

00:28:04.000 | Previous to that, I had. I sold some to Masa.

00:28:07.000 | No, no, no, no, no, no. I'm asking you a question. When Uber went public and

00:28:10.000 | you got distributed from Uber, your stock.

00:28:13.000 | So you've never sold a single share of Uber ever?

00:28:15.000 | Not since it's public.

00:28:16.000 | No, I sold it before in private markets.

00:28:18.000 | Okay, but I'm sorry, but that doesn't make Jason a partner of Uber.

00:28:21.000 | You're just a stockholder.

00:28:22.000 | Yeah, this is slightly different, I think. But okay, fine. You guys, if you

00:28:26.000 | guys want to consider.

00:28:27.000 | No, he was the first investor in the company.

00:28:29.000 | They were business partners.

00:28:31.000 | That might have been a historical situation, but their status at the time

00:28:33.000 | that CZ tweeted was that they were competitors.

00:28:36.000 | Okay, fine. I mean, they became competitors. I agree.

00:28:39.000 | And by the way, to Tomas point about this rug pulling language, I think

00:28:42.000 | we're getting kind of down a rabbit hole here.

00:28:44.000 | We're getting on a wheeze here.

00:28:45.000 | CZ did perform a service in this sense.

00:28:48.000 | Okay.

00:28:49.000 | SBF claims that 4 billion more was about to come in.

00:28:52.000 | I personally don't believe that. Sounds like bullshit to me.

00:28:54.000 | But if it is true, that would have been a bad thing.

00:28:57.000 | The more money that came in to that operation, SBF proved that he was a

00:29:01.000 | very poor custodian of customer funds.

00:29:04.000 | For sure. For sure. I'm not defending SBF.

00:29:07.000 | The longer this went on.

00:29:08.000 | I'm just highlighting the language that you use is sort of like, again,

00:29:12.000 | part of that establishment elite narrative.

00:29:14.000 | And I'm just questioning, you should maybe steal man.

00:29:17.000 | Take a second to just steal man.

00:29:19.000 | Yeah.

00:29:20.000 | A more dispassionate view, which is here's a counterparty.

00:29:23.000 | Okay. Yeah.

00:29:24.000 | Who, when he left the cap table, was given half cash, half tokens.

00:29:29.000 | Okay.

00:29:30.000 | And he decided to sell his tokens.

00:29:33.000 | Yeah.

00:29:34.000 | And tweet it publicly and cause a run on the bank.

00:29:36.000 | So it was not a run on the bank.

00:29:38.000 | There was no bank. Where's the bank.

00:29:40.000 | This run on the bank language. Okay.

00:29:42.000 | Is something this was in the semaphore coverage.

00:29:45.000 | Okay. Did it publicly. That's my point.

00:29:47.000 | So I'll steal man.

00:29:49.000 | No, no, no, no.

00:29:50.000 | These tokens are worthless.

00:29:51.000 | I need to liquidate them as fast as possible.

00:29:53.000 | But why would you do that publicly?

00:29:55.000 | Why would you do it privately?

00:29:56.000 | You're about to move the market.

00:29:57.000 | He wanted to move the market to zero.

00:29:59.000 | Yes.

00:30:00.000 | He's letting people know why,

00:30:01.000 | but he wanted to kill his competitor as Chamath was just saying.

00:30:04.000 | He's he wanted to kill his competitor.

00:30:06.000 | He told the market.

00:30:07.000 | I am.

00:30:08.000 | He was trying to do that.

00:30:09.000 | We got to back up here because I think we've done a lot of like 30,000 foot,

00:30:12.000 | like lessons and like takeaways from this whole thing,

00:30:15.000 | but we haven't really established what it is that SBF did wrong.

00:30:18.000 | So I think we need to sort of take a second to unmoddy the waters.

00:30:21.000 | Okay.

00:30:22.000 | And part of that,

00:30:24.000 | I think we should start with this idea of a run on the bank because the

00:30:27.000 | favorite, the press, you've been writing puff pieces about SBF.

00:30:30.000 | I'd say mainly Semaphore, which he was a big donor to.

00:30:33.000 | Yeah.

00:30:34.000 | Even trying to frame it as a run on the bank.

00:30:35.000 | And then that implies that it's not really his fault.

00:30:37.000 | It could happen to anybody.

00:30:38.000 | Lots of banks have had this problem.

00:30:40.000 | Okay.

00:30:41.000 | First of all, they're not a bank.

00:30:42.000 | Banks actually have the legal right under certain conditions to take customer

00:30:46.000 | deposits and loan them out.

00:30:48.000 | Okay.

00:30:49.000 | Yes, they did not.

00:30:50.000 | Their terms of use did not allow that as Stephanopoulos pointed out.

00:30:53.000 | SBF's answer to that was, well, we had this like margin account program.

00:30:57.000 | There were other provisions and other terms of use, but most of the customers

00:31:00.000 | who lost money, the vast majority did not opt into that program.

00:31:03.000 | They never agreed to that.

00:31:04.000 | So that's, that's point number one.

00:31:06.000 | Point number two is I think we need to, to look at this language of margin

00:31:10.000 | account.

00:31:11.000 | Okay.

00:31:12.000 | SBF's explanation of how customer money was siphoned off for his own

00:31:18.000 | personal use, IE to Alameda is that Alameda had a margin account.

00:31:22.000 | So I think we could perform a service here by explaining why it wasn't a

00:31:26.000 | margin account.

00:31:27.000 | And, you know, Jamal and you guys understand this really well.

00:31:30.000 | The way that a margin account works is the following.

00:31:34.000 | Okay.

00:31:35.000 | Because I think some of us have them set up with investment banks.

00:31:38.000 | You go to an investment bank, say Morgan Stanley, and you over, you post

00:31:43.000 | collateral, you actually over collateralize.

00:31:46.000 | So for example, you might take a hundred million dollars of stock posted at

00:31:50.000 | the investment bank, and then they will let you loan a certain percentage,

00:31:55.000 | nowhere near a hundred percent, maybe 50% of that.

00:31:57.000 | So if you have a very, very liquid security, you may get 50% coverage,

00:32:04.000 | which means if you posted a hundred million dollars, you could get a 50

00:32:07.000 | million dollar loan.

00:32:08.000 | Okay.

00:32:09.000 | So you have a hundred million in Amazon stock, you're some Amazon VP, you

00:32:12.000 | can get 50 million loans.

00:32:13.000 | And if it's a private asset, it's anywhere as high as 30%, 35%.

00:32:19.000 | But typically it's about 25%.

00:32:21.000 | My expectation is in a liquid token like this would have basically gotten

00:32:25.000 | five or 10% coverage ratio at the best of it.

00:32:28.000 | And then what happens is you have these maintenance values.

00:32:31.000 | So if all of a sudden the value of these entities multiplied by that

00:32:36.000 | percentage that you're allowed to loan falls below, you have to post money.

00:32:39.000 | That's how a margin account works.

00:32:40.000 | It's just, there is no free lunch in that.

00:32:42.000 | Yes, exactly.

00:32:44.000 | Let me just say quite simply, very simply what I, it appears this guy did.

00:32:48.000 | He took customer deposits in us dollars.

00:32:52.000 | He then converted those dollars into some other asset.

00:32:57.000 | And he had a mark on that asset.

00:32:59.000 | Let's call it a dollar a token.

00:33:01.000 | And then those dollars were moved to somewhere else.

00:33:04.000 | No, this is someone transferred in some other token.

00:33:08.000 | But we need to finish the explainer around the margin account.

00:33:13.000 | Okay.

00:33:14.000 | Because what SPF did is this.

00:33:17.000 | He took customer deposits, gave them to himself.

00:33:21.000 | Just be clear.

00:33:22.000 | No, he gave, he took customer deposits in us dollars.

00:33:25.000 | They were wired in.

00:33:27.000 | Correct.

00:33:28.000 | He took those dollars out and he put a fake token in and he called that.

00:33:32.000 | That's right.

00:33:33.000 | He said, therefore he said, the balance sheet is good.

00:33:37.000 | But the value of that token, it turns out isn't a dollar.

00:33:40.000 | It's 10 cents.

00:33:41.000 | That's right.

00:33:42.000 | It was his, it was his sort of, it was sort of his made up token that

00:33:45.000 | he tightly controlled the trading of and artificially prompted the price.

00:33:50.000 | By the way, it wasn't just FTT.

00:33:52.000 | Yeah.

00:33:53.000 | But here's the thing.

00:33:54.000 | It wasn't just the fact that his collateral was no good.

00:33:56.000 | It was also the fact that, and this is from the bankruptcy filing by the

00:34:01.000 | Enron trustee guy.

00:34:05.000 | He specifically said that Alameda, unlike every other margin account on

00:34:09.000 | the platform, had the auto liquidation provisions turned off.

00:34:13.000 | So wait, we have to finish the thought around how margin works.

00:34:15.000 | So like Tomas said, you over post collateral.

00:34:18.000 | And if the value of that collateral goes down or the, the, the, the, the

00:34:23.000 | position, your trading account, the value of that goes down, you either

00:34:26.000 | have to post more collateral or they will actually liquidate your collateral

00:34:31.000 | to pay off the loan.

00:34:33.000 | So Morgan Stanley will never lose money on a margin account.

00:34:36.000 | Never.

00:34:37.000 | The whole point is because they don't make money on it.

00:34:40.000 | They loan you the money at like, you know, a few percents, like very cheap.

00:34:44.000 | Yeah.

00:34:45.000 | Loan LIBOR plus.

00:34:46.000 | So exactly.

00:34:47.000 | That is not a risk account to them.

00:34:49.000 | And so in the, in the example, let's use an example.

00:34:52.000 | In the case of the VP at Amazon, who's got a hundred million in Amazon,

00:34:56.000 | they have a $50 million loan.

00:34:58.000 | If Amazon loses half its value, then that triggers the automatic selling

00:35:02.000 | of Amazon shares to get it back down to 50% coverage.

00:35:06.000 | So now you have to sell 25 million of Amazon shares.

00:35:10.000 | If the full 50 million was pulled down to get back down to 25 to 50%

00:35:13.000 | leverage.

00:35:14.000 | Yes.

00:35:15.000 | And they don't wait until like Amazon stock is at the exact level where

00:35:19.000 | now the collateral equals a hundred percent of the loan.

00:35:21.000 | They will keep that 50% loan to value and they will liquidate you, you

00:35:25.000 | know, and, and by the way, you can lose your entire amount, right?

00:35:28.000 | So, you know, this is why I'm trading a margin is so risky.

00:35:31.000 | Is that you can get wiped out completely, wiped out very, very quickly

00:35:34.000 | with a small move down because they are the custodians of that Amazon

00:35:38.000 | stock.

00:35:39.000 | They are holding it for you now and they have the right to sell it to

00:35:42.000 | cover your margin.

00:35:43.000 | You're saying that governor, that basic tenant, that basic safety

00:35:47.000 | control was turned off by Alameda and it's even more sinister.

00:35:51.000 | Alameda controlled like 90 or 95% of these FTT tokens and was owned

00:35:56.000 | by Sam Bankman fraud.

00:35:59.000 | So he owned that company.

00:36:00.000 | Then he claims he had no operating.

00:36:02.000 | Listen, what should have happened is with that collateral is that as

00:36:07.000 | the value of their position was going down and or as the value of the

00:36:11.000 | collateral was going down, it should have been liquidated to pay off

00:36:14.000 | the margin loan.

00:36:15.000 | And that did not happen.

00:36:16.000 | And the reason it didn't happen is that Alameda got a special

00:36:19.000 | exception on the platform to turn off auto liquidation.

00:36:22.000 | Therefore it was never a margin account.

00:36:24.000 | If even if it was a margin account, okay.

00:36:27.000 | And, and FTX somehow misadministered the margin account.

00:36:31.000 | It should never have taken other customers deposits and use them to

00:36:35.000 | pay back that money.

00:36:36.000 | What should have happened is if FTX was going to lose money on a margin

00:36:40.000 | account, that would hit the FTX corporate treasury.

00:36:44.000 | Okay.

00:36:45.000 | And when the FTX corporate treasury ran out, the company falls for

00:36:48.000 | bankruptcy then, and then all the other customers hold on their account

00:36:52.000 | is still there.

00:36:53.000 | Their money is there in segregated accounts and in bankruptcy, they get

00:36:57.000 | their money back.

00:36:58.000 | The idea that a margin account could ever cause another customer to

00:37:03.000 | lose money that like, whatever that is, that's not a margin.

00:37:06.000 | There's a great article.

00:37:07.000 | There's a great article.

00:37:08.000 | This one was a journalist that did his job properly.

00:37:11.000 | His name is David Z.

00:37:12.000 | Morris.

00:37:13.000 | He wrote an article in coin desk that summed up for anybody that's

00:37:17.000 | interested, all of the actual fraud and all of the crimes that were

00:37:21.000 | committed in excruciating detail.

00:37:24.000 | And what's so sad about all these interviews in this press tour is if

00:37:27.000 | anybody would just read this article, you can construct the right

00:37:30.000 | questions to ask this guy just based on this one article.

00:37:33.000 | But the point I wanted to make is that one of the most interesting

00:37:37.000 | insights was these guys had lost an enormous amount of money already in

00:37:43.000 | calendar year 21.

00:37:45.000 | And so this is what's so crazy, Jason, about you using language like

00:37:51.000 | rug pulling and nobody actually trying to be clear.

00:37:56.000 | Like you guys are giving this guy a hall pass.

