E122: Is AI the next great computing platform? ChatGPT vs. Google, containing AGI & RESTRICT Act

00:00:00.000 | Oh, J Cal's here. Hello, J Cal.

00:00:02.000 | Hey, how are you?

00:00:03.000 | Thanks for showing up.

00:00:04.000 | I've been here the whole time. I was just, I was just having some of these beautiful

00:00:09.600 | salted roasted pistachios. The only problem is when I went to the store, I kid you not,

00:00:14.440 | there was a shelf of these. All flavors available except one flavor.

00:00:18.600 | Salt and vinegar.

00:00:19.600 | Sea salt and vinegar.

00:00:20.600 | The entire-

00:00:21.600 | We moved the market. We moved the market.

00:00:23.480 | I am not kidding. I go to the fancy, you know, bespoke-

00:00:28.440 | The Raley's? You went to the Raley's?

00:00:29.440 | You went to the Raley's and Truckee?

00:00:30.800 | I went to the Raley's and Truckee, the artisanal, and they have, you know, all these overpriced-

00:00:35.280 | First of all, it's called artisanal?

00:00:37.200 | That's what I said. The art stuff. The artistic food. The artisanal row where they had this,

00:00:45.440 | I kid you not, spicy, salty, no salt, every shelf packed. Then there's one shelf I can

00:00:52.800 | see straight through to the ice cream.

00:00:54.360 | But not sea salt and vinegar.

00:00:55.440 | And I look at the tiny little sign, salt and vinegar, shelled nuts.

00:00:59.560 | Sea salt and vinegar.

00:01:00.560 | Sea salt and vinegar. Chamat's shelled nuts. Sold out across the country.

00:01:05.680 | You know, I cannot recommend these more highly. They're incredible.

00:01:08.560 | You can't recommend your salty nuts more?

00:01:11.160 | They are delicious. My salty nuts are delicious.

00:01:14.840 | Did you see Joe Manchin's high heater op-ed in the Wall Street Journal?

00:01:35.080 | Uh-oh.

00:01:35.560 | Oh my God.

00:01:36.520 | Yep. Joe Manchin went for it.

00:01:38.680 | But Joe Manchin's running for president.

00:01:40.600 | He is. I think. Okay, so let me ask Saks right there.

00:01:43.720 | Saks, Joe Manchin, Nikki Haley, and who's the guy from Florida?

00:01:48.600 | What's your question?

00:01:49.720 | By the way, there was a big dissection that was leaked this week. Ron Lauder

00:01:54.200 | flipped from Trump to DeSantis. That's a big one because Lauder's good for

00:01:57.560 | a lot of money. Five to 10 million at least.

00:02:00.360 | Joe Manchin, what impact would he have coming into the race? I'm not trolling you.

00:02:03.400 | Looking for your honest opinion.

00:02:04.520 | Well, it depends how he comes in. What did he say in the op-ed?

00:02:07.160 | He was talking about the insincerity of the Biden administration to control costs and how

00:02:10.600 | everybody was incompetent. And it's certainly there's some waste and we can control some

00:02:14.040 | spending and everybody needs to grow up and get in a room and just manage the budget for

00:02:18.360 | the American people and stop playing politics.

00:02:19.960 | Yeah, I think the headline of the article actually, to your point, Jake,

00:02:23.000 | was much worse than the substance of the article, Saks. But if you see the headline,

00:02:26.360 | I don't know, Nick, if you can just throw it up there, it was brutal.

00:02:29.000 | The headline and the byline of the article, I think, was more damaging than the substance

00:02:33.880 | of the article.

00:02:34.440 | Biden's Inflation Reduction Act betrayal. Instead of implementing the law as intended,

00:02:39.800 | his administration subverts it for ideological, ideological ends.

00:02:43.880 | I have to think that Joe was responsible for that, for the titling of that article.

00:02:46.920 | You don't, you know,

00:02:47.480 | He would get permission to approve it.

00:02:49.720 | Right.

00:02:50.220 | And by the way, I think if you guys remember, we talked about this when that act was first

00:02:55.320 | published. And if you guys remember, I think I pulled up the CBO data, the CBO model,

00:02:59.960 | and it showed for the first five years, this thing burns a couple hundred billion dollars.

