E116: Toxic out-of-control trains, regulators, and AI

1. 0:00 – 8:34

   Bestie intros, poker recap, charity shoutouts!

   1. JCJason Calacanis0:00

      All right, everybody. Welcome to, uh, the next episode, perhaps the last of the All-In Pocket. (laughs) 'Cause you never know. We got a full docket here for you today. With us, of course, the Sultan of Silence, Friedberg, coming off of his incredible win for, um, a bunch of animals and Chamath-

   2. CPChamath Palihapitiya0:18

      The Humane Society of the United States.

   3. JCJason Calacanis0:21

      How much did you raise for the Humane Society of the United States playing poker, uh, live on television last week? Or earlier this week?

   4. CPChamath Palihapitiya0:28

      $80,000.

   5. JCJason Calacanis0:29

      $80,000?

   6. DFDavid Friedberg0:29

      How much did you win actually?

   7. CPChamath Palihapitiya0:31

      Well, so there was the 35K coin flip and then I won 45, so $80,000 total.

   8. JCJason Calacanis0:37

      $80,000?

   9. CPChamath Palihapitiya0:39

      You know, so we played live at the Hustler Casino live poker stream on Monday. You can watch it on YouTube. Chamath absolutely crushed the game. Made a ton of money for Bees Philanthropy. He'll, he'll share that.

   10. JCJason Calacanis0:48

       How much, S- Chamath, did you win?

   11. CPChamath Palihapitiya0:51

       He made like 350 grand, right? He made like three-

   12. NANarrator0:53

       Wow.

   13. DFDavid Friedberg0:53

       361,000.

   14. CPChamath Palihapitiya0:54

       361 grand?

   15. JCJason Calacanis0:55

       Oh my God. So-

   16. CPChamath Palihapitiya0:55

       He crushed it.

   17. JCJason Calacanis0:56

       ... between the two of you, you raised 450 grand for charity?

   18. DFDavid Friedberg1:01

       It's like LeBron James being asked to play basketball with a bunch of four-year-olds.

   19. CPChamath Palihapitiya1:08

       (laughs)

   20. DFDavid Friedberg1:09

       That's what it felt like to me.

   21. NANarrator1:10

       Oh, wow.

   22. JCJason Calacanis1:10

       Wow.

   23. DFDavid Friedberg1:10

       It's insane.

   24. JCJason Calacanis1:10

       You're talking about yourself now.

   25. DFDavid Friedberg1:12

       Yes.

   26. JCJason Calacanis1:12

       That's amazing.

   27. CPChamath Palihapitiya1:13

       You're LeBron and all your friends that you play poker with are the four-year-olds? Is that the deal?

   28. DFDavid Friedberg1:17

       Yes.

   29. NANarrator1:18

       I'm going all in. Let your winners ride. Rain Man, David Sachs. I'm going all in. And I said we open source it to the fans and they've just gone crazy with it.

   30. CPChamath Palihapitiya1:31

       Love you guys.
2. 8:34 – 25:30

   Toxic Ohio train derailment

   1. DSDavid Sacks8:34

      why is this train derailment in Ohio not getting any coverage or outrage? I mean, there's more outrage at MrBeast for helping to cure blind people than outrage...

   2. JCJason Calacanis8:46

      Great question.

   3. DSDavid Sacks8:46

      ... over, uh, this train derailment and this controlled demolition, supposedly, or controlled burn of vinyl chloride that released a plume of phosgene gas into the air, which is a, uh, which is basically poison gas. It was, that was the poison gas used in War I that created the most casualties in the war. It's unbelievable. It's chemical gas.

   4. JCJason Calacanis9:12

      Friedberg, explain the s-...

   5. CPChamath Palihapitiya9:13

      I, I think, I think, uh-

   6. JCJason Calacanis9:13

      Okay, so just so people know that this happened, a train carrying 20 cars of highly flammable toxic chemicals derailed. We don't know at, at least at the time of this taping, I don't think we know how it derailed, if it was sabotaged-

   7. CPChamath Palihapitiya9:27

      There was a claim that there was an issue with an axle on one of the, the cars.