00:37:58.000 | I'm not giving any, any industrious reporter could have found an

00:38:02.000 | employee who said, wait a minute, we just blew a $3 billion hole in

00:38:06.000 | our balance sheet and calendar year 21, 20.

00:38:08.000 | And now we're sitting here at the end of 22.

00:38:11.000 | Hold on.

00:38:14.000 | I need to respond.

00:38:15.000 | I am not giving him a pass.

00:38:17.000 | And for you to blame journalists who are reflecting the crime and not

00:38:22.000 | putting any light on VCs and the capital allocators who made this

00:38:26.000 | investment and who did no diligence and now put governance in it is the

00:38:29.000 | height of arrogance.

00:38:30.000 | Shama.

00:38:31.000 | This is not the press is not doing that.

00:38:32.000 | This is the VCs.

00:38:34.000 | Is the capital allocators fault.

00:38:36.000 | You're blaming the people who are telling the story after the crime

00:38:39.000 | story.

00:38:40.000 | They're covering the story.

00:38:42.000 | I can't handle the truth.

00:38:43.000 | Jason, they're covering the story.

00:38:45.000 | Can I get in here?

00:38:46.000 | Can I get in here?

00:38:47.000 | All right, listen, Jason, I will defend you against tomorrow saying

00:38:50.000 | that somehow you're, you know, that you're letting SPF off the hook.

00:38:54.000 | I know you don't want to let SPF.

00:38:56.000 | However, you are letting the press off the hook.

00:38:58.000 | And the reason why, hold on a second.

00:39:00.000 | The reason why you're using this inaccurate language like rug pulling

00:39:04.000 | and run on the bank when there was no run and there was no bank is

00:39:07.000 | because you've been infected by this language that the media has inserted

00:39:11.000 | into the discourse.

00:39:12.000 | The media, listen, hold on a second.

00:39:15.000 | Investors may have got it wrong last year.

00:39:17.000 | Investors may have got it wrong when they did that last round, but I

00:39:20.000 | think investors now understand what's happening, but the media is still

00:39:23.000 | covering for SPF by miss explaining what happened.

00:39:27.000 | Okay.

00:39:28.000 | Give me a percentage, Saks, of who's to blame here.

00:39:30.000 | VCs who invested and didn't set up any governance regulators who did

00:39:34.000 | not set rules around crypto and then three, the media, what percentage

00:39:40.000 | out of a hundred percent is the investors, the regulators and the

00:39:44.000 | press go three numbers.

00:39:47.000 | I would say that before the fraud got exposed, one third, one third,

00:39:50.000 | one third, one third each before the fraud got exposed, but they were

00:39:54.000 | all jointly and severally liable.

00:39:56.000 | But after the fraud has been exposed, no investor is still defending

00:40:00.000 | SPF.

00:40:01.000 | But I think that the investors who were swindled by him, they feel

00:40:04.000 | bad about it.

00:40:05.000 | So 30, 30, 30 is absurd.

00:40:08.000 | The press had no way to know the fraud was going on.

00:40:11.000 | Just like the VCs who put the money in.

00:40:14.000 | Jason, are you stupid?

00:40:16.000 | Like, wasn't it your stupid journalist that exposed the fraud at

00:40:20.000 | Theranos?

00:40:21.000 | He's the guy that went and did all the work.

00:40:22.000 | John Kerry, you should be celebrated.

00:40:24.000 | Hold on a second.

00:40:25.000 | John Kerry, you went and found this thing when nobody, when everybody

00:40:29.000 | else was like, this is perfect.

00:40:30.000 | It meets all of our priors.

00:40:32.000 | Let me finish, please.

00:40:33.000 | It meets all of our priors.

00:40:34.000 | This is great.

00:40:35.000 | And John Kerry was like, this doesn't pass the smell test to me.

00:40:39.000 | Let me go do some work.

00:40:40.000 | And he pulled one little string.

00:40:42.000 | And over the course of 18 months, he exposed the whole bloody thing.

00:40:45.000 | So hold on a second.

00:40:46.000 | So what is incredible to me is that it was possible to expose this

00:40:50.000 | thing before nobody did.

00:40:52.000 | I agree with David.

00:40:53.000 | It's about equal responsibility before, but afterwards, the bulk of

00:40:57.000 | the responsibilities now sits with regulators to clean it up and

00:41:00.000 | journalists to tell the truth.

00:41:01.000 | Okay.

00:41:02.000 | And now may I respond to that since you call me stupid?

00:41:05.000 | You are delusional.

00:41:07.000 | Number one, every one of those investors in Theranos could have

00:41:11.000 | taking a fucking blood test at two different places like Jean-Louis

00:41:15.000 | Gassier did and write a blog post and prove that Theranos didn't

00:41:19.000 | work.

00:41:20.000 | And they withheld disbelief.

00:41:22.000 | Investors putting in a hundred million dollars, including Rupert

00:41:25.000 | Murdoch, didn't even take a fucking blood test or tell one of their

00:41:29.000 | diligence teams to do it.

00:41:30.000 | The same thing happened here with the investors in FTX.

00:41:34.000 | They did zero diligence.

00:41:36.000 | They set up zero governance.

00:41:37.000 | This was a failure of the investors and the governance for 99% of

00:41:43.000 | the problem.

00:41:44.000 | And then regulators should have caught it.

00:41:45.000 | And the regulators in fact did catch Theranos.

00:41:48.000 | So you're completely wrong.

00:41:49.000 | Chamath again, the journalists come in after the fraud is happening.

00:41:54.000 | The investors and governance is responsible for stopping these

00:41:57.000 | things.

00:41:58.000 | FTX was a failure of governance and investors.

00:42:00.000 | And so was Theranos.

00:42:02.000 | The end.

00:42:03.000 | You're completely wrong.

00:42:04.000 | The question is post post exposure.

00:42:07.000 | Why are you guys obsessed with post?

00:42:09.000 | How about avoiding these things?

00:42:10.000 | You guys are blaming the story is ongoing because the story is

00:42:13.000 | ongoing journalists for something that is capital allocators

00:42:16.000 | responsibility.

00:42:17.000 | It is our responsibility to do diligence.

00:42:19.000 | It is our responsibility to create a board of directors that checks on

00:42:23.000 | Elizabeth Holmes and Sam.

00:42:25.000 | I didn't disagree.

00:42:26.000 | I just call me stupid.

00:42:29.000 | You just call me stupid for pointing out something that you refuse to

00:42:32.000 | accept.

00:42:33.000 | What are you talking about?

00:42:34.000 | I'm the biggest.

00:42:35.000 | I'm the older brother.

00:42:37.000 | I think that maybe you guys are giving a pass to the investors.

00:42:44.000 | You're doing a good impression again.

00:42:46.000 | I'm not doing Fredo.

00:42:47.000 | I'm not doing Fredo.

00:42:48.000 | You guys are being absurd.

00:42:49.000 | This is why people say that we're delusional on this podcast.

00:42:53.000 | Don't call him dumb.

00:42:54.000 | The reason people say we're delusional is that we won't take

00:42:57.000 | acceptance of this issue.

00:42:58.000 | Remember in Fish Called Wanda, like the guy you can't call dumb.

00:43:01.000 | Totally.

00:43:02.000 | He loses it.

00:43:03.000 | He goes berserk.

00:43:04.000 | No, I'm not going berserk.

00:43:05.000 | You guys have a blind spot.

00:43:06.000 | Capital allocators are responsible.

00:43:08.000 | He's like, don't call me dumb.

00:43:10.000 | I'm not going to defend any single one of those investors.

00:43:14.000 | I think that they did a horrible job too.

00:43:17.000 | It's a great episode.

00:43:18.000 | But the reality is I think that if you think that you can -- it's your

00:43:24.000 | decision to defend the mainstream media.

00:43:26.000 | I think that that's fine.

00:43:27.000 | I'm not defending them.

00:43:28.000 | No, you are.

00:43:29.000 | You said they have no responsibility.

00:43:30.000 | No, I'm blaming the VCs.

00:43:31.000 | It's different.

00:43:32.000 | The culpability is with the investor class that has not had proper

00:43:37.000 | governance and diligence.

00:43:38.000 | Jason, how many articles have been written excoriating them?

00:43:41.000 | Yeah, some.

00:43:43.000 | A lot.

00:43:44.000 | They look like fools.

00:43:45.000 | Yeah.

00:43:46.000 | Show me the Washington Post.

00:43:48.000 | Type into Capital.

00:43:49.000 | No, show me the Washington Post, New York Times that's like digging

00:43:52.000 | in to that malfeasance or that lack of oversight and holding them

00:43:56.000 | accountable in a way that you feel exposes this problem to create

00:44:00.000 | change.

00:44:01.000 | Well, if we look at Theranos, those people who invested, including

00:44:06.000 | Draper and --

00:44:09.000 | Just show me the examples.

00:44:10.000 | Rupert Murdoch, they really went after them for sure.

00:44:13.000 | Yeah.

00:44:14.000 | What about here?

00:44:15.000 | Wall Street Journal, one day ago, Sequoia Capital apologizes to its

00:44:19.000 | fund investors for FTX loss.

00:44:20.000 | Venture capital firm tells fund investors that --

00:44:22.000 | That's hard hitting.

00:44:23.000 | -- it will improve due diligence on future investments after 100

00:44:25.000 | billion loss.

00:44:26.000 | They really got them.

00:44:27.000 | They really got them.

00:44:28.000 | Let me read you.

00:44:29.000 | I'm going to read you a sentence from the New York Times coverage of

00:44:33.000 | SBF.

00:44:34.000 | I'm not defending that.

00:44:35.000 | Sam Banquet-Fried is neither a visionary nor a criminal mastermind.

00:44:39.000 | He is a human who made the same poor choice that generations of money

00:44:43.000 | managers have made before him.

00:44:45.000 | Are you effing kidding me?

00:44:46.000 | No, I'm not defending them.

00:44:48.000 | Yes, you are.

00:44:49.000 | I am not defending them.

00:44:50.000 | I just called Andrew Horne's token a suit.

00:44:53.000 | And then Semaphore, who was on the take, who received millions of

00:44:56.000 | dollars in his money --

00:44:57.000 | They're on the payroll.

00:44:58.000 | They're on the bank.

00:44:59.000 | Okay.

00:45:00.000 | How much did they get?

00:45:01.000 | 5 million?

00:45:02.000 | Now, where is their apology?

00:45:03.000 | Hold on a second.

00:45:04.000 | Where is their apology?

00:45:05.000 | Sequoia has apologized.

00:45:06.000 | Where is their apology?

00:45:07.000 | Oh, it has to come.

00:45:08.000 | I'm not defending the press.

00:45:09.000 | Yes, you are.

00:45:10.000 | I'm just saying --

00:45:11.000 | Yes, you are.

00:45:12.000 | No, I am not.

00:45:13.000 | I am literally telling you that the New York Times has been asleep at

00:45:15.000 | the wheel and throwing --

00:45:16.000 | Where's the New York Times apology?

00:45:17.000 | The press is always demanding an apology from everybody else.

00:45:19.000 | I'm agreeing with you.

00:45:20.000 | The press has done a shit job.

00:45:21.000 | Hold on a second.

00:45:22.000 | The press is always demanding --

00:45:23.000 | The press is so incompetent on this.

00:45:24.000 | The Twitter spaces yesterday did a better job of trying to ask

00:45:26.000 | questions and getting to the truth than a single journalist has

00:45:29.000 | done or the collective body of all of journalists.

00:45:31.000 | Absolutely.

00:45:32.000 | Literally, randos on Twitter spaces did a better job than Sorkin.

00:45:35.000 | Let me tell you why no one trusts the press, Jason.

00:45:38.000 | First of all, they have an agenda.

00:45:39.000 | I am in agreement.

00:45:40.000 | But second of all, when they make a mistake, they never admit it.

00:45:42.000 | When's the last time they did an apology or retraction?

00:45:45.000 | When's the last time they did what Sequoia did?

00:45:47.000 | I don't know.

00:45:48.000 | And they need to apologize about New York Times.

00:45:50.000 | Exactly.

00:45:51.000 | You came to a point to it.

00:45:52.000 | I am in agreement with you on that, but I think we have to first

00:45:54.000 | point, and this is where you guys have a blind spot, is what is

00:45:57.000 | the responsibility of capital allocators and governance

00:46:00.000 | and regulators?

00:46:02.000 | I think it's one, two, three.

00:46:04.000 | Our industry is responsible for setting up proper governance.

00:46:07.000 | The regulators are responsible for making sure that scientific

00:46:10.000 | claims are backed up, and then press is a distant third.

00:46:14.000 | You know who I think is responsible?

00:46:16.000 | Go.

00:46:17.000 | One, two, and three.

00:46:18.000 | And then we can talk about four, five, and six.

00:46:20.000 | Okay, four, five, and six.

00:46:22.000 | Capital allocators, regulators, the press a distant sixth.

00:46:25.000 | I agree.

00:46:26.000 | Let's go on to China.

00:46:27.000 | God, it's so spicy today in here.

00:46:29.000 | God, it's so hot.

00:46:31.000 | I do think when we attack the mainstream media, Jason feels a

00:46:34.000 | little tinge of like insecurity and illegitimacy because he were

00:46:39.000 | a journalist.

00:46:40.000 | My personal perception is I think that's bullshit.

00:46:42.000 | No.

00:46:43.000 | I think that you have an incredibly romantic view of the

00:46:48.000 | craft as you practiced it back then, which I think was fully

00:46:51.000 | of integrity.

00:46:52.000 | Yes, I think that's true.

00:46:53.000 | I think that you don't adequately realize how massively the

00:46:58.000 | industry has changed in the last 20 years since you've become

00:47:00.000 | a professor.