00:03:04.680 | And then there's some expectation that there'll be some sudden boom in revenue

00:03:08.200 | in the out years, and then you make the money back in the out years.

00:03:11.160 | So it's total like accounting shenanigans for him to have made the claim in the first place

00:03:15.880 | that the IRA was actually going to be like a net deficit reduction or debt reduction.

00:03:20.040 | In fact, it's all just accounting shenanigans. And it's just a massive spend package,

00:03:23.720 | particularly in the near term when it matters most.

00:03:26.360 | I think I told you guys this, but I think this was like,

00:03:28.200 | when was the last time I was in Washington? Probably, what is it March now? So maybe it

00:03:33.000 | was January I was there. And I saw Schumer and Mark Warner, and I spent about two hours with

00:03:41.320 | Manchin. He is really impressive. He's cool. He's interesting. He's thoughtful. He's moderate.

00:03:49.320 | Manchin's like a formidable guy. So this will be really interesting if he steps in there and

00:03:55.800 | tries to take on the-

00:03:56.200 | Between Nikki Haley and Manchin, where do you write your check?

00:03:59.080 | I'd probably write a check to both, to be honest.

00:04:01.080 | Feels like a good ticket to me. I've always wanted to see the cross-

00:04:04.120 | Could you imagine a Democrat and Republican merging somehow and like running together?

00:04:08.200 | It would be the greatest.

00:04:08.920 | Oh my God.

00:04:10.520 | I've been pitching that for years. I think that's like a clear path.

00:04:13.800 | David Freeberg may have just come up with one of the most disruptive ideas in American politics

00:04:18.840 | that's ever been floated. Oh my God.

00:04:21.880 | Manchin Haley?

00:04:22.680 | Haley Manchin.

00:04:23.160 | Keep dreaming.

00:04:23.640 | Manchin Haley, yeah.

00:04:24.360 | Just my comment on this. So first of all, I remember when, you know, Manchin did a good

00:04:28.280 | job stopping Biden's $3.5 trillion build back better. Remember, it was him and Sinema that

00:04:32.680 | were the holdouts. But then Manchin compromised and gave Biden a $750 billion version of it.

00:04:39.400 | And I guess now he's complaining that Biden didn't live up to his end of the bargain in

00:04:43.480 | doing the deficit reduction. But quite frankly, many commentators said at the time that the

00:04:49.160 | bill's claims to deficit reduction were preposterous and that would never happen.

00:04:53.560 | So quite frankly, you know, Manchin shouldn't have been euchared or hoodwinked by Biden.

00:04:58.680 | Everyone was basically saying there'll never be any deficit reduction out of this bill.

00:05:01.720 | It's just more spending. So I don't really feel bad for Manchin here saying that somehow he was

00:05:07.160 | betrayed by Biden. He should have known better. Now, in terms of him running, yeah, I think as

00:05:12.840 | a Democrat who's figured out how to get himself elected in West Virginia, which is a plus 20

00:05:17.720 | red state, he obviously knows how to appeal to the center. The problem for him is just how do

00:05:22.280 | you get the Democratic Party nomination? Because he's far to the right of your average Democratic

00:05:27.720 | Party voter. If he wants to run as an independent, that's a different story. And that would really

00:05:32.440 | throw a curveball into the race. But I don't see him doing that. I think it's kind of a stretch.

00:05:37.480 | And this is the problem with a lot of these fantasy candidates is that, you know, centrist or

00:05:43.240 | moderate voters might like them, but they can't get the nomination of their party.

00:05:46.200 | And unless you mean like Trump and Obama, those who are fantasy candidates?

00:05:49.240 | I don't think so. I mean...

00:05:50.440 | Trump was not a fantasy candidate. He's the ultimate fantasy candidate.

00:05:52.360 | Well, he was an outsider, but he appealed to the base of the party. He appealed to the

00:05:55.960 | base of the party. What I'm saying is in order to get the nomination of a major party, you have

00:06:00.040 | to appeal to its base. And I don't think Manchin appeals to the base of the Democratic Party. He's

00:06:04.440 | out of step with it. He's out of step with it in ways that I like, don't get me wrong,

00:06:08.360 | but I just, I don't see how he's going to get a nomination.

00:06:11.720 | Chris Christie? What do you think of him? It seems like he's about to come in the race too, David.