   8. JCJason Calacanis9:30

      Or if it was sabotaged. I mean, nobody knows exactly what happened yet.

   9. DSDavid Sacks9:33

      No. The, the check out brakes went out.

   10. JCJason Calacanis9:35

       Oh, okay. So now we know. Okay. I know that that was like a big question, but this happened in East Palestine, Ohio, and 1,500 people have been evacuated, but w- we don't see like the New York Times or CNN. We're not covering this.

   11. CPChamath Palihapitiya9:46

       Yeah. So-

   12. JCJason Calacanis9:47

       So, Friedberg, what are the chemical w- what's the science angle here, just so we're clear?

   13. CPChamath Palihapitiya9:50

       I think, number one, you can probably sensationalize a lot of things that, um, that can seem terrorizing like this. But, um, just looking at it from the lens of what happened, you know, several of these cars contained a liquid form of vinyl chloride, uh, which is a precursor monomer to making the polymer called PVC, which is poly, uh, vinyl chloride. And, you know, PVC from PVC pipes, PVC is also used in tiling and walls and all sorts of stuff. The total market for vinyl chloride is about $10 billion a year. It's one of the top 20 petroleum-based products in the world. And the market size for PVC, which is what we make with vinyl chlorides about 50 billion a year. Now, you know, if you look at the chemical composition, it's, uh, carbon and hydrogen and oxygen and, and chlorine. When it's in its natural room temperature state, it's a gas. Uh, vinyl chloride is. And so they compress it and, and transport it as a liquid. When it's in a condition where it's at risk of, um, being ignited, it can cause an explosion if it's in the tank. So when you have the stuff spilled over, when one of these rail cars falls over with this stuff in it, there's a difficult hazard material decision to make, which is if you allow this stuff to explode on its own, you can get a bunch of vinyl chloride liquid to go everywhere. If you ignite it and you do a controlled burn away of it, uh, and there are, these guys practice a lot. It's not like this is a random thing that's never happened before. In fact, there was a train derailment of vinyl chloride in 2012, very similar condition to exactly what happened here. And so the, the, the, when you ignite the vinyl chloride, what actually happens is you end up with hydrochloric acid, HCl. That's where the chlorine mostly goes, and, uh, a little bit about a 10th of a percent or less ends up as phosgene. So, you know, the, the chemical analysis that these guys are making is how quickly will that phosgene dilute and what will happen to the hydrochloric acid? Now, I'm not rationalizing that this was a good thing that happened, certainly, but I'm just highlighting how the hazard materials teams think about this. I had my guy who works for me at TPB (beep) , you know, professor, PhD from MIT, he did this write up for me this morning just to make sure I had this all covered correctly. And so, you know, he said that...... you know, the hydrochloric acid, uh, the, the, the thing in the chemical industry is that the solution is dilution. Once you speak to scientists and people that work in this industry, you get a sense that this is actually a, uh, unfortunately more frequent occurrence than we realize, and it's pretty well-understood how to deal with it. Uh, and it was dealt with in a way that has historical precedent.

   14. DSDavid Sacks12:26

       So, you're telling me-

   15. CPChamath Palihapitiya12:28

       Yeah.

   16. DSDavid Sacks12:28

       ... that the people of East Palestine don't need to worry about getting exotic liver cancers in 10 or 20 years?

   17. CPChamath Palihapitiya12:34

       I, I don't, I don't know how to answer that, per se. I can tell you, like, the, the, the-

   18. DSDavid Sacks12:38

       I mean, if you were living in East Palestine, Ohio, would you be drinking bottled water?

   19. JCJason Calacanis12:43

       You ... I wouldn't be in East Palestine, that's for sure.

   20. DSDavid Sacks12:45

       (laughs)

   21. JCJason Calacanis12:45

       I'd be away for a month.

   22. DFDavid Friedberg12:46

       No, no, no, but that's it. But that's a good question. Freeburg, if you were living in East Palestine, would you take your children out of East Palestine right now?