00:47:01.000 | I realize it more than you do.

00:47:02.000 | I fully realize that the media has absolutely become biased.

00:47:05.000 | Are they corrupt?

00:47:06.000 | And they have lost, in some cases, yes.

00:47:08.000 | I mean, if you're taking money from SBF and then giving him--

00:47:11.000 | Are they blinded?

00:47:12.000 | Are they biased?

00:47:13.000 | They are corrupt if they're taking money from SBF and then

00:47:16.000 | giving him Kigalov coverage.

00:47:17.000 | Absolutely.

00:47:18.000 | That is the definition of corruption in my mind.

00:47:20.000 | What is it called when you don't take money necessarily like

00:47:23.000 | the New York Times and still treat them with kid gloves?

00:47:25.000 | What is that?

00:47:26.000 | It is extreme bias.

00:47:27.000 | And the New York Times became extremely biased.

00:47:29.000 | Why do you think that bias exists?

00:47:31.000 | They were always left-leaning, but I can tell you why.

00:47:34.000 | They-- when Trump came in, a generation of new journalists

00:47:39.000 | became activist journalists.

00:47:41.000 | They didn't want to tell stories and take it straight down the

00:47:44.000 | middle and let the facts tell the story and let the audience

00:47:46.000 | make their own decision.

00:47:47.000 | They felt an existential risk when Trump came into office.

00:47:50.000 | They got Trump derangement syndrome.

00:47:52.000 | They picked a side like MSNBC and Fox did, and the business

00:47:55.000 | model became, for the New York Times, pick a side and get

00:47:59.000 | the subscribers.

00:48:00.000 | It was a deliberate, cynical choice on the New York Times

00:48:03.000 | part to go full MSNBC or full Fox, the two extremes in

00:48:08.000 | mainstream media, in order to get the subs.

00:48:10.000 | And they literally rallied the troops there to do anti-tech,

00:48:14.000 | anti-Trump coverage, and they became activists.

00:48:18.000 | And when journalists become activists, they are no longer

00:48:21.000 | journalists, they're activists or commentators.

00:48:23.000 | And that's the problem.

00:48:24.000 | It's being presented as journalism when in fact it's

00:48:27.000 | activism.

00:48:28.000 | And that's the problem.

00:48:29.000 | So you're right.

00:48:30.000 | So shout out to Matt Taibbi, who just did a monk debate--

00:48:33.000 | That's well said, by the way.

00:48:34.000 | --on this very topic.

00:48:35.000 | Thank you so much.

00:48:36.000 | And he has a great, great sub stack, basically saying what

00:48:39.000 | you're saying, Jason.

00:48:40.000 | And the best quote is, "The story is no longer the boss.

00:48:43.000 | Instead, we sell narrative."

00:48:44.000 | He's a lifelong journalist, whose father was a lifelong

00:48:47.000 | journalist, and he understands the way the business has

00:48:49.000 | changed.

00:48:50.000 | And it's like what you're saying.

00:48:51.000 | And this is why independent media, whether it's sub stacks,

00:48:54.000 | whether it's call-in shows, whether it's all-in podcasts or

00:48:58.000 | other podcasts--Joe Rogan, Sam Harris, whoever it is--

00:49:01.000 | independent voices are now what consumers are seeking out

00:49:05.000 | because they can sense the bias.

00:49:07.000 | They know Rachel Maddow and Tucker have an axe to grind,

00:49:10.000 | and they're left and right.

00:49:12.000 | They didn't expect the New York Times, Washington Post,

00:49:15.000 | and Wall Street Journal to--they knew they were leaning.

00:49:18.000 | They didn't expect them to pick a side.

00:49:20.000 | Do you think we should cancel--do you think folks are

00:49:23.000 | better off keeping their New York Times subscription or

00:49:26.000 | replacing that New York Times subscription with a basket

00:49:29.000 | of sub stacks?

00:49:31.000 | You answered your own question.

00:49:32.000 | It's the latter.

00:49:33.000 | I think you're on your own as a consumer now.

00:49:35.000 | You're going to have to--and I think this podcast and the

00:49:39.000 | nuance we have--shout out to Freeberg for nuance--

00:49:42.000 | what we've done on this podcast is--

00:49:44.000 | Wasn't that funny, Sax?

00:49:45.000 | --to explain to people--

00:49:46.000 | Freeberg's not the only one with nuance, Jake Helms.

00:49:49.000 | Nobody would describe David Sax with the word nuance.

00:49:51.000 | What, he's the nuance department on this podcast?

00:49:53.000 | What am I?

00:49:54.000 | 100% he is, but you would know that because you leave when

00:49:56.000 | science course starts.

00:49:57.000 | I'm the truth department, okay?

00:49:58.000 | Sometimes nuance--

00:49:59.000 | You're truth bombs.

00:50:00.000 | I'm the truth department.

00:50:01.000 | You can take the both sides departments.

00:50:02.000 | The point is, consumers need to become extremely literate,

00:50:06.000 | and they have to do their own search for truth in today's age.

00:50:09.000 | They shouldn't trust New York Times.

00:50:11.000 | They shouldn't trust us.

00:50:12.000 | They should trust themselves.

00:50:13.000 | They shouldn't trust necessarily the CDC or the World Health

00:50:17.000 | Organization.

00:50:18.000 | They should trust themselves and come up with their own process

00:50:21.000 | for figuring out the truth in the middle of this mess.

00:50:24.000 | By the way, this is a good reflection on what's happened

00:50:27.000 | with the rest of media with respect to the creator class.

00:50:31.000 | Where it used to be the movie studios and a handful of

00:50:35.000 | aggregated creators that made all of the content,

00:50:38.000 | the record labels, and now independent artists,

00:50:41.000 | independent producers, independent creators,

00:50:44.000 | and now independent journalists are going to become the bulk

00:50:47.000 | of volume that's going to be consumed.

00:50:49.000 | It's just a different consumption model.

00:50:51.000 | We've already seen it happen with music.

00:50:52.000 | We saw it happen with movies, and we've seen this disruption

00:50:54.000 | happen across all of these other media classes.

00:50:57.000 | Journalism and what we call the press is very likely going to

00:51:01.000 | be that next layer of disruption.

00:51:04.000 | I would trust having a conversation with you about

00:51:06.000 | science topics over reading a science article.

00:51:09.000 | 100%.

00:51:10.000 | You know, in the New York Times or Wall Street Journal,

00:51:12.000 | if I'm being honest.

00:51:13.000 | 100%.

00:51:14.000 | I would much prefer to talk to you about it,

00:51:15.000 | and if it was markets, I'd rather talk to Chamath,

00:51:17.000 | and if it was SaaS, I would talk to Saks or operating a company.

00:51:20.000 | Speaking of operating a company.

00:51:22.000 | Or politics.

00:51:24.000 | [Laughter]

00:51:27.000 | We'll see. We'll see.

00:51:30.000 | But I think it's about having unique inside insight, right?

00:51:33.000 | Like, that wasn't the case.

00:51:35.000 | And what's interesting is that the people who are the professionals

00:51:37.000 | that have the knowledge and the touch points are also becoming

00:51:41.000 | journalists in the sense that they're also becoming speakers

00:51:44.000 | of their truth.

00:51:45.000 | And I think Twitter is a good enabling platform for this.

00:51:48.000 | We see it on YouTube where like scientists are putting out their

00:51:51.000 | own videos, or market actors, like people that are traders in the

00:51:54.000 | market go out and they put out their own videos,

00:51:57.000 | and they put out their own podcasts,

00:51:58.000 | and I think we're probably a good reflection of that.

00:52:01.000 | In the sense that we are the actors in the market,

00:52:04.000 | and we're not just the independent observer that has kind of a

00:52:08.000 | surface level view.

00:52:09.000 | We have the depth to be able to talk about the things that we choose

00:52:12.000 | to talk about.

00:52:13.000 | And I think that's where consumers find value and will continue to

00:52:16.000 | find value in terms of who the journalist or speaker is that they're

00:52:19.000 | going to start to trust for their information.

00:52:22.000 | I saw that interview you did with Newcomer.

00:52:25.000 | Oh, yeah.

00:52:27.000 | Yeah, I saw that coverage.

00:52:30.000 | I mean, he speaks a lot about this phenomenon of going direct.

00:52:34.000 | And of course he's against it.

00:52:36.000 | Now he interprets going direct as an attempt by newsmakers to avoid

00:52:41.000 | answering tough questions or take tough questions.

00:52:44.000 | I think that's ridiculous because, for example,

00:52:47.000 | I go on CNBC all the time.

00:52:49.000 | I go on Emily Ching and Bloomberg all the time.

00:52:53.000 | I submit to like really tough questions.

00:52:55.000 | I actually like those sort of sparring sessions.

00:52:58.000 | I did hard talk this week.

00:53:00.000 | Have you ever done that?

00:53:01.000 | Yeah, exactly.

00:53:02.000 | So that's not what's going on here.

00:53:04.000 | I think what's going on is we have expertise.

00:53:06.000 | We want to communicate them.

00:53:08.000 | And we do feel like the media has become a very unreliable narrator.

00:53:12.000 | There is too much bias and sloppiness.

00:53:14.000 | Not all of it is agenda.

00:53:16.000 | Some of it's just pure sloppiness.

00:53:18.000 | And there's no reason why we shouldn't go direct.

00:53:20.000 | And people want to hear from us.

00:53:22.000 | The audience wants to hear from us.

00:53:24.000 | Look at Draymond.

00:53:25.000 | Look at Draymond and the success he's had with his podcast.

00:53:27.000 | No basketball player has ever gone direct

00:53:30.000 | and created content like Draymond's created.

00:53:32.000 | And it's totally changed the game.

00:53:34.000 | And he was so clear.

00:53:35.000 | He's like, "We are. I am the media now."

00:53:37.000 | JJ read it.

00:53:39.000 | Old Man and the Three.

00:53:40.000 | Amazing podcast.

00:53:42.000 | I was going to tell you guys a story.

00:53:43.000 | So I was in the Middle East last week--

00:53:45.000 | or this week, sorry.

00:53:47.000 | And I had this crazy experience

00:53:49.000 | where I was trying to understand what was going on in China.

00:53:52.000 | And so I started on CNN.

00:53:54.000 | And the whole thing was the propaganda machine

00:53:58.000 | around a democratic revolt,

00:54:01.000 | pushing for democracy and trying to depose Xi.

00:54:04.000 | Then I moved to Al Arabiya.

00:54:06.000 | So one channel up, I went from channel 10 to channel 11.

00:54:10.000 | And instead, what they were actually doing

00:54:12.000 | was interviewing people on the ground.

00:54:14.000 | And what they were talking about was

00:54:16.000 | literally how these PCR tests have become far too burdensome.

00:54:20.000 | And they just wanted it to end

00:54:22.000 | and more reasonable restrictions to get in and out of quarantine.

00:54:25.000 | Then I went from there to BBC.

00:54:28.000 | And at BBC, they had a China scholar

00:54:30.000 | who was talking about how for decades, actually,

00:54:33.000 | the Communist Party supports local-level protests and demonstrations

00:54:37.000 | because they've realized that it is a part of their political system

00:54:42.000 | to make sure that people feel like they have a say.

00:54:45.000 | And I was, like, taking a step back.

00:54:47.000 | And I'm like, "If you listen to the US narrative

00:54:49.000 | and even Jason, like, in our group chat,

00:54:51.000 | people fomenting for, like, revolution,

00:54:53.000 | and this is Tiananmen 2.0."

00:54:55.000 | And I'm like, "Well, I'm reading two other channels

00:54:58.000 | that tell us a completely different set of things."

00:55:01.000 | And I just thought, "Man, people just really fit the data

00:55:04.000 | to fit their bias."

00:55:06.000 | Yeah, we are projecting.

00:55:08.000 | We want to see a revolution in China.

00:55:10.000 | The people in China want to have their lives back.

00:55:12.000 | Well, I would love to see more democracy in the world.

00:55:14.000 | Yes, guilty as charged.

00:55:17.000 | I would like to see people be more free in the world.

00:55:20.000 | Dictator.

00:55:21.000 | I think most people just want to improve their condition.

00:55:24.000 | And I don't think people are as tied up

00:55:26.000 | on the philosophy of the government

00:55:28.000 | as they are about improving their condition.

00:55:30.000 | And as long as their condition is improving,

00:55:32.000 | they're willing to put up with any form of government.

00:55:34.000 | And history shows that.

00:55:35.000 | By the way, the conditions in China have improved

00:55:37.000 | better than everybody's in the last--

00:55:39.000 | It's better than anyone in history.

00:55:41.000 | They've moved 500 million people out of abject poverty,

00:55:43.000 | and that's the great success of engagement.

00:55:47.000 | Isolationism would not have created that amazing outcome.

00:55:52.000 | 500 million people going out of abject--

00:55:54.000 | You're referring to us throwing--

00:55:56.000 | Apple building factories is what I'm referring to.

00:55:59.000 | Oh, okay, fine.

00:56:00.000 | If they want to build something over there,

00:56:02.000 | I guess that's better than us throwing open our markets

00:56:04.000 | and giving China MFN status to destroy American manufacturing

00:56:09.000 | and build up their economy so they can become

00:56:11.000 | a peer competitor to the United States.

00:56:14.000 | I mean, this is the balance of engagement.

00:56:16.000 | If you engage too much, you give everything up.

00:56:19.000 | Sax, which of the current Republican agenda

00:56:24.000 | do you disagree with most strongly, just as an aside?