00:06:15.240 | Pointless.

00:06:16.440 | He's viable?

00:06:17.240 | Pointless.

00:06:17.800 | Pointless.

00:06:18.520 | He's just clutter.

00:06:19.800 | Okay.

00:06:20.360 | Pointless. All right. Listen, everybody, welcome to the all-in podcast. It's like episode 100

00:06:25.560 | something with me again today. The Rain Man himself. Yeah. David Sacks is here. Friedberg

00:06:32.440 | is in his garden at his home in Paris. Spring has sprung. The Queen of Quinoa. And of course,

00:06:39.960 | the dictator himself, Chamath Palihapitiya, the Silver Fox. Look at that little tuft of silver

00:06:45.560 | hair. So distinguished.

00:06:47.240 | I got a haircut from somebody recently who said that people go to her and ask her to put the

00:06:53.320 | silver thing in their hair. Really?

00:06:56.600 | I don't have to worry about that.

00:06:58.840 | Yeah.

00:06:59.640 | Friedberg looks like he's in Smurf Village there. What is that background?

00:07:04.760 | That is a scene from [censored]

00:07:07.800 | Oh, okay.

00:07:08.680 | I like most of my backgrounds. I think it reflects the mood and the moment of the week.

00:07:13.000 | You guys just totally, totally denied half the beta mails in the YouTube comments from

00:07:18.040 | being able to guess what the background was. Thanks a lot, Sacks.

00:07:20.840 | You ruined it for them.

00:07:22.120 | Well, he'll break it up.

00:07:22.920 | Actually, I did a reverse image search and then I used a chat GPT plug-in to automatically figure

00:07:27.640 | out Friedberg's background each week.

00:07:29.720 | Oh, okay. All right. Well, let's get started. Come on.

00:07:33.080 | Let's get started. Okay. Listen.

00:07:34.280 | Look at Sacks. I got to get out of here.

00:07:37.080 | The alpha is spoken. The alpha is spoken.

00:07:38.760 | Sacks feels like he's in a good mood. I like this.

00:07:40.760 | Welcome to the world's greatest podcast.

00:07:43.000 | OpenAI launched a bunch of chat GPT plugins and I don't know if you saw it, but David Sacks

00:07:52.600 | wrote a blog post with chat GPT. It's an amazing back and forth. I read this back and forth.

00:07:58.680 | Explain what you did, Sacks. This was really one of the best conversations I've seen with chat GPT.

00:08:05.880 | It'll pop it up here on the screen, but explain what you did.

00:08:08.200 | Well, I had an idea for a blog post about the use of a, I guess, tactic you could call give to get.

00:08:16.360 | I thought it would be an interesting tactic for AI startups to use if they're trying to get a

00:08:21.640 | hold of proprietary training data. For example, if you want to create an architect AI, you need

00:08:27.560 | a lot of plans. Or if you're going to create a doctor AI, you need a lot of lab results or medical

00:08:34.120 | reports to train the AI on. Those are hard to get. OpenAI doesn't necessarily have them yet.

00:08:39.080 | There is an opportunity, I think, for startups to create these AIs in different,

00:08:43.560 | you'd call them professional verticals. The give to get technique would be you give points to your

00:08:50.200 | users for uploading that data and then they can spend those points by using the AI. Anyway,

00:08:55.560 | the company that came up with this give to get tactic was a company called Jigsaw almost 20

00:09:01.560 | years ago. No one remembers this company. I'm kind of dating myself because I remembered it.

00:09:05.400 | But I just had this idea, gee, I wonder if the Jigsaw approach could be used for AI startups.

00:09:09.800 | I started by going into chat GPT and I said, "Hey, have you heard of Jigsaw?" Then it had.

00:09:16.200 | Then I said, "Tell me about its give to get approach." Then I said, "Would this approach

00:09:21.880 | work for AI startups that want proprietary training data sets?" It said, "Yes, this is a good idea."

00:09:29.960 | Then I gave the architect example and I said, "Can you give me more examples like this?" It gave me

00:09:33.800 | like 20 more examples. Then I asked it just to flesh out various kinds of details. I went down

00:09:39.560 | some cul-de-sacs I didn't use. Then at the end, I said, "Can you summarize everything we've just

00:09:44.120 | talked about in a blog post?" It gave me the first draft of a blog post. I then did a substantial

00:09:49.560 | amount of editing on most of the blog posts, although some of it I just used verbatim.