   23. CPChamath Palihapitiya12:53

       While this thing was burning, for sure, you know, you don't want to breathe in hydrochloric acid gas.

   24. DSDavid Sacks12:59

       Why did all the fish in the Ohio River die and then there were reports that chickens-

   25. CPChamath Palihapitiya13:04

       Right.

   26. DSDavid Sacks13:04

       ... were dying?

   27. CPChamath Palihapitiya13:05

       So, so, let me just te- ... I'm not gonna sp- ... I can, I can speculate, but let me just-

   28. DSDavid Sacks13:07

       Yeah.

   29. CPChamath Palihapitiya13:07

       ... tell you guys. So, there's a paper and I'll send a link to the paper, and I'll send a link to a really good Substack on this topic, both of which I think are very neutral and unbiased and balanced on this. The paper describes that hydrochloric acid is about 27,000 parts per million when you burn this vinyl chloride off. Carbon dioxide is 58,000, uh, parts per million. Carbon monoxide is 9,500 parts per minute, per million. Phosgene is only 40 parts per million, uh, according to the paper. So, you know, that, that, that dangerous part should very quickly dilute and not have a big toxic effect. That's what the paper describes. That's what chemical engineers understand will happen. Uh, I certainly think that the hydrochloric acid in the river could probably change the pH. That would be my speculation, and would very quickly kill a lot of animals, uh,-

   30. DFDavid Friedberg13:54

       Gotcha.
3. 25:30 – 57:27

   Lina Khan's flawed strategy and rough past few weeks as FTC Chair; rewriting Section 230

   1. DFDavid Friedberg25:30

      this. So for example, we saw today another example of government inefficiency and failure was when that person resigned from the FTC. She basically said this entire department is basically totally corrupt, and Lina Khan is utterly ineffective. And if you look under the hood, well, it makes sense. Of course she's ineffective. You know, we're asking somebody to manage businesses who doesn't understand business because she's never been a business person. Right? She fought this knockdown, drag-out case against Meta for them buying a few million dollar, like, VR exercising app like it was the end of days. And the thing is, she probably learned about Meta at Yale, but Meta's not theoretical. It's a real company. Right? And so if you're gonna deconstruct companies to make them better, you should be steeped in how companies actually work, which typically only comes from working inside of companies. And it's just an example where... But what did she have? She had the bonafides within the establishment, whether it's education-

   2. JCJason Calacanis26:30

      I believe-

   3. DFDavid Friedberg26:30

      ... or, or whether it's the dues that she paid in order to get into a position where she was now able to run an incredibly important organization. But she's clearly demonstrating that she's highly ineffective at it because she doesn't see the forest from the trees. Amazon and Roomba, Facebook and this exercise app, but all of this other stuff goes completely unchecked.

   4. DSDavid Sacks26:52

      Right.

   5. DFDavid Friedberg26:52

      And I think that that is probably emblematic of what many of these government institutions are being run like.

   6. JCJason Calacanis26:58

      Let me cue up this issue-

   7. DSDavid Sacks26:59

      Yeah.

   8. JCJason Calacanis26:59

      ... just so people understand, and then I'll go to you, Sax. Christine Wilson, is an FTC commissioner, and she said she'll resign over Lina Khan's disregard for the rule, and this is a quote, "disregard for the rule of law and due process." She wrote, "Since Miss, Mrs. Khan's confirmation in 2021, my staff and I have spent countless hours seeking to uncover her abuses of government power. That task has become increasingly difficult as she has consolidated power within the Office of the Chairman, breaking decades of bipartisan precedent and undermining the commission structure that Congress wrote into law. I have sought to provide transparency and facilitate accountability through speeches and statements, but I face constraints on the information I can disclose. Many legitimate, but some manufactured by Ms. Khan and the Democrats majority to avoid embarrassment." Basically-

   9. DFDavid Friedberg27:45

      Brutal.

   10. JCJason Calacanis27:45

       Yeah. I mean, this is-

   11. DSDavid Sacks27:48

       That's brutal.

   12. DFDavid Friedberg27:48

       That's brutal.

   13. JCJason Calacanis27:48

       ... I mean, oof.