00:56:27.000 | Well, most Republicans are in favor of our Ukraine policy,

00:56:31.000 | this sort of unlimited appropriation of weapons

00:56:34.000 | and aid to them.

00:56:35.000 | Don't you disagree with immigration policy

00:56:37.000 | of Republicans and Democrats?

00:56:38.000 | Well, I have a more nuanced position on immigration,

00:56:40.000 | which is I think we need to have a border.

00:56:43.000 | It can't be just like an open border,

00:56:45.000 | which is the de facto policy we have now.

00:56:47.000 | But at the same time, I do think that we should have H-1B visas

00:56:50.000 | and we want to, like Jamal said,

00:56:52.000 | we want to be an all-star team for the world.

00:56:54.000 | We want to have the best people who want to come here.

00:56:56.000 | So there's a balance. It's a balance.

00:56:58.000 | And then, look, I think that I was happy to see

00:57:00.000 | the marriage equality bill finally pass the Senate.

00:57:03.000 | Yes, they did get about a dozen--

00:57:05.000 | 12 Republicans voted for it. That's great.

00:57:07.000 | So that's great.

00:57:08.000 | But that's not the majority, unfortunately.

00:57:11.000 | You know, look, on what I would categorize

00:57:13.000 | as the old social issues,

00:57:16.000 | you know, like gay marriage, like cannabis legalization,

00:57:20.000 | I was on the liberal side.

00:57:22.000 | Yeah, I don't think, you know, banning abortion entirely

00:57:26.000 | and total abolition is going to work for this country.

00:57:28.000 | I think Republicans will lose elections

00:57:30.000 | if they insist on that,

00:57:31.000 | and I think they're getting that message.

00:57:33.000 | So, yeah, I mean, look, I think that--

00:57:37.000 | I've always considered myself to be pretty centrist.

00:57:40.000 | And so you're not a globalist.

00:57:42.000 | You don't believe in open global markets with the U.S.?

00:57:45.000 | In general, I understand the benefits of free trade,

00:57:48.000 | and I don't think we should be isolationist

00:57:51.000 | with respect to trade.

00:57:52.000 | I don't think that we can be a successful country

00:57:54.000 | if we are-- we isolate our economy.

00:57:57.000 | So I do want to trade.

00:57:59.000 | However, with China in particular,

00:58:02.000 | I think we made a mistake

00:58:04.000 | in throwing open our markets to their products,

00:58:06.000 | giving them MFN status, while--

00:58:08.000 | Most favored nation.

00:58:10.000 | Yes, enriching them to the point

00:58:11.000 | where they became a peer competitor to the United States.

00:58:14.000 | Now, look, I understand why we made that mistake 20 years ago,

00:58:17.000 | because everybody thought the theory was

00:58:19.000 | that if we helped China become rich,

00:58:21.000 | that China would inevitably become more democratic,

00:58:24.000 | and they'd be filled with gratitude

00:58:26.000 | towards the United States,

00:58:27.000 | and they'd actually become more--

00:58:30.000 | more hospitable towards us, more westernized.

00:58:32.000 | And I think that theory has just proven to be wrong.

00:58:35.000 | I mean, they have not--

00:58:36.000 | Or it's going very slowly, one or the other.

00:58:39.000 | Let's own our clips here.

00:58:41.000 | Here is Shamat's prediction from episode 61,

00:58:44.000 | and we'll see you on the other side of this quick clip.

00:58:47.000 | My worldwide biggest political winner for 2022 is Xi Jinping.

00:58:53.000 | I think this guy is--

00:58:56.000 | He's firing on all cylinders,

00:58:59.000 | and he is basically ascendant.

00:59:02.000 | So 2022 marks the first year

00:59:04.000 | where he's essentially really ruler for life.

00:59:06.000 | And so I don't think we really know

00:59:08.000 | what he's capable of and what he's going to do.

00:59:10.000 | And so that's just going to play out.

00:59:11.000 | You think he's the biggest political winner, really?

00:59:13.000 | Oh, my God. I think it's going to be a--

00:59:16.000 | He's going to run roughshod,

00:59:17.000 | not just domestically, but also internationally,

00:59:19.000 | because you have to remember,

00:59:21.000 | he controls so much of the critical supply chain

00:59:24.000 | that the Western world needs to be--

00:59:26.000 | I think you're completely wrong.

00:59:28.000 | I think you're completely wrong.

00:59:29.000 | I think he's losing his power. He's scared.

00:59:31.000 | That's why he took out all these CEOs.

00:59:33.000 | He's consolidating power,

00:59:34.000 | because he fears that they're going to win too big

00:59:38.000 | and then displace him.

00:59:39.000 | And he has massive real estate problems over there

00:59:42.000 | that could blow up at any moment in time.

00:59:43.000 | He could face a civil war there.

00:59:44.000 | I think he's totally isolated himself.

00:59:46.000 | Civil war? They don't even have guns.

00:59:48.000 | Every major country is removing their factories

00:59:51.000 | and removing its dependency.

00:59:52.000 | They don't even have any guns there.

00:59:53.000 | What are you talking about?

00:59:54.000 | What are they going to riot with?

00:59:56.000 | Did you not see Tiananmen Square?

00:59:58.000 | Did you not see the riots in Hong Kong?

01:00:00.000 | Are you not paying attention, Shamath?

01:00:01.000 | There's been many riots in China.

01:00:03.000 | Jason, Zulus were crushed.

01:00:05.000 | I'm not saying they will be crushed,

01:00:07.000 | but he still will have massive amounts of,

01:00:10.000 | I believe, protests.

01:00:12.000 | And yeah, he'll have to kill people.

01:00:13.000 | I think the bigger risk is that

01:00:16.000 | China gets better for Xi Jinping,

01:00:18.000 | but worse for everybody else in China.

01:00:20.000 | It's already worse for all the billionaires over there.

01:00:22.000 | It's worse for the tech industry.

01:00:24.000 | You've now got Evergrande,

01:00:25.000 | that whole gigantic debt implosion.

01:00:29.000 | I think there could be contagion from China next year.

01:00:32.000 | I don't think Xi's going to lose his grip in any way,

01:00:35.000 | but I'm not sure China's going to have a good year next year.

01:00:38.000 | Wow. Nailed it.

01:00:40.000 | I think all three of us kind of got this right.

01:00:42.000 | What are you talking about?

01:00:43.000 | You got none of it right.

01:00:44.000 | I said there were going to be riots

01:00:47.000 | and they're going to have a recession.

01:00:49.000 | Jason, let's be honest.

01:00:50.000 | You said that they would be squashed.

01:00:51.000 | That's exactly what happened.

01:00:53.000 | Both things happened.

01:00:54.000 | I actually think I have a pretty decent ability

01:00:56.000 | to steel man pretty concretely the details.

01:00:59.000 | I think that at best,

01:01:00.000 | when it comes to things like democracy

01:01:02.000 | and your belief in US exceptionalism

01:01:04.000 | in a specific political worldview,

01:01:06.000 | at best you straw man.

01:01:08.000 | I think that you get very biased

01:01:09.000 | without seeing the forest from the trees.

01:01:12.000 | The reason I said that is not because I'm like

01:01:14.000 | some huge Xi supporter.

01:01:16.000 | I'm just trying to steel man what happens

01:01:18.000 | when one individual person

01:01:20.000 | gets anointed leader for life

01:01:23.000 | of 1.3 billion people

01:01:25.000 | that then controls 20% of the world's GDP.

01:01:28.000 | There is no other single human being

01:01:31.000 | as powerful as him as of this month.

01:01:36.000 | Can I just say that this show

01:01:38.000 | is going to become insufferable

01:01:40.000 | if every time you sort of said something

01:01:42.000 | in the past that was sort of correct,

01:01:45.000 | we're going to have to replay it?

01:01:47.000 | I was actually playing that one for you, Sax.

01:01:50.000 | I think you're the one who nailed it.

01:01:51.000 | I was giving that was a softball to you.

01:01:53.000 | I nail it every week, J. Cal.

01:01:55.000 | This show is going to slow down

01:01:56.000 | if we play every clip that I got right.

01:01:58.000 | You guys are asking to pull clips all the time.

01:02:00.000 | I just pulled one clip about China.

01:02:02.000 | You nailed it.

01:02:03.000 | No, no, no, look.

01:02:04.000 | I think like and also look what he did.

01:02:05.000 | Chamath's the king of pull a clip.

01:02:06.000 | The Evergrande thing, look what he did this week.

01:02:09.000 | They said, okay, you know what?

01:02:10.000 | The real estate industry can now issue

01:02:12.000 | secondary stock sales,

01:02:13.000 | raise equity and equitize themselves.

01:02:15.000 | So they're going to find a soft landing

01:02:17.000 | for the equity part

01:02:18.000 | for the real estate industry in China.

01:02:21.000 | And now they're reopening.

01:02:23.000 | So I don't know.

01:02:24.000 | I mean, like I'm not sure

01:02:25.000 | what we're supposed to comment.

01:02:26.000 | What I what I will stand by is what I said,

01:02:28.000 | which is I don't think we have a very clear view

01:02:31.000 | about what's going on,

01:02:33.000 | what the substance of these protests are

01:02:35.000 | and what people actually want.

01:02:37.000 | If you're only consuming US media.

01:02:39.000 | And so if you find a way to get a diet

01:02:42.000 | from a bunch of different sources all around the world,

01:02:44.000 | you may get a better sense.

01:02:46.000 | I had an accidental window into that

01:02:48.000 | by being in a completely different part of the world.

01:02:50.000 | This past week,

01:02:51.000 | freeburg, any final thoughts on China?

01:02:53.000 | And what's going on there

01:02:55.000 | before we move on to chat GPT?

01:02:57.000 | Your favorite story?

01:02:58.000 | I think one of the things we often miss

01:03:00.000 | is that China, the CCP

01:03:03.000 | does have their hand on the throttles,

01:03:08.000 | like they throttle up and down.

01:03:10.000 | We always think that it's a linear line

01:03:13.000 | and that it's super dogmatic and fixed,

01:03:15.000 | but there's certainly responsiveness

01:03:17.000 | and the release of the lockdowns

01:03:19.000 | in Guangzhou and Beijing this week

01:03:21.000 | seems to have been a pretty good indication

01:03:24.000 | that when things do get,

01:03:25.000 | when things when the tides do change,

01:03:28.000 | leadership there seems to respond,

01:03:30.000 | not always, but enough to kind of keep things going.

01:03:34.000 | So should they reopen?

01:03:35.000 | I think it's 60% of the population or so

01:03:37.000 | is vaccinated with obviously vaccines

01:03:40.000 | that maybe aren't as don't have the same efficacy

01:03:43.000 | as the ones here in the United States.

01:03:45.000 | Do you think they should open up

01:03:47.000 | and just let it rip?

01:03:49.000 | Or do you think they should still try to

01:03:51.000 | maintain the zero COVID policy?

01:03:52.000 | Because that is the debate right now.

01:03:54.000 | From what point of view, what's the objective?

01:03:57.000 | Because obviously from the objective

01:03:59.000 | of economic growth, they need to open up

01:04:01.000 | and they need to keep their economy working

01:04:02.000 | and they need to keep their labor force engaged

01:04:05.000 | or else they're going to continue to suffer.

01:04:07.000 | So if economic growth is the objective,

01:04:11.000 | they need to open up, right?

01:04:13.000 | If the long term health cost of the nation

01:04:16.000 | balanced against that is the calculus

01:04:19.000 | that they're kind of weighing,

01:04:20.000 | there's probably some more nuance to that.

01:04:23.000 | And certainly my understanding is there may be

01:04:28.000 | a precedent setting, which is,

01:04:31.000 | hey, we've said that it's a zero COVID policy,

01:04:33.000 | therefore we have to hold strict to it,

01:04:35.000 | hold toe to the line,

01:04:37.000 | or else it looks like we're weak.

01:04:39.000 | And so there's also this, you know,

01:04:40.000 | maintaining the authority of the CCP objective.

01:04:44.000 | So there's a lot of maybe competing objectives right now.

01:04:47.000 | Certainly don't have a sense of how they're weighing them all.

01:04:50.000 | But I think that once all those videos came out this week,

01:04:54.000 | you guys saw them, but people were screaming,

01:04:56.000 | there was an apartment on fire,

01:04:57.000 | or the doors were locked with steel beams

01:04:59.000 | on the base of the building.

01:05:01.000 | At least that's what the video said.

01:05:03.000 | I don't know how much truth there is to that,

01:05:05.000 | but that's what was said.

01:05:06.000 | And clearly people are extremely distraught

01:05:08.000 | and unhappy with the conditions of the lockdown.

01:05:12.000 | At some point enough people with enough loud voices,

01:05:15.000 | you know, something's going to change.

01:05:19.000 | I mean, let's just remember like the bargain

01:05:21.000 | that struck in that country and with all countries

01:05:24.000 | is that, you know, the citizenry to some extent

01:05:27.000 | are willing to tolerate

01:05:29.000 | their government so long as their conditions

01:05:34.000 | continue to improve.

01:05:35.000 | And there's a bargain, there's some bargain that struck.

01:05:38.000 | But as that bargain starts to go south for the citizenry,

01:05:43.000 | then that governing entity is at risk.

01:05:46.000 | And I think that that's what we sort of started

01:05:48.000 | to see this week was the conditions are getting

01:05:51.000 | far, far worse and far less livable

01:05:53.000 | for so many people in that country

01:05:55.000 | that the government had to shift.