00:09:53.960 | Then I had a couple of people in my firm look at it. They made some good suggestions. It's not like

00:09:58.120 | the human is completely out of the loop. Then I copied and pasted my edited version back into

00:10:02.920 | chat. GPT said, "Here's my edit." Then I asked for some suggestions. It made a few small edits.

00:10:09.080 | I said, "Okay, great. Just incorporate the edits yourself." Gave me that final output.

00:10:13.560 | Then I posted on Substack, a blog that probably would have taken me a week to research and write.

00:10:19.640 | If I had done it at all, I was able to do in a day. I can't see myself going back now. I think

00:10:25.080 | this is just how I'm going to write all my blog posts is use chat GPT as my researcher,

00:10:31.960 | as a writing partner, some cases an editor, but I'm definitely going to run it through.

00:10:38.360 | The thing that I was struck by was just how kind and generous and thoughtful this conversation was.

00:10:45.160 | I just thought, "I've never seen Sax have a conversation where he was so kind to the

00:10:48.520 | other person and thoughtful." Right about now, all your friends and family are like,

00:10:51.880 | "How do we get Sax to have this conversation with us?" You were super kind to the AI.

00:10:55.960 | Because it's not a person, it was a robot.

00:10:58.760 | Oh, no.

00:10:59.260 | Well, just in case it takes over the world, J-Cal, you can't be too careful. No, I think,

00:11:04.920 | listen, it's important to give the AI-

00:11:06.680 | Look at him, he's so kind. He's like, "Oh, perfect, thanks."

00:11:08.840 | Perfect, thanks. Exclamation mark.

00:11:10.920 | I've never once gotten a thanks from Sax.

00:11:12.840 | Well, show that example, actually.

00:11:15.160 | Scroll up and show that example. The AI actually gave me some information about Jigsaw's point

00:11:20.200 | system, again, the rewards that they used. And it was just in text. So I said down below, "Hey,

00:11:26.440 | can you spit that out as a table?" And it did instantly, perfectly.

00:11:29.960 | It's like a day's work, right? You would have to have an analyst or researcher do a day's work.

00:11:33.160 | It's incredible.

00:11:33.880 | Then I just screenshot of that and I made it an exhibit in my blog post.

00:11:36.840 | But you also said thank you.

00:11:38.040 | Well, yeah.

00:11:38.760 | And then it was delightful back to you. I mean, this is a-

00:11:41.400 | Yes, it was delightful back to me.

00:11:43.080 | This is a literal road to you being a kind human being. All the money that you've spent on therapy

00:11:50.120 | and just trying coaching to be nice to people, you're just nice naturally to people.

00:11:55.080 | No, J. Kell, Sax is in a good mood today. I don't know why you're instigating him.

00:11:59.160 | He's laughing. Come on, it's fun. You have to think it's funny,

00:12:01.640 | Sax saying thank you to the AI. Perfect.

00:12:04.440 | This is confirmatory of what we know. David wants to live in a set of highly

00:12:09.000 | transactional relationships, ideally with a machine,

00:12:12.200 | who can then eventually help him make him money.

00:12:17.320 | Can I ask you a question, though? Sincerely, Sax. What did you enjoy more,

00:12:21.000 | working with your team of humans on this or working with

00:12:24.680 | ChatGPT? Which one was more enjoyable for you, just personally?

00:12:29.320 | Well, I think they both were. I would say that the human contributions were essential.

00:12:34.760 | Okay. But what did you enjoy more?

00:12:38.120 | It's not about enjoyment. This is just a job to get done, but it definitely sped things up

00:12:44.120 | enormously. I personally find the hardest part of writing a blog is when you're staring at that

00:12:48.280 | blank sheet of paper and just having to spit out the first thousand words. It's just so time

00:12:54.360 | consuming to do that. But again, if you start with the first draft, even if it's not very good,

00:12:59.240 | then you can just edit it and it speeds things up a lot.

00:13:02.360 | Ideation. It's great for ideation.