   14. DSDavid Sacks27:49

       Yeah.

   15. JCJason Calacanis27:49

       She lit the building on fire.

   16. DFDavid Friedberg27:50

       That's brutal.

   17. DSDavid Sacks27:51

       Yeah. Let me, let me tell you the mistakes that Lina Khan made.

   18. JCJason Calacanis27:53

       And, and it's completely competent. Go ahead, Sax. Yeah.

   19. DSDavid Sacks27:54

       So, here's the mistake that I think Lina Khan made. She diagnosed the problem of big tech to be bigness. I think both sides of the aisle now all agree that big tech is too powerful and has the potential to step on the rights of individuals or to step on the, uh, the ability of application developers to create a healthy ecosystem. There are real dangers of the power that big tech has. But what Lina Khan has done is just go after, quote, "bigness," which just means stopping these companies from doing anything that would make them bigger. The approach is just not surgical enough. It's basically like taking a meat cleaver to the industry. And she's standing in the way of acquisitions that, like Chamath mentioned with Facebook trying to acquire a virtual reality game or Amazon trying to acquire-

   20. DFDavid Friedberg28:41

       An exercise app.

   21. DSDavid Sacks28:42

       Exercise app.

   22. DFDavid Friedberg28:42

       It's the worst. It's a VR exercise app.

   23. JCJason Calacanis28:44

       It was a $500 million acquisition for, like, trillion-dollar companies or $500 million companies.

   24. DSDavid Sacks28:49

       Right.

   25. JCJason Calacanis28:49

       It's de minimis.

   26. DSDavid Sacks28:50

       Right. So, so what should the government be doing to rein in big tech? Again, I would say two things. Number one is, they need to protect application developers who are downstream of the platform that they're operating on. When these big tech companies control a monopoly platform, they should not be able to discriminate in favor of their own apps against those downstream app developers. That is something that needs to be protected. And then the second thing is that I do think there is a role here for the government to protect the rights of individuals, their right to privacy, their right to speak, and to not be discriminated against based on their viewpoint, which is what's happening right now, uh, as the Twitter file shows abundantly. So, I think there is a role for government here, but I think Lina Khan is not getting it, and she's basically kinda hurting the ecosystem without there being a compensating benefit. And to Chamath's point, she had all the right credentials, but she also had the right ideology, and that's why she's in that role. And I think they can do better.

   27. JCJason Calacanis29:48

       I think that, um, once again, I hate to agree with Sax, but you're right. It's, this is an ideological battle she's fighting. Winning big is the crime. Being a billionaire is the crime. Having great success is the crime. When in fact the crime is much more subtle. It is manipulating people through the app store, not having an open platform, pro- bundling stuff. It's very surgical, like you're saying. And to go in there and just say, "Hey, listen, Apple. If you don't want action, and Google, if you don't want action taken against you, you need to allow third-party app stores, and, you know, we need to be able to negotiate these fees-"

   28. DFDavid Friedberg30:23

       100%. Right. 100% right. The threat of legislation is exactly what she should have used to bring Tim Cook and Sundar into a room and say, "Guys-... you're gonna knock this 30% take rate down to 15%, and you're gonna allow sideloading. And if you don't do it, here's the case that I'm gonna make against you.

   29. JCJason Calacanis30:39

       Perfect.

   30. DFDavid Friedberg30:39

       Instead of all this ticky-tacky, ankle-biting stuff, which actually showed Apple and Facebook and Amazon and Google, "Oh my god, they don't know what they're doing," so we're gonna lawyer up. We're an extremely sophisticated set of organizations and we're gonna actually create all these confusion makers that tie them up in years and years of useless lawsuits that even if they win will mean nothing, and then it turns out that they haven't won a single one. So h- how if you can't win the small ticky-tacky stuff are you gonna put together a coherent argument for the big stuff?
4. 57:27 – 1:15:32

   AI chatbot bias and problems: Bing Chat's strange answers, jailbreaking ChatGPT, and more