01:05:58.000 | - Do you think Chamath, this COVID strategy,

01:06:01.000 | and then we'll move on,

01:06:02.000 | was basically Xi Jinping wanting to get to that Congress,

01:06:06.000 | to get to that coronation.

01:06:08.000 | And now that that's over, maybe he can change gears.

01:06:11.000 | And then--

01:06:12.000 | - Like I said, my belief is that I have a very poor access

01:06:16.000 | to enough data to have a, to steel man

01:06:19.000 | what is actually going on there.

01:06:22.000 | But one explanation could actually be that

01:06:25.000 | in the absence of enough hospital infrastructure

01:06:28.000 | and ventilators and a bunch of these other things,

01:06:31.000 | a pretty severe approach to this disease,

01:06:36.000 | I don't know what they know or didn't know.

01:06:38.000 | Maybe they understood the virulence of it.

01:06:42.000 | Maybe that they have a slightly different

01:06:44.000 | aging characteristics of their population.

01:06:46.000 | Maybe they genetically responded to the SARS-CoV-2 virus.

01:06:50.000 | I don't know any of these things enough to tell you, Jason.

01:06:53.000 | - Yeah.

01:06:54.000 | - But the reality is what Freeberg says is right,

01:06:56.000 | which is that you cannot grow an economy

01:06:58.000 | if people are inside locked in their apartments.

01:07:01.000 | And it looks like they have decided

01:07:04.000 | that that's coming to an end and they're going to,

01:07:07.000 | you know, deconstruct all of these things.

01:07:10.000 | So, you know, the Chinese growth engine is coming back.

01:07:14.000 | And I think that that's going to be an important factor

01:07:17.000 | economically that we're going to have to figure out

01:07:20.000 | because it's going to have a huge implication

01:07:23.000 | to American growth and American inflation.

01:07:26.000 | - Sax, if in fact, if the lab league theory is correct,

01:07:31.000 | China might have some insights into this disease

01:07:34.000 | that maybe the West didn't.

01:07:36.000 | Maybe that plays into their policy a bit.

01:07:39.000 | Any final thoughts here?

01:07:40.000 | - Jake, I'm just surprised to hear

01:07:42.000 | that you have a problem with their lockdown policy over there

01:07:46.000 | because aren't they just implementing

01:07:48.000 | Democratic Party orthodoxy?

01:07:50.000 | I mean, isn't this the policy that Tony Fauci

01:07:53.000 | and Barbara Ferrer, you know, all the health experts.

01:07:56.000 | - I was not in favor of lockdowns.

01:07:57.000 | What are you talking about?

01:07:58.000 | You're the one who had all your masks

01:07:59.000 | and had ventilators day one.

01:08:01.000 | - Isn't this the lockdown policy?

01:08:03.000 | - Why are you painting this?

01:08:04.000 | Why am I getting a stray here?

01:08:06.000 | I'm just asking a question as the moderator.

01:08:08.000 | - Isn't this basically--

01:08:10.000 | - Why am I getting a stray?

01:08:11.000 | - Isn't this what Gavin Newsom--

01:08:13.000 | - Don't deflect Jake, I'll answer the question.

01:08:14.000 | - Yes, hold on, let me finish the question.

01:08:16.000 | - I'm not in favor of lockdowns.

01:08:18.000 | I was in favor of if people wanted to stay home, stay home.

01:08:21.000 | And then if people wanted to go out and take the risk,

01:08:23.000 | take the risk.

01:08:24.000 | I was always in personal choice.

01:08:25.000 | - Isn't this what Gretchen Whitmer in Michigan

01:08:28.000 | and Gavin Newsom in California subscribed to,

01:08:31.000 | the idea that the way to fight COVID was through lockdowns.

01:08:34.000 | Now, yes, Newsom had 10 pages of exceptions

01:08:37.000 | for his political donors.

01:08:38.000 | - Absolutely.

01:08:39.000 | - And he didn't use the police

01:08:40.000 | to lock people in apartment buildings.

01:08:41.000 | He may have wanted to, but they didn't actually do that.

01:08:44.000 | But can you really tell me that this lockdown policy

01:08:47.000 | has been disavowed by people like Fauci

01:08:50.000 | or by the health authorities

01:08:52.000 | like the Barbara Ferrer's of this world?

01:08:54.000 | They still subscribe to this view.

01:08:56.000 | - Do they really?

01:08:57.000 | Is anybody doing lockdown now?

01:08:58.000 | - Show me, well, they're not able to do it

01:09:00.000 | because no one agrees anymore.

01:09:02.000 | But tell me where anybody,

01:09:04.000 | tell me where any of the health experts

01:09:06.000 | who said that lockdowns were the correct response

01:09:08.000 | have repented and disavowed that view.

01:09:10.000 | - Yeah, I don't know.

01:09:13.000 | I haven't been tracking their mea copas.

01:09:15.000 | You yourself were in favor of lockdowns

01:09:18.000 | for a period of time.

01:09:19.000 | Yes, you were.

01:09:20.000 | You absolutely were.

01:09:21.000 | I'll pull the tape for the next episode.

01:09:22.000 | - No, I was in favor of a mask mandate.

01:09:23.000 | No, that's not true.

01:09:24.000 | I was in favor of a mask mandate.

01:09:26.000 | And I said the mask mandate

01:09:27.000 | was the alternative to lockdowns.

01:09:29.000 | - Okay.

01:09:30.000 | - I was saying that by May of 2020.

01:09:32.000 | - All right, listen, let's move on to the next thing.

01:09:34.000 | I don't represent, I'm an independent.

01:09:35.000 | I don't represent the Democratic Party.

01:09:37.000 | I don't represent Fauci.

01:09:39.000 | - No, you represent mainstream media.

01:09:41.000 | - I do not represent mainstream media.

01:09:43.000 | - You just said I was an old school journalist

01:09:45.000 | who's mortified with where it is today.

01:09:47.000 | - I'm giving you a hard time.

01:09:48.000 | I thought your explanation was fabulous.

01:09:51.000 | - Oh, thank you.

01:09:52.000 | - Obviously, Jake, I'm giving you a hard time too.

01:09:54.000 | I know that you were not a big lockdown proponent,

01:09:56.000 | but you understand the point I was making,

01:09:58.000 | which is there were a lot of proponents in the US.

01:10:00.000 | - Let's stop making me the Democratic spokesperson.

01:10:03.000 | - I just want to let you guys know

01:10:04.000 | I'm feeling a lot better after my beer.

01:10:06.000 | Let's talk about open AI.

01:10:08.000 | Come on, people.

01:10:09.000 | - All right, let's go.

01:10:10.000 | - Let's talk about open AI.

01:10:11.000 | - All right, open AI.

01:10:12.000 | All right, open AI is a company that builds

01:10:17.000 | artificial intelligence software and platforms.

01:10:21.000 | They have one platform called GPT.

01:10:24.000 | It is on its third version.

01:10:26.000 | As part of GPT-3, they created ChatGPT,

01:10:29.000 | which is a chat interface

01:10:31.000 | where you can ask questions to AI.

01:10:34.000 | The results are nothing short of stunning when they hit.

01:10:38.000 | Some of them are a little bit mixed,

01:10:40.000 | but Freeberg has spent the last 48 hours

01:10:44.000 | drinking white Russians with Oatly milk

01:10:47.000 | and playing with ChatGPT,

01:10:49.000 | including his question to the ChatGPT,

01:10:52.000 | which was write a script of Chamath, Sax, and J. Cal

01:10:56.000 | talking about the future of AI

01:10:58.000 | in the style of a Quentin Tarantino movie.

01:11:01.000 | And I have to say, it was pretty great, the result.

01:11:05.000 | Go ahead, Freeberg, tell us what you discovered.

01:11:08.000 | - I think you guys should read this real quick.

01:11:09.000 | We can cut it out if it's stupid, but okay.

01:11:11.000 | - Just actually read the script?

01:11:12.000 | - Yeah, read the script.

01:11:13.000 | - I don't think they got the Quentin Tarantino part right.

01:11:15.000 | I mean, I think it was impressive,

01:11:16.000 | but there's no F-bombs, there's no Tarantino-esque dialogue.

01:11:19.000 | - And by the way, they excluded all of that from open AI.

01:11:22.000 | They took out all cussing and violence from open AI.

01:11:25.000 | - That explains it, because if this was a Tarantino script,

01:11:28.000 | there would be some choice language in here.

01:11:29.000 | - Totally, but it's interesting.

01:11:30.000 | It's like the PG Tarantino script,

01:11:32.000 | with no cussing, no violence, no sex.

01:11:35.000 | - There's no pop culture references.

01:11:37.000 | - Yeah, you do the narrative.

01:11:38.000 | Go ahead.

01:11:39.000 | (gentle music)

01:11:43.000 | - The scene opens on Chamath Palihapitiya.

01:11:45.000 | David Sachs and Jason Calacanis

01:11:47.000 | sitting at a dimly lit table in a smoky bar.

01:11:49.000 | A bottle of whiskey sits in the center of the table, half empty.

01:11:53.000 | Chamath leaning forward, his eyes intense.

01:11:55.000 | - So, what do you guys think about this chat GPT thing?

01:11:59.000 | - David Sachs takes a sip of his drink.

01:12:01.000 | - It's impressive, no doubt about it.

01:12:03.000 | A large language model trained by open AI

01:12:05.000 | that can generate human-like texts

01:12:06.000 | in a variety of styles and formats.

01:12:08.000 | It's a game changer.

01:12:09.000 | - Jason nods.

01:12:10.000 | - Definitely.

01:12:11.000 | It's one of the most advanced AI systems we've seen to date.

01:12:14.000 | It's gonna have a huge impact on the way we use

01:12:17.000 | and interact with AI.

01:12:19.000 | - Chamath frowns.

01:12:21.000 | - That's what I'm worried about.

01:12:22.000 | With a technology like this,

01:12:23.000 | the possibilities are endless,

01:12:25.000 | but so are the potential dangers.

01:12:27.000 | - Sachs leans back in his chair.

01:12:29.000 | - Oh, come on, Chamath.

01:12:30.000 | You're not still stuck on that whole

01:12:32.000 | AI is gonna kill us all thing, are you?

01:12:34.000 | - Chamath leans forward again,

01:12:36.000 | his voice low and serious.

01:12:38.000 | - Look, I'm not saying that chat GPT

01:12:41.000 | is going to turn into Skynet overnight,

01:12:44.000 | but we need to be careful

01:12:46.000 | about the way we develop and deploy AI.

01:12:49.000 | We need to make sure that we're using it for good

01:12:51.000 | and not just for our own gain.

01:12:53.000 | (laughs)

01:12:54.000 | - J. Cal nods.

01:12:55.000 | - I agree.

01:12:56.000 | We need to make sure that we're responsible

01:12:58.000 | in the way we use AI,

01:12:59.000 | and that includes making sure it's accessible

01:13:02.000 | and affordable to everyone,

01:13:04.000 | not just a select few.

01:13:05.000 | - Chamath with a dirty smirk.

01:13:07.000 | - Ha ha ha ha.

01:13:08.000 | And let's not forget about the importance of regulation.

01:13:11.000 | Ha ha.

01:13:12.000 | We need to make sure that we have the right laws,

01:13:14.000 | ha ha, in place, ha ha.

01:13:15.000 | - Okay, okay, this is the most.

01:13:16.000 | - To prevent the misuse of AI.

01:13:17.000 | - I'm calling, the director's calling, cut.

01:13:19.000 | - Cut, that's it, scene.

01:13:21.000 | - Cut.

01:13:22.000 | - That's it, scene.

01:13:23.000 | We're not gonna finish it.

01:13:24.000 | - I mean, it's not that far off.

01:13:25.000 | I mean, it's 60% of the way there.

01:13:27.000 | - Actually, there isn't a better one.

01:13:29.000 | - All you have to do is put in a Biden for Sachs.

01:13:31.000 | If you blame Biden, it would have been perfect.

01:13:33.000 | - Let me tell you guys something stunning about this,

01:13:35.000 | this platform.

01:13:37.000 | So this is GPT 3.5,

01:13:39.000 | which is an interim model to the,

01:13:41.000 | what people are saying is the long awaited GPT 4.0 model,

01:13:45.000 | which I think they announced in 2020,

01:13:47.000 | and has been in development for some time.

01:13:50.000 | So the model, this GPT 3.5 model was trained in three steps.

01:13:54.000 | They do a great job explaining it on the OpenAI blog site,

01:13:59.000 | where they collect some data,

01:14:01.000 | and then there's a supervised model,

01:14:03.000 | meaning that there are humans that are involved in tagging.

01:14:06.000 | And then the model kind of learns from that system.

01:14:09.000 | Then you ask the model questions, you get output.

01:14:12.000 | And then humans rank the output.

01:14:14.000 | And so the model learns through that ranking system.

01:14:16.000 | And then there's kind of this third optimization thing,

01:14:18.000 | and then it's fine tuned.

01:14:20.000 | So the model itself has several steps

01:14:23.000 | of kind of human involvement,

01:14:24.000 | and it sources its own data and builds it.

01:14:28.000 | You know what's incredible about this model?

01:14:30.000 | The total size of the software package

01:14:33.000 | that runs the model is about 100 gigabytes.

01:14:35.000 | Isn't that amazing?

01:14:36.000 | Like you could fit this model on probably what,

01:14:39.000 | 20% of the storage space on your iPhone,

01:14:42.000 | and you could run this thing,

01:14:43.000 | and you could probably just talk to it

01:14:44.000 | for the rest of your life.

01:14:46.000 | And it's really kind of an incredible milestone.