00:13:03.960 | But the contributions of people in my firm were important. I actually trusted it. I know that you

00:13:09.960 | probably should fact check it in a way because it can hallucinate. But the things that we're

00:13:14.280 | saying made so much sense to me that I didn't think it was hallucinating.

00:13:17.160 | Well, this is a great moment to pivot into what OpenAI did with plugins. These came fast and

00:13:23.640 | furious this week. And a bunch of folks who had started verticalized ChatGPT-based projects,

00:13:30.840 | MVPs were like, "Oh, maybe my project MVP is now dead," because Instacart, OpenTable, Shopify,

00:13:38.040 | Slack, Zapier, and Zapier, obviously, like if then this, then that is a very wide ranging tool

00:13:44.440 | that allows you to connect APIs from a multitude of sources. And what this all lets you do at the

00:13:51.080 | end of the day is have ChatGPT ping one of these sources, just like an app might do or some custom

00:13:58.600 | software might do, ping the API and return data. So, hey, what tables are open on OpenTable?

00:14:07.640 | Maybe Shopify, find me things to buy in this category, etc. And so people have started building

00:14:14.280 | little scripts. We used to call these when magic leap was out, internet agents, and the concept of

00:14:22.600 | a software agent that's existed for a long time, actually, in computer science, I'm sure Freebird

00:14:28.200 | will give us some examples of that. But also ChatGPT can now use a browser. So that means you

00:14:33.000 | can get around the dated nature of the content in the corpus. Somebody did things like, "Hey,

00:14:39.720 | build me a meal plan, book me a reservation for Friday night in OpenTable, source other ingredients

00:14:44.600 | and buy it for Saturday night on Instacart and then use something like Wolfram Alpha to, you

00:14:50.360 | know, calculate the calories, etc." So when you saw this drop, Sax, what did you think in terms of

00:14:58.120 | the opportunity for startups and to build these intelligent agents, things that will do if that,

00:15:05.320 | if this, then that, or just background tasks over time, and you could actually leave them running?

00:15:11.400 | Yeah, I mean, I think this is the most important developer platform

00:15:15.160 | since the iPhone and the launch of iOS in the App Store, and I would argue maybe ever in our

00:15:22.360 | industry, certainly since the beginning of the internet. I think there was a question when

00:15:26.520 | ChatGPT launched on November 30th, and people started playing with in December, what exactly

00:15:31.720 | OpenAI's product strategy was going to be? Was this just like a proof of concept or a demo?

00:15:38.120 | And they even kind of called it like a demo. And initially, it looked like what their business

00:15:44.280 | model was going to be was providing an intelligence API that other websites, other applications could

00:15:51.080 | incorporate. And we saw some really cool demos like that Notion demo of other applications

00:15:56.200 | incorporating AI capabilities. And so initially, it looked like what OpenAI was going to be was

00:16:02.280 | more like Stripe, where in the same way that Stripe made payments functionality available

00:16:07.720 | very easily through a developer platform, they were going to make AI capabilities available

00:16:12.520 | through their developer platform. And then I think a funny thing happened on the way to this

00:16:17.320 | announcement, which is they became the fastest growing application of all time, talking about

00:16:21.560 | ChatGPT, over 100 million users in two months. Nobody else has ever done that before. I think

00:16:26.760 | it took the iPhone, you know, two years plus, Gmail, Google, those products all took, I think,

00:16:33.720 | well over a year. So this became the fastest growing site of all time. And I think with

00:16:39.480 | plugins, what they're indicating is that they will become a destination site. This is not just a

00:16:44.360 | developer platform, this is a destination site. And through plugins, they are now incorporating

00:16:49.640 | the ability to basically, you know, anything you could do through an application, you will now be

00:16:56.120 | able to do through a plugin, you'll just tell ChatGPT what you want done. If you say, hey,

00:17:01.400 | book me a plane ticket on this date, it will go into Kayaks plugin and do that. You say,

00:17:05.960 | book me a plane ticket and then an Airbnb.

00:17:07.640 | So the promise of Siri and Alexa realized because those were very rigid, they had no intelligence,

00:17:14.280 | right, Friedberg? If you wanted Siri to do something specific, like use Waze, or to go

00:17:20.680 | get you an open table, it needed to be pretty specific. And it didn't have any kind of natural

00:17:26.280 | language model behind it. So this is taking existing APIs and putting a natural language

00:17:31.160 | layer in front of it, which makes it, you know, perform a little more naturally. Is that what

00:17:36.760 | we're seeing here, Friedberg?