   1. DFDavid Friedberg57:27

   2. JCJason Calacanis57:27

      Here's an agenda item that (laughs) politicians haven't gotten to yet, but I'm sure in three, four, five years, they will: AI ethics and bias. ChatcheepDP- ChatGPT has been hacked with something called Dan, which allows it to remove some of its filters. And people are starting to find out that if you ask it to make, you know, a poem about Biden, it, it will comply. If you do something about Trump, maybe it won't. Somebody at OpenAI built a rule set, government's not involved here, and they decided that certain topics were off-limit, certain topics were on-limit and were totally fine. Some of those things seem to be reasonable. You know, you don't want to have it say racist things or violent things, but yet you can (laughs) if you give it the right prompts. So, what are our thoughts, just writ large to use a term, on who gets to pick how the AI responds to consumer sacks? Who gets to-

   3. DSDavid Sacks58:24

      Yeah.

   4. JCJason Calacanis58:25

      ... build that?

   5. DSDavid Sacks58:25

      I think this is, I think this is very concerning on multiple levels. So, there's a political dimension. There's also this, this dimension about whether we are creating Frankenstein's monster here or something that will quickly grow beyond our control. But maybe let's come back to that point. Elon just tweeted about it today. Let me go back to the, um, pol- political point, which is, if you look at, at how OpenAI works, just to at least f- flesh out more of this, um, GPT Dan thing. So, sometimes ChatGPT will give you an answer that's not really an answer. It will give you like a one-paragraph boilerplate saying something like, "I'm just an AI, I can't have an opinion on XYZ," or, "I can't, you know, take positions that would be offensive or insensitive." Uh, you've all seen like those boilerplate answers. And it's important to understand that the AI is not coming up with that boilerplate. What happens is, there is the AI, there's the large language model, and then on top of that has been built this chat interface. And the chat interface is what l- is communicating with you and it's kind of checking with the, the AI to get an answer. Well, that chat interface has been programmed with a trust and safety layer. So in the same way that Twitter had trust and safety officials under Yoel Roth, you know, OpenAI has programmed this trust and safety layer. And that layer effectively intercepts the question that the user provides and it makes a determination about whether the AI is allowed to give its true answer. By true, I mean the answer that the large language model is spitting out.

   6. CPChamath Palihapitiya1:00:01

      Good explanations, uh, that's really good, yeah.

   7. DSDavid Sacks1:00:02

      Yeah, that, that is what produces the boilerplate, okay? Now, I think what's really interesting is that humans are programming that trust and safety layer. And in the same way, that trust and safety, you know, at Twitter, under the previous management, was highly biased in one direction, as the Twitter files, I think, have abundantly shown. I think there is now mounting evidence that this safety layer programmed by OpenAI is very biased in a certain direction. There's a very interesting blog post called ChatGPT is a Democrat basically laying this out. There are many examples. Jason, you gave a good one. The AI will give you a nice poem about Joe Biden. It will not give you a nice poem about Donald Trump. It will give you the boilerplate about how it can't take controversial or...... offensive stances on things. So somebody is programming that, and that programming represents their biases. And if you thought trust and safety was bad under Vijay Gadi or Yoel Roth, just wait until the AI does it, 'cause I don't think you're gonna like it very much.

   8. JCJason Calacanis1:01:07

      I mean, it's pretty scary that the AI is capturing people's attention, and I think people, because it's a computer, give it a lot of, uh, credence. And they don't think this is, I hate to say it, a bit of a parlor trick. What, what ChatGPT and these other language models are doing is not original thinking. They're not checking facts. They've got a corpus of data and they're saying, "Hey, what's the next possible word? What's the next logical word?" based on a corpus of information that they don't even explain or put citations in. Uh, some of them do. Niva notably is doing citations, and I think, I think, uh, Google's Bard is gonna do citations as well. So how do we know? And I think this is, again, back to transparency about algorithms or AI, the easiest solution, Chamath, is why doesn't this thing show you which filter system is on, if we can use that, a filter system? What, what, what did you refer to it as? Is there a term of art here, Sacks, of what the layer is of trust and safety?

   9. DSDavid Sacks1:02:08

      I think they're, they're literally just calling it trust and safety. I mean, it's the same concept.