01:14:49.000 | But I think what was so stunning to me about this,

01:14:52.000 | I know you guys are probably expecting something

01:14:54.000 | to be said like this,

01:14:55.000 | but you could see so many human knowledge worker roles

01:15:02.000 | and functions being replaced by this extraordinary interface.

01:15:06.000 | So kids can do homework, that's easy.

01:15:09.000 | Software engineers can get their code optimized

01:15:11.000 | and can get their code written for them.

01:15:13.000 | There's great examples of how software code

01:15:15.000 | has been written by this interface.

01:15:17.000 | You could see real estate insurance salespeople

01:15:21.000 | being replaced by some sort of software-like interface.

01:15:24.000 | Copywriters.

01:15:25.000 | Copywriters, you know, make me 100 versions

01:15:29.000 | of a commercial or an ad that I can then use.

01:15:32.000 | Customer support.

01:15:33.000 | Customer support completely replaced.

01:15:34.000 | If you guys remember,

01:15:35.000 | there were these automated customer support companies

01:15:37.000 | that started two decades ago.

01:15:40.000 | Never worked.

01:15:41.000 | And there was this great flurry.

01:15:42.000 | All BPO businesses were all about lower cost human labor.

01:15:44.000 | Now the cost of human labor goes to zero.

01:15:46.000 | My prediction, which is,

01:15:48.000 | so everyone's got the obvious prediction,

01:15:50.000 | which is there's going to be 100,000 startups

01:15:52.000 | that are going to emerge.

01:15:53.000 | I mean, this is kind of like this moment

01:15:55.000 | where the internet came along

01:15:56.000 | and everyone's like, this changes everything.

01:15:58.000 | I do think everyone thinks and feels that.

01:16:00.000 | So the obvious next step is a bubble will form.

01:16:03.000 | - Can I ask a technical question though, Freeberg?

01:16:06.000 | And then Chamath, you're probably thinking the same thing.

01:16:08.000 | - Let me just finish my market prediction

01:16:10.000 | and then we'll do the,

01:16:11.000 | but I think because everyone's so hyped about this

01:16:13.000 | and we all know this,

01:16:14.000 | all the VC attention, all the investor attention

01:16:16.000 | is shifting to this capability.

01:16:18.000 | And how do you apply this sort of capability

01:16:20.000 | across all of these different industries

01:16:22.000 | and all these different applications?

01:16:24.000 | And as a result, my guess is the next hype cycle,

01:16:26.000 | the next bubble cycle in Silicon Valley

01:16:28.000 | will absolutely be this generative AI business.

01:16:31.000 | - Okay, this is a little technical,

01:16:33.000 | but how would it know the difference

01:16:35.000 | between like, Y-O-U-R and U-R

01:16:39.000 | when it is processing natural language

01:16:42.000 | if you were to do like your anus or your anus?

01:16:45.000 | How would that, Freeberg,

01:16:47.000 | how would it know the difference

01:16:48.000 | between your anus and your space anus?

01:16:51.000 | - It'll learn that, you know.

01:16:53.000 | - It's a joke, it didn't land.

01:16:55.000 | It was a joke about your anus, it didn't land.

01:16:57.000 | - Try again, let's try again.

01:16:59.000 | - The AI probably would have made a better joke than that.

01:17:01.000 | - It would have made a better joke, for sure.

01:17:02.000 | - So, somebody did it in our group chat and said,

01:17:04.000 | "Do intros like J-Cal," and they were terrible.

01:17:06.000 | So at least they have a job for another year.

01:17:08.000 | - Avoyasi AI to pretend you're the all-in-pod besties

01:17:11.000 | telling Uranus jokes.

01:17:13.000 | - That would be pretty hilarious.

01:17:14.000 | - What would it come up with?

01:17:15.000 | - Sorry, let me just say one more thing

01:17:16.000 | about this OpenAI thing.

01:17:17.000 | I do think that the biggest and most interesting

01:17:20.000 | thing to think about is how this will

01:17:24.000 | disrupt the search box.

01:17:26.000 | The search, you know, the way search works at Google,

01:17:29.000 | you know, and the internet search,

01:17:31.000 | is there are these kind of servers,

01:17:33.000 | these web crawlers that go out and gather data.

01:17:35.000 | Some are structured data feeds,

01:17:37.000 | and some of them are just crawlers.

01:17:39.000 | And then that data is indexed, or in the structured way,

01:17:42.000 | it's kind of made available for serving

01:17:44.000 | directly on the search page.

01:17:47.000 | And so much of that is indexing.

01:17:50.000 | So I search for a bunch of keywords, those keywords,

01:17:52.000 | and perhaps there's some natural language context

01:17:54.000 | or match to a result page, and I click on that,

01:17:57.000 | and it's linked out.

01:17:58.000 | Years ago, Google started a product called the OneBox,

01:18:00.000 | where they could take structured data,

01:18:02.000 | like what is the weather in San Francisco today,

01:18:05.000 | and that top of the search result page

01:18:07.000 | just presented that data,

01:18:08.000 | 'cause it knows with high certainty

01:18:10.000 | the question you're asking,

01:18:11.000 | and it knows with high certainty the answer it can give you.

01:18:13.000 | - Yeah, clipped it from somebody's website, right.

01:18:15.000 | - So if that starts to become everything,

01:18:18.000 | then that OneBox interface,

01:18:20.000 | and it's not just Google's ability

01:18:22.000 | to access all this data and index it

01:18:24.000 | and serve it and store it,

01:18:26.000 | there could be a lot of competitors to the OneBox

01:18:29.000 | and a lot of competitors ultimately to search.

01:18:32.000 | And ultimately, Google's core product,

01:18:34.000 | their search engine,

01:18:36.000 | could be radically disrupted

01:18:39.000 | by an alternative set system or set of systems

01:18:42.000 | that have more of a natural language chat interface.

01:18:45.000 | - Which is literally why Google bought DeepMind,

01:18:50.000 | and there were a collection

01:18:51.000 | of human-powered search engines,

01:18:52.000 | Mahalo included, ChaCha, Answers.com,

01:18:55.000 | who were trying to do the human-based version of this.

01:18:57.000 | It just didn't scale.

01:18:58.000 | - We don't want to get ahead of ourselves,

01:18:59.000 | because one of the things we don't know

01:19:01.000 | is how much is going on in DeepMind.

01:19:03.000 | They're not very open like OpenAI is.

01:19:05.000 | They talk about some of the advanced frontier stuff

01:19:07.000 | like AlphaFold and so on,

01:19:09.000 | and they've been public about that,

01:19:10.000 | but a lot of that is really to generate interest

01:19:12.000 | and hype in what's next.

01:19:14.000 | But my understanding is DeepMind's been applied

01:19:16.000 | to everything from--

01:19:17.000 | - Ads.

01:19:18.000 | - Ad optimization,

01:19:20.000 | but also the ranking on YouTube videos

01:19:22.000 | to get people more engagement on YouTube,

01:19:24.000 | et cetera, et cetera.

01:19:25.000 | So there's all these ways that DeepMind's been applied

01:19:27.000 | within Google services--

01:19:28.000 | - That we don't see.

01:19:29.000 | - And certainly within search.

01:19:31.000 | But the question is,

01:19:32.000 | is there an entirely new interface for search

01:19:35.000 | that risks Google's core search business?

01:19:38.000 | And I think that there certainly will be

01:19:40.000 | a lot of money thrown at this,

01:19:41.000 | and if anyone has any interesting ideas,

01:19:43.000 | send me an email.

01:19:44.000 | - Saks and then Shamath.

01:19:45.000 | - Yeah, I think that's a really interesting point.

01:19:47.000 | I saw a thread on this

01:19:49.000 | where somebody was asking GPT

01:19:54.000 | a bunch of questions.

01:19:56.000 | They were generally coding questions,

01:19:58.000 | and they were actually comparing

01:19:59.000 | the result in Google versus GPT,

01:20:02.000 | and Google would just give you a reference

01:20:04.000 | to a link to some page,

01:20:06.000 | whereas GPT-3 would actually construct the answer,

01:20:09.000 | like a multi-paragraph answer

01:20:11.000 | that was far more detailed

01:20:13.000 | and in a way user-friendly.

01:20:15.000 | - Yeah.

01:20:16.000 | - Whereas the Google page

01:20:17.000 | would kick you over to a reference

01:20:19.000 | where it was like this one, two, three,

01:20:21.000 | sort of maybe someone had created a checklist,

01:20:23.000 | but it just wasn't that detailed.

01:20:24.000 | It really is pretty interesting.

01:20:26.000 | I thought Andresen tweeted

01:20:29.000 | a really interesting example as well

01:20:32.000 | where he asked GPT to create a scene from a play

01:20:38.000 | starring a New York Times journalist

01:20:40.000 | and a Silicon Valley tech entrepreneur.

01:20:41.000 | They were arguing about free speech,

01:20:43.000 | and each person asserts the view

01:20:45.000 | associated with his profession and social circle.

01:20:48.000 | We don't need to read the whole thing,

01:20:49.000 | but I thought this was spot on

01:20:51.000 | where I was actually like,

01:20:53.000 | both sides are making their best arguments,

01:20:55.000 | and it's like to each other

01:20:57.000 | in a conversation that seems intelligible.

01:20:59.000 | Like they're making their points

01:21:01.000 | at the right time in the conversation.

01:21:04.000 | It's like they're playing off each other.

01:21:06.000 | In other words,

01:21:07.000 | it actually reads like a conversation.

01:21:08.000 | I actually thought this one was more impressive

01:21:10.000 | than the one with the bestie impersonation

01:21:13.000 | because I actually thought that

01:21:14.000 | the one about all in

01:21:15.000 | didn't really capture our personalities per se,

01:21:19.000 | but this one actually does a pretty good job

01:21:21.000 | capturing the arguments in this debate.

01:21:24.000 | - Chamath?

01:21:25.000 | - Pretty impressive.

01:21:26.000 | - Any thoughts here?

01:21:27.000 | - Yeah, lots.

01:21:28.000 | I've spent a lot of time learning about this area.

01:21:31.000 | Six years ago,

01:21:32.000 | a team that I partnered with

01:21:35.000 | who was at Google that built TPU,

01:21:37.000 | we've been building silicon for this space,

01:21:39.000 | so we've been kind of going from the ground up

01:21:40.000 | for the last six years.

01:21:42.000 | A couple things that I'll say.

01:21:43.000 | The first is that I think

01:21:46.000 | we're going to replace SAS

01:21:48.000 | with what I call MAS,

01:21:50.000 | which is Models as a Service.

01:21:52.000 | And so, you know,

01:21:53.000 | a lot of what software will be,

01:21:55.000 | particularly in the enterprise,

01:21:56.000 | will get replaced with a single-use model

01:21:59.000 | that allows you to solve a function.

01:22:01.000 | So these chat examples are one,

01:22:03.000 | and you can name a bunch of SAS companies

01:22:05.000 | that were purveyors of SAS

01:22:07.000 | that'll get replaced by essentially GPT-3

01:22:10.000 | or some other language model.

01:22:12.000 | And then there'll be a whole bunch

01:22:14.000 | of other things like that.

01:22:15.000 | If it's a, you know,

01:22:16.000 | expense management company,

01:22:18.000 | they'll have a model

01:22:19.000 | that'll allow them to actually do

01:22:21.000 | expense management or blah, blah, blah,

01:22:23.000 | forecasting better.

01:22:25.000 | So I think SAS will get replaced over time

01:22:28.000 | with these models incrementally.

01:22:30.000 | That's phase one.

01:22:31.000 | But the problem with all of these models,

01:22:33.000 | in my opinion,

01:22:34.000 | is that they're still largely brittle.

01:22:36.000 | They are good at one thing.

01:22:39.000 | They are a single-mode way

01:22:42.000 | of interfacing with data.

01:22:44.000 | The next big leap,

01:22:46.000 | and I think it will come from

01:22:47.000 | one of the big tech companies

01:22:48.000 | or from OpenAI,

01:22:50.000 | is, and we talked about this,

01:22:51.000 | I talked about this a few episodes ago,

01:22:53.000 | is a multimodal model,

01:22:55.000 | which then allows you to actually

01:22:57.000 | bring together and join

01:22:59.000 | video voice data in a unique way

01:23:02.000 | to answer real substantive problems.

01:23:05.000 | So if I had to steel man

01:23:07.000 | the opposite side reaction,

01:23:08.000 | so I think there's a lot of people gushing

01:23:10.000 | over the novelty of GPT-3,

01:23:12.000 | if I had to, or chat GPT,

01:23:14.000 | if I had to steel man the opposite,

01:23:16.000 | what I would say is

01:23:17.000 | it's going to get somewhere

01:23:19.000 | between 95 to 99%

01:23:21.000 | of all of these very simple questions right

01:23:24.000 | because they're kind of cute and simple.

01:23:26.000 | There is no consequence of saying

01:23:28.000 | "Write a play" because there is no wrong answer.

01:23:30.000 | Right? You either kind of,

01:23:32.000 | it tickles your fancy or it doesn't,

01:23:34.000 | it kind of entertains you or it doesn't.

01:23:37.000 | When this stuff becomes very valuable

01:23:39.000 | is that when you really need a precise answer

01:23:42.000 | and you can guarantee that

01:23:44.000 | to be overwhelmingly right,

01:23:45.000 | that's the last 1% to 2%

01:23:47.000 | that is exceptionally hard.

01:23:49.000 | And I don't think that we're at a place yet

01:23:51.000 | where these models can do that.

01:23:53.000 | But when we get there,

01:23:55.000 | all of these models as a service

01:23:57.000 | will be very much commoditized.