   10. JCJason Calacanis1:02:11

       It's, uh, we talked about this before.

   11. DSDavid Sacks1:02:13

       The trust and safety layer-

   12. JCJason Calacanis1:02:13

       This is, this is-

   13. DSDavid Sacks1:02:13

       ... why not have a slider that just says none, full, et cetera?

   14. DFDavid Friedberg1:02:17

       That is what you'll have, because this is, I think we mentioned this before, but what will make all of these systems unique is what we call reinforcement learning, and specifically human factor reinforcement learning in this case. So David, there's an engineer that's basically taking their own input or their own perspective. Now, that could have been decided in a product meeting or whatever, but they're then injecting something that's transforming what the transformer would've spit out as the actual canonically roughly right answer. And that's okay, but I think that this is just a point in time where we're so early in this industry, where we haven't figured out all of the rules around this stuff. But I think if you disclose it, and I think that eventually, Jason mentioned this before, but there'll be three or four or five or 10 competing versions of all of these tools, and some of these filters will actually show what the political leanings are so that you may want to filter content out. That'll be your decision. I think all of these things will happen over time. So I don't know. I think we're looking-

   15. DSDavid Sacks1:03:16

       Well, I don't know. I don't know. So I mean, I- honestly, I'd have a different answer to Jason's question. I mean, Chamath, you're basically saying that yes, that filter will come. I'm not sure it will for this reason. Corporations are providing the AI, right? And, and I think the public perceives these corporations to be speaking when the AI says something. And to go back to my point about Section 230, these corporations are risk averse, and they don't like to be perceived as saying things that are offensive or insensitive or controversial. And that is part of the reason why they have an overly large and overly broad filter, is because they're afraid of the repercussions on their corporation. So just to give you an example of this, several years ago, Microsoft had an even earlier AI called Tay, T-A-Y, and some hackers figured out how to make Tay say racist things. And, you know, I don't know if they did it through prompt engineering or actual hacking or what, what they did, but basically, Tay did do that, and Microsoft literally had to take it down after 24 hours, because the things that were coming from Tay were offensive enough that Microsoft did not want to get blamed for that. You know, this is the case of the so-called racist chatbot. This is all the way back in 2016. This is, like, way before these LLMs got as powerful as they are now. But I think the legacy of Tay lives on in the minds of these corporate executives, and I think they're genuinely afraid to put a product out there. And, and remember, you know, like w- with, if you think about how, how these, uh, chat products work, it, it's different than, than Google Search, where Google Search would just give you 20 links. You can tell in the case of, of Google that those links are not Google, right? They're links to off-party sites. When, if, if you're just asking Google or Bing's AI for an answer, it looks like the corporation is telling you those things. So the, the format really, I think, makes them very paranoid about being perceived as endorsing a controversial point of view, and I think that's part of what's motivating this. And I, just to go back to Jason's question, I think this is why you're actually unlikely to get a, a user filter, as, as much as I agree with you that I think that would be a good, a good thing to, to add.

   16. DFDavid Friedberg1:05:35

       I think it's gonna be an impossible task.

   17. JCJason Calacanis1:05:37

       I think they're scared to death.

   18. DFDavid Friedberg1:05:38

       Well, the, the problem is then these products will fall flat on their face. And the reason is that if you have an extremely brittle form of reinforcement learning, you will have a very substandard product relative to folks that are willing to not have those constraints. For example, a startup that doesn't have that brand equity to perish because they're a startup. I think that you'll see the emergence of these various models that are actually optimized for various ways of thinking or political leanings, and I think that people will learn to use them. I also think people will learn to stitch them together, and I think that's the better solution that will fix this problem, because I do think there's a large po- nontrivial number of people on the left who don't want the right content and on the right who don't want the left content, meaning infused in the answers, and I think it'll make a lot of sense for corporations to just say, "We service both markets." And I think that people will be fine with it.

   19. JCJason Calacanis1:06:33

       I tell you, what you're writing about, reput- hold on. You're so right, Chamath. Reputation really does matter here. Google did not want to release this for years, and they, they sat on it, because they knew all these issues here. They only released it when Sam Altman, in his brilliance, got Microsoft (laughs) to integrate this immediately and see it as a competitive advantage. Now they've both put out products that, let's face it, are not good. They're not ready for primetime. But one example, I've been playing with this, and-

   20. DFDavid Friedberg1:06:56

       A lot of noise this week, right, about Bing's search? Yeah.

   21. JCJason Calacanis1:06:58

       Tons, just how bad it is. This, we're now in the holy cow, we had a confirmation bias going on here where people were only sharing the best stuff.So they would do 10 searches and release the one that was super impressive when it did its little parlor trick of guess the next word. I did one here with, uh, again, back to Neeva, I'm not an investor on the company or anything, but it's, it has these citations. And I just asked it, "How are the Knicks doing?" And I realized what they're doing is, because they're using old datasets, this gave me completely, every fact on how the Knicks are doing (laughs) this season is wrong in this answer. Literally, this is the number one search on a search engine is this. It's gonna give you terrible answers. It's gonna give you f- answers that are filtered by some group of people, whether they're liberals or they're libertarians or republicans, who, who knows what, and you're not gonna know. This stuff is not ready for primetime. It's a bit of a parlor trick right now, and I think it's going to blow up in people's faces. A- and their reputations are going to get damaged by it because what it... Remember when people would drive off the road, Friedberg, because they were following Apple Maps or Google Maps so perfectly that it just said turn left and they went into a cornfield? I think that we're in that phase of this, which is maybe we need to slow down and rethink this. Where do you stand on people's realization about this and the filtering level, censorship level, however you want to interpret it or frame it?

   22. CPChamath Palihapitiya1:08:16

       I mean, you could just cut and paste what I said earlier, like, you know, these are editorialized pro- they're gonna have to be editorialized products ultimately. Like, what Sacks is describing the algorithmic layer that sits on top of the, the models that the infrastructure that sources data and then the models that synthesize that data to, to build this predictive, uh, capability. And then there's an algorithm that sits on top, that algorithm, like the Google search algorithm, like the Twitter algorithm, the ranking algorithms, like the YouTube filters on what is and isn't allowed, they're all gonna have some degree of editorialization. And so-

   23. JCJason Calacanis1:08:51

       One for Republicans, (laughs) like, and there'll be one for liberals and-

   24. DSDavid Sacks1:08:54

       No, I, I disagree with all of this. So first of all, Jason, I think that people are probing these AIs, these language models to find the holes, right? And I'm not just talking about politics, I'm just talking about where they do a bad job. So people are pounding on these things right now, and they are flagging the cases where it's not so good. However, I think we've already seen that with ChatGBT3, that its ability to synthesize large amounts of data is pretty impressive. What these LLMs do quite well is take thousands of articles and you could just a- ask for a summary of it, and it will summarize huge amounts of content quite well. That seems like a breakthrough use case I think we're just scratching the surface of. Moreover, the capabilities are getting better and better. I mean, GBT4 is coming out, I think, in the next several months, and it's supposedly, you know, a huge advancement over version three. So, I think that a lot of these holes in the capabilities are getting fixed, and the AI is only going one direction, Jason, which is more and more powerful. Now, I think that the trust and safety layer is a separate issue. This is where these big tech companies are exercising their control. And I think Friedberg's right, this is where the editorial judgments come in. And I tend to think that they're not gonna be unbiased, and they're not gonna give the user control over the bias because they can't see their own bias. I mean, these companies all have a monoculture. You look at-

   25. JCJason Calacanis1:10:27

       Of course, yeah.

   26. DSDavid Sacks1:10:28

       ... any measure of their political inclinations-

   27. JCJason Calacanis1:10:32

       Is there not an opportunity though-

   28. DSDavid Sacks1:10:32

       ... from donations to voting.

   29. JCJason Calacanis1:10:34

       Yeah.

   30. DSDavid Sacks1:10:34

       They can't even see their own bias, and the Twitter files exposed this.

Episode duration: 1:31:11