01:24:00.000 | And I think the real value is finding

01:24:03.000 | non-obvious sources of data

01:24:06.000 | that feed it.

01:24:08.000 | So it's all about training.

01:24:09.000 | So meaning you can break down

01:24:11.000 | machine learning and AI into two simple things.

01:24:13.000 | There's training,

01:24:14.000 | which is what you do asynchronously,

01:24:16.000 | and then there's inference,

01:24:17.000 | which is what you're doing in real time.

01:24:19.000 | So when you're typing something into chat API

01:24:21.000 | or a chat GPT,

01:24:23.000 | that's an inference that's running

01:24:25.000 | and then you're generating an output.

01:24:26.000 | But the real key is where do you find

01:24:28.000 | proprietary sources of data

01:24:31.000 | that you can learn on top of?

01:24:33.000 | That's the real arms race.

01:24:35.000 | So one example would be,

01:24:37.000 | let's say you build a model

01:24:39.000 | to detect tumors.

01:24:41.000 | Right? There's a lot of people doing that.

01:24:43.000 | Well, the company that will win

01:24:45.000 | may be the company that actually

01:24:47.000 | then vertically integrates,

01:24:49.000 | buys a hospital system,

01:24:51.000 | and get access to patient data

01:24:53.000 | that is completely proprietary to them

01:24:55.000 | and covers the most number of women

01:24:57.000 | of all age groups and of all ethnic

01:24:59.000 | categories.

01:25:01.000 | Those are the kinds of moves

01:25:03.000 | in business that we will see in the next

01:25:05.000 | five to ten years that I find much more

01:25:07.000 | exciting and trying to figure out

01:25:09.000 | how to play in that space.

01:25:11.000 | But I do think that chat GPT is a wonderful

01:25:13.000 | example to point

01:25:15.000 | us in that direction. But I'm sort

01:25:17.000 | of more of that case, which is

01:25:19.000 | it's a cute toy, but we

01:25:21.000 | haven't yet cracked the 1 to 2% of

01:25:23.000 | use cases that makes it super useful.

01:25:25.000 | But I think the first step,

01:25:27.000 | but the first step will be the transformation

01:25:29.000 | of SAS to MAS.

01:25:31.000 | And then from there, we think we can try to figure

01:25:33.000 | this out. It reminds me of in a way when you

01:25:35.000 | give that description of like, "Hey, this is really

01:25:37.000 | interesting, but it's not complete."

01:25:39.000 | Is remember when GPS came out and

01:25:41.000 | people were doing turn-by-turn

01:25:43.000 | navigation, they'd drive off the road because they were

01:25:45.000 | trusting it too much. And then

01:25:47.000 | over 20 years of GPS, we're

01:25:49.000 | kind of like, "Yeah, it's pretty bulletproof, but keep your eyes on

01:25:51.000 | the road." Same thing that's happening.

01:25:53.000 | And these changes tend to be slow.

01:25:55.000 | You said it right. These last 100 or 200 basis

01:25:57.000 | points literally takes decades.

01:25:59.000 | Exactly. So the last 15% of self-driving

01:26:01.000 | is like the decade-long

01:26:03.000 | slog. That may take a century. 15% may

01:26:05.000 | take a century. But the last 2% will take a few

01:26:07.000 | decades. It's like the change happens very slowly and then

01:26:09.000 | it all happens at once. For people who don't know

01:26:11.000 | what a TPU is, that's a

01:26:13.000 | Tensor Processing Unit. This is Google's

01:26:15.000 | application-specific

01:26:17.000 | circuits, right? Custom silicon

01:26:19.000 | that they invented for TensorFlow at the time.

01:26:21.000 | Yeah, so if you want to try to... Although now the modality

01:26:23.000 | of AI, we've changed that as well. So now we're totally

01:26:25.000 | in the world of transformers. So we're not even using...

01:26:27.000 | You're not letting the tensors flow the way they used to.

01:26:31.000 | All right, there's been a slowdown in SAS.

01:26:33.000 | Sax, what is happening

01:26:35.000 | in the software-as-a-service world?

01:26:37.000 | This is a good update by Jamin

01:26:39.000 | Ball, who works for Altimeter, our friend Brad Gerstner.

01:26:41.000 | He does these really great updates on

01:26:43.000 | what's happening in the SAS world. The big thing

01:26:45.000 | this week is that Salesforce had

01:26:47.000 | its quarter, and I would consider Salesforce

01:26:49.000 | to be the bellwether for the whole

01:26:51.000 | SAS category. I think they're the largest

01:26:53.000 | pure SAS company. They were

01:26:55.000 | the first multi-tenant SAS

01:26:57.000 | company at scale.

01:26:59.000 | What they've shown is

01:27:01.000 | a huge slowdown. Basically,

01:27:03.000 | their net new ARR

01:27:05.000 | that they just added in the previous quarter

01:27:07.000 | dropped two-thirds compared

01:27:09.000 | to the previous quarter.

01:27:11.000 | But because their sales and marketing spend

01:27:13.000 | was the same as the previous quarter,

01:27:15.000 | it exploded their

01:27:17.000 | CAC payback, which means the amount of time

01:27:19.000 | it takes to pay back your

01:27:21.000 | customer acquisition costs for a given

01:27:23.000 | customer. So you see there, 155

01:27:25.000 | months it would

01:27:27.000 | take now to pay back

01:27:29.000 | the customer acquisition costs.

01:27:31.000 | That's over 10 years. That doesn't work.

01:27:33.000 | I think before this quarter

01:27:35.000 | it was more like

01:27:37.000 | two and a half years. That's something that you can

01:27:39.000 | afford. A company can't

01:27:41.000 | if you're spending 10 years of

01:27:43.000 | gross profit on a customer

01:27:45.000 | to acquire them. The business doesn't make sense.

01:27:47.000 | So now, I'm not saying any of this

01:27:49.000 | to pick on Salesforce. It's an exceptionally

01:27:51.000 | run company.

01:27:53.000 | One of the absolute best.

01:27:55.000 | Marc Benioff, fantastic

01:27:57.000 | CEO, founder, great human

01:27:59.000 | being. But I think the point here is that

01:28:01.000 | what you're seeing is

01:28:03.000 | the whole SaaS industry

01:28:05.000 | is really slowing down here.

01:28:07.000 | In the first half of the year, you saw

01:28:09.000 | SaaS valuations correct.

01:28:11.000 | Now we're actually seeing SaaS

01:28:13.000 | top line correct.

01:28:15.000 | There's an interesting question

01:28:17.000 | here. If your CAC payback goes from

01:28:19.000 | two and a half years to 10 years, you have to

01:28:21.000 | bring your CAC down. How do you do that?

01:28:23.000 | You can either reduce marketing

01:28:25.000 | or you can reduce sales.

01:28:27.000 | So in other words, you can reduce... You mean cut the sales

01:28:29.000 | team? You can either cut

01:28:31.000 | people and headcount from your own team

01:28:33.000 | or you can cut spending you do

01:28:35.000 | on advertising or

01:28:37.000 | events or money that you spend

01:28:39.000 | on other companies. Either way,

01:28:41.000 | there's going to be a big contraction

01:28:43.000 | in jobs, basically,

01:28:45.000 | around this industry. And I think

01:28:47.000 | that what that could do is cause

01:28:49.000 | a vicious cycle where

01:28:51.000 | we start seeing... Death spiral?

01:28:53.000 | I wouldn't say death spiral. I think this vicious cycle

01:28:55.000 | for the next year or so where

01:28:57.000 | seat contraction becomes the

01:28:59.000 | norm instead of seat expansion.

01:29:01.000 | So if you go back over the last 10

01:29:03.000 | years, a major tailwind

01:29:05.000 | at the backs of SaaS startups has

01:29:07.000 | been that every year

01:29:09.000 | you start with 120%, 130%,

01:29:11.000 | 150% of last year's

01:29:13.000 | revenue just from your existing customers.

01:29:15.000 | Why? Because they were hiring more

01:29:17.000 | and more people and they needed to buy more and more

01:29:19.000 | seats. But now headcount growth

01:29:21.000 | is frozen and in fact, companies

01:29:23.000 | are doing major layoffs. So the

01:29:25.000 | baseline for next year could be

01:29:27.000 | seat contraction. So instead of starting

01:29:29.000 | with 120% of last year's revenue,

01:29:31.000 | you might start with 80% or 90%

01:29:33.000 | because there's going to be so much churn.

01:29:35.000 | So I think that SaaS companies need

01:29:37.000 | to take this into account. This idea

01:29:39.000 | that growth is on autopilot,

01:29:41.000 | that could start to go in reverse. I don't

01:29:43.000 | think permanently, but I think for the next year

01:29:45.000 | or so. This is why I also tweeted

01:29:47.000 | 2X is the new 3X. If you can

01:29:49.000 | grow 2X year over year in this environment,

01:29:51.000 | that is as good as

01:29:53.000 | better than growing 3X last year.

01:29:55.000 | Clearly, it's better. Like a

01:29:57.000 | lot of companies that weren't that great could

01:29:59.000 | grow 3X last year because it was

01:30:01.000 | so times are so frothy. Everyone was

01:30:03.000 | buying everything. But now it is going to be

01:30:05.000 | really hard to even double year over

01:30:07.000 | year. Companies need to take that into account

01:30:09.000 | into their financial planning.

01:30:11.000 | You need to restrain your burn because

01:30:13.000 | a lot of the revenue that you predict is going to be there

01:30:15.000 | may not be there.

01:30:17.000 | All right. Thanks so much

01:30:19.000 | to the Secretary of SaaS.

01:30:21.000 | I think you got a board meeting. I do. I got to run.

01:30:23.000 | One of the interesting things I saw in terms of use cases

01:30:25.000 | is somebody used the

01:30:27.000 | chat GPT to describe

01:30:29.000 | rooms. Then they took the descriptions

01:30:31.000 | of those rooms and then they put them into like

01:30:33.000 | Dolly or Stable Diffusion, one of those

01:30:35.000 | and it created the visual. I'm curious if

01:30:37.000 | you think the self-driving

01:30:39.000 | APIs and machine

01:30:41.000 | learning that's going on.

01:30:43.000 | Then you got images, then you got chat.

01:30:45.000 | Maybe you have proteins going on with

01:30:47.000 | the AlphaFold stuff. When

01:30:49.000 | these things start talking to each other,

01:30:51.000 | is that going to be the emergent

01:30:53.000 | behavior that we see

01:30:55.000 | of general AI and that's how we'll interpret

01:30:57.000 | it in our world is these

01:30:59.000 | 100 different vertical

01:31:01.000 | AIs

01:31:03.000 | hitting some level of reasonableness

01:31:05.000 | to Chamath's point on data sets

01:31:07.000 | and then all of a sudden the self-driving AI

01:31:09.000 | is talking to the one that's looking at

01:31:11.000 | cancer and tumor diagnosis in the chat

01:31:13.000 | and the image ones and maybe Stable Diffusion,

01:31:15.000 | the protein AI

01:31:17.000 | and the one that's looking at cancer

01:31:19.000 | cells start talking to each other. Yeah,

01:31:21.000 | I'm not sure that's as

01:31:23.000 | likely as

01:31:25.000 | the... there's a lot of

01:31:27.000 | solutions that will emerge

01:31:29.000 | within verticals

01:31:31.000 | and I think you can distinguish them.

01:31:33.000 | So I kind of gave this example a few

01:31:35.000 | months ago.

01:31:37.000 | If you remember Kai's Power Tools

01:31:39.000 | was a plugin for Adobe Photoshop

01:31:41.000 | came out in 1993, I

01:31:43.000 | believe. Of course. And Kai's Power Tools

01:31:45.000 | completely transformed the

01:31:47.000 | potential of Adobe Photoshop. Because

01:31:49.000 | Photoshop had all the basic brushing and

01:31:51.000 | editing capabilities within it.

01:31:53.000 | Kai's Power Tools was statistical models

01:31:55.000 | that basically took the

01:31:57.000 | matrix of the pixels

01:31:59.000 | and created some

01:32:01.000 | evolution of them into some visual

01:32:03.000 | output like a blur. And so you

01:32:05.000 | could blur, motion blur something and you could change

01:32:07.000 | the parameters and now your photo

01:32:09.000 | looked like it was going through a motion blur.

01:32:11.000 | Ultimately, Photoshop bought and

01:32:13.000 | implemented those tools. But those

01:32:15.000 | were similar. They were statistical models

01:32:17.000 | that made some representation of the input

01:32:19.000 | which was the image and then created an

01:32:21.000 | output which was an adjusted image.

01:32:23.000 | I would argue that that is very similar

01:32:25.000 | although the models behind

01:32:27.000 | it are very different in

01:32:29.000 | terms of the contextual application

01:32:31.000 | of statistical

01:32:33.000 | models in software. And

01:32:35.000 | you could see stuff like, for example,

01:32:37.000 | a chatbot that replaces

01:32:39.000 | "Help me figure out

01:32:41.000 | whether my credit card charges are correct or not."

01:32:43.000 | Instead of having a customer service

01:32:45.000 | agent, an offshore customer service

01:32:47.000 | agent helping you resolve that.

01:32:49.000 | Or "Help me return my item."

01:32:51.000 | Or there are very specific

01:32:53.000 | kind of verticalized applications

01:32:55.000 | that can plug in

01:32:57.000 | that ultimately replace what was

01:32:59.000 | manual and human driven before.

01:33:01.000 | Because humans used to manually make the motion

01:33:03.000 | blur in Photoshop and then it was automated

01:33:05.000 | with these software packages. And I think you

01:33:07.000 | can kind of think about it in that same way that these

01:33:09.000 | are known knowns. They don't require

01:33:11.000 | necessarily human

01:33:13.000 | physical labor or some

01:33:15.000 | human responsiveness. That if 95%

01:33:17.000 | of the work can be handled, it will

01:33:19.000 | get handled by some verticalized

01:33:21.000 | solution. So I think the

01:33:23.000 | physical labor versus the non-physical labor is

01:33:25.000 | one way to think about the distinction. Meaning is there some

01:33:27.000 | change in the physical world?

01:33:29.000 | Driving is absolutely a change in the

01:33:31.000 | physical world. You have to move physically through

01:33:33.000 | space. So that one is a very

01:33:35.000 | distinct class. All the stuff that's like

01:33:37.000 | communication,

01:33:39.000 | imagery, static imagery, audio,

01:33:41.000 | and then visual, video,

01:33:43.000 | there's some stacking that

01:33:45.000 | happens there. And some of those will be kind

01:33:47.000 | of siloed and then some of them will merge

01:33:49.000 | and you'll have these kind of unique kind of combo models.

01:33:51.000 | And so look, as

01:33:53.000 | they start to work together, I think we'll see them

01:33:55.000 | completely rewrite

01:33:57.000 | some of these verticals like movie

01:33:59.000 | production or music production,

01:34:01.000 | right, or advertising,

01:34:03.000 | or we're seeing it now with

01:34:05.000 | with video and creative

01:34:07.000 | arts with

01:34:09.000 | some of the visual stuff on

01:34:11.000 | OpenAI. And to be honest, a lot of journalism, a lot

01:34:13.000 | of creative arts have become the wisdom of the

01:34:15.000 | crowds over the last two decades where

01:34:17.000 | artists were looking

01:34:19.000 | at the collective works of

01:34:21.000 | the internet, interpreting it,

01:34:23.000 | and then coming up with content, which is

01:34:25.000 | kind of what these AIs are doing. And

01:34:27.000 | then who legally owns

01:34:29.000 | the collective content is

01:34:31.000 | going to be a big question, Chamath. You talked about

01:34:33.000 | datasets, and Microsoft is being

01:34:35.000 | sued right now, and GitHub, because

01:34:37.000 | they used open source

01:34:39.000 | to create tools in AI

01:34:41.000 | to help augment

01:34:43.000 | programmers, like while they're programming, writing code,

01:34:45.000 | it gives them suggestions, and now the open source

01:34:47.000 | community is suing them for using their

01:34:49.000 | datasets. So what do you think about the legality

01:34:51.000 | of datasets, Chamath, and should

01:34:53.000 | they get some kind of protection

01:34:55.000 | if you make a GPT-3

01:34:57.000 | based on Quora or based on Wikipedia?

01:34:59.000 | Should you have to get approval

01:35:01.000 | to use that data?

01:35:03.000 | I think it's the exact opposite.

01:35:05.000 | It's the exact opposite. They say that

01:35:07.000 | this is actually your work.

01:35:09.000 | And I think that that's the right legal

01:35:11.000 | framework. But the answer to your other question

01:35:13.000 | is, this is why I think the hunt

01:35:15.000 | for proprietary data

01:35:17.000 | actually becomes the hunt that matters.

01:35:19.000 | All of this other stuff, I think,

01:35:21.000 | is a lot less important because I think you have

01:35:23.000 | to assume that all of these models will

01:35:25.000 | eventually just get commoditized.

01:35:27.000 | So there'll be a, you know, like you see

01:35:29.000 | like Jasper AI, and you see a bunch of these

01:35:31.000 | generative AI companies. It's really interesting.

01:35:33.000 | But the problem is, when you sit it

01:35:35.000 | on top of the same substrate, you'll have a

01:35:37.000 | convergence. Right?

01:35:39.000 | Everybody's chat model will eventually look and sound

01:35:41.000 | and feel like the same thing, unless

01:35:43.000 | you're giving it a few

01:35:45.000 | special ingredients that other

01:35:47.000 | people are not. And so it's the

01:35:49.000 | hunt for those ingredients that will make

01:35:51.000 | this next generation of

01:35:53.000 | models really valuable.

01:35:55.000 | So to give an example, you'd have Wikipedia,

01:35:57.000 | which is creative comments anybody can

01:35:59.000 | use, but Quora as a data set, not

01:36:01.000 | everybody can use that's owned by a company.

01:36:03.000 | So Quora would have an advantage. Take an

01:36:05.000 | extreme example. If Quora

01:36:07.000 | didn't allow themselves

01:36:09.000 | to be crawled,

01:36:11.000 | okay, which they don't.

01:36:13.000 | But then they

01:36:15.000 | developed their own language model, which used

01:36:17.000 | the best of the internet, so

01:36:19.000 | call it, you know, GPT

01:36:21.000 | and Quora,

01:36:23.000 | maybe they are slightly

01:36:25.000 | better in certain domains than others.

01:36:27.000 | The other extreme example is the one that I

01:36:29.000 | used in healthcare, which is, you know,

01:36:31.000 | if you have access to patient data

01:36:33.000 | that you will not license to anybody

01:36:35.000 | else, you know, it stands to

01:36:37.000 | reason that that model

01:36:39.000 | actually then has much

01:36:41.000 | better chances of highly

01:36:43.000 | effective clinical outcomes versus any

01:36:45.000 | other model. Apple Watch comes to mind,

01:36:47.000 | right? Apple has all that watch

01:36:49.000 | data. If they could pair that with epics,

01:36:51.000 | another data set.

01:36:53.000 | What could they do together?

01:36:55.000 | So this is going to be like, this is the new oil

01:36:57.000 | is going to be data. And by the way,

01:36:59.000 | to talk about Apple

01:37:01.000 | for a second, the smart thing is they've gotten

01:37:03.000 | so methodically, they've never

01:37:05.000 | touted the AI. You know, they introduce

01:37:07.000 | one or two distinguishing

01:37:09.000 | features every year, right?

01:37:11.000 | So like the ECG,

01:37:13.000 | which was introduced many, many years ago,

01:37:15.000 | has only gotten

01:37:17.000 | slightly more usable like five or six years

01:37:19.000 | later, but in the meantime, there's, you

01:37:21.000 | know, 10s of millions of watches

01:37:23.000 | collecting this kind of data. So to your

01:37:25.000 | point, it's it's using

01:37:27.000 | these devices as Trojan horses to

01:37:29.000 | collect training data. That is the

01:37:31.000 | oil. Uber and

01:37:33.000 | Tesla have all this data

01:37:35.000 | of the data being collected by,

01:37:37.000 | you know, the

01:37:39.000 | phones or the cameras in the

01:37:41.000 | cars. The other difference though, is that

01:37:43.000 | you have to be in a realm

01:37:45.000 | where you don't need regulators

01:37:47.000 | to go the last mile. So the

01:37:49.000 | problem with ADAS, I think, or

01:37:51.000 | Level 5 autonomy, is that

01:37:53.000 | eventually you get to a point where even if

01:37:55.000 | the model becomes quote unquote

01:37:57.000 | "perfect,"

01:37:59.000 | you still need regulatory approval. And

01:38:01.000 | what I'm saying is I think you have to focus on

01:38:03.000 | areas of the economy

01:38:05.000 | that are not subject to that or

01:38:07.000 | where the regulatory pathway is already defined.

01:38:09.000 | So for example, if you use that healthcare

01:38:11.000 | example, let's say that you had the largest corpus of

01:38:13.000 | breast cancer image data and you could actually build an

01:38:15.000 | AI that was a much better classifier

01:38:17.000 | of tumors versus other things,

01:38:19.000 | the FDA

01:38:21.000 | actually has a pathway to get that approved

01:38:23.000 | very quickly. The problem with, you know,

01:38:25.000 | Level 5 autonomy is that there is no clear

01:38:27.000 | pathway. It's not, again, we go back to almost

01:38:29.000 | a crypto example.

01:38:31.000 | We don't really know who will govern that decision

01:38:33.000 | and we don't know how that will be governed.

01:38:35.000 | So I think the

01:38:37.000 | thing that investors have to do

01:38:39.000 | and entrepreneurs, entrepreneurs have to pick their end

01:38:41.000 | market very carefully, and investors

01:38:43.000 | have to realize that this dynamic exists

01:38:45.000 | as well. If you're going to do this right, make money.

01:38:47.000 | Imagine the Robin Hood trading, you know,

01:38:49.000 | trader data set, watching

01:38:51.000 | people sell in shares and then predicting

01:38:53.000 | markets with it, with AI, I mean, it could be crazy.

01:38:55.000 | Well, you have that payment for order flow that's used

01:38:57.000 | by Citadel and the other big... But not AI.

01:38:59.000 | Right. Or who knows? Maybe they are using

01:39:01.000 | AI on their side. I don't know if they are. I can

01:39:03.000 | tell you as somebody who sells,

01:39:05.000 | we sell a lot

01:39:07.000 | of machine learning hardware into this market.

01:39:09.000 | The biggest buyers are

01:39:11.000 | the US government and

01:39:13.000 | these ultra high frequency trading organizations.

01:39:15.000 | Freeberg, any final thoughts here? I'll give you the final word.

01:39:17.000 | How could this affect

01:39:19.000 | astronomy? How could this affect

01:39:21.000 | our search of the galaxies,

01:39:23.000 | you know, going out past Pluto,

01:39:25.000 | Saturn, breaching Uranus,

01:39:27.000 | any of those things, how could it impact?

01:39:29.000 | Any...

01:39:33.000 | I'm trying to get a Uranus joke

01:39:35.000 | to land. Help me out there, Tramont.

01:39:37.000 | You got me. It's not that... I think

01:39:39.000 | you need to have more

01:39:41.000 | space related...

01:39:43.000 | Yeah, workshop this one with me.

01:39:45.000 | Or gut biome related,

01:39:47.000 | you know... Gut biome, yeah. So how would this affect

01:39:49.000 | super gut? Use the promo code TWIST.

01:39:51.000 | You have to trick Freeberg into thinking

01:39:53.000 | we're asking a serious question. Get him down

01:39:55.000 | the science path and then rug pull him.

01:39:57.000 | Now that's the right use of rug pull. That's a good proper rug pull.

01:39:59.000 | Exactly. Okay, let's do it. Here we go. Let's workshop. That's the rug pull.

01:40:01.000 | So tell us, you know, when you're doing like

01:40:03.000 | super gut, use promo code TWIST to get 25% off.

01:40:05.000 | When you're doing super gut,

01:40:07.000 | you're analyzing people's guts.

01:40:09.000 | How would you then

01:40:11.000 | have machine learning in this, you know,

01:40:13.000 | API, this chat API

01:40:15.000 | in GPT-3? How could that help

01:40:17.000 | with processing all of that,

01:40:19.000 | especially when it passes through Uranus?

01:40:21.000 | Freeberg. You okay up there, Freeberg?

01:40:25.000 | You are

01:40:27.000 | hung over. I'm hung over, but I

01:40:29.000 | also had like a 7am board

01:40:31.000 | meeting. So I'm also just a little

01:40:33.000 | beat up. Were you grumpy on the board meeting?

01:40:35.000 | Did you get a little cantankerous with your... No, no, it was fine.

01:40:37.000 | You were fine? You kept the rage in control?

01:40:39.000 | I think I had my caffeine

01:40:41.000 | fuel and then I kind of

01:40:43.000 | cranked down afterwards.

01:40:45.000 | Alright, everybody. We will see you next time

01:40:47.000 | for the Secretary of

01:40:49.000 | SAS, the Dictator,

01:40:51.000 | and the Sultan of Hungover.

01:40:53.000 | We will see you

01:40:55.000 | next time. Bye-bye.

01:40:57.000 | Love you guys. Bye-bye.

01:40:59.000 | [outro music]

01:41:01.000 | [outro music]

01:41:03.000 | [outro music]

01:41:05.000 | ♪ I'm going all in ♪

01:41:07.340 | And they said,

01:41:08.140 | "We open source it to the fans,

01:41:09.780 | and they've just gone crazy with it."

01:41:11.240 | Love you West, baby!

01:41:12.380 | I'm queen of Kinwa!

01:41:13.480 | ♪I'm going all in ♪

01:41:14.940 | ♪ What your winners like? ♪

01:41:16.120 | ♪ What-What are your winners like? ♪

01:41:18.180 | ♪ What your winners like? ♪

01:41:20.380 | Besties are gone!

01:41:21.480 | Go, 13!

01:41:22.820 | That is my dog taking a noise in your driveway.

01:41:25.020 | No, no, no, no, no!

01:41:26.520 | [Laughing]

01:41:28.060 | Oh, man!

01:41:29.000 | My avatars will meet me at "Lakeligt".

01:41:30.820 | We should all just get a room

01:41:32.000 | and just have one big huge orgy,

01:41:33.440 | cause they're all just useless.

01:41:34.860 | It's like this sexual tension

01:41:36.200 | that they just need to release somehow.

01:41:37.740 | ♪

01:41:38.800 | ♪ What your beat? ♪

01:41:40.540 | ♪ What your beat? ♪

01:41:42.440 | ♪ What your beat? ♪

01:41:44.040 | We need to get merch!

01:41:44.940 | Besties are back!

01:41:45.780 | ♪ I'm going all in ♪

01:41:47.440 | ♪

01:41:53.480 | ♪ I'm going all in ♪

01:41:55.480 | you

E106: SBF's media strategy, FTX culpability, ChatGPT, SaaS slowdown & more

Chapters