00:17:37.640 | I think it provides access to a corpus of data and a suite of services that are not well

00:17:45.400 | integrated into a search or chat interface anywhere today. So, you know, knowing what

00:17:52.760 | restaurants have what seats available is in a closed service, it's in a data warehouse,

00:17:57.480 | operated by OpenTable. And now what OpenTable can do is provide an API into that data

00:18:04.280 | via an interface and they can allow chat GPT to make a request to figure that data out

00:18:09.640 | to give a response to a user where they can ultimately benefit from transacting and allowing

00:18:16.840 | a service. This closes the loop between search and commerce in a way that Google cannot and

00:18:24.840 | does not do today. And I think that's what makes it very powerful. We've seen this

00:18:29.240 | attempted in a number of important ways in the last couple of years with Alexa

00:18:33.320 | and Apple Home and Google Home kind of integration via the chat services that they offer,

00:18:39.400 | you know, where you speak to the device. But the deep integration that's possible now,

00:18:44.200 | and the natural language way that you can go from the request all the way through to the

00:18:49.080 | transaction is what makes this so extremely powerful. And I think, you know, the points

00:18:54.760 | I made a few weeks ago, when we first talked about, you know, search, having so many searches

00:19:00.440 | that are done, where the human computer interface presents a table or presents a chart, or presents

00:19:07.160 | a shopping list in a matrix. That's what makes search such a defensible product, I think,

00:19:12.600 | could theoretically be completely obviated or destroyed with an interface like this,

00:19:17.240 | where you can write the ability for chat GPT or whatever the core centralized services to

00:19:23.560 | actually present results in a table in a matrix in an interface in a shopping list, and actually

00:19:29.000 | close the transaction loop. It's really disruptive to things like commerce providers,

00:19:34.200 | it's really disruptive. You know, some of these commerce platforms, it's really disruptive

00:19:38.040 | to a lot of different industries, but also introduces a lot of real opportunity to build

00:19:42.840 | on top of that capability and that functionality to rewrite and ultimately make things easier and

00:19:47.480 | better for consumers on the internet. What do you think Chamath you're looking at this,

00:19:50.920 | and it seems to be moving at a very fast pace. Over 100 million users, they put a business model

00:19:56.440 | on it already 20 bucks a month, they have a secondary business model of, hey, use the API

00:20:01.720 | and we'll charge you for usage. And then you layer on what Zapier and if this than that had

00:20:06.120 | already sort of established in the world, which is API's, but nobody ever really wanted to write

00:20:11.720 | scripts. So that seemed to be the blocker, you go into Zapier, if this and that, it's where 5%

00:20:16.840 | of the audience people want to customize stuff, people who want to tinker. But this seems to now

00:20:20.600 | with the chat GPT chat interface, open it up to a lot of people. So is this super significant? Or

00:20:25.960 | is this a commodity product that, you know, 10 people will have for sitting here next year on

00:20:30.680 | all in Episode 220? I think you are asking the exact right question. And you use the great term,

00:20:36.760 | like in poker, if there are three hearts on the board, and you have the ace of hearts,

00:20:43.640 | you have what's called the nut blocker, right? Which means that nobody else, even if anybody

00:20:50.360 | else has a flush, they never have the best flush. And if flush is the best hand, there's a lot of

00:20:55.480 | ways that you can manipulate the pot and eventually win the pot because you have that ace of hearts

00:20:59.320 | and nobody else has it. The concept of blocker, I think is very important to understand here,

00:21:06.360 | which is what are the real blockers for this capability to not be broadly available.

00:21:12.040 | So I think you have to segregate, you have the end user destination,

00:21:16.360 | you have the language model, and then you have the third party services. And so if you ask the

00:21:23.160 | question, what is the incentive of the third party service? Well, the shareholders of

00:21:27.720 | a travel site, right? They're not interested in doing an exclusive deal with any distribution

00:21:36.440 | endpoint, they want their services integrated as broadly as possible. Right? So I think the

00:21:42.040 | the answer for the service providers is just like they build an app for iOS and for Google. And,

00:21:48.520 | you know, if they could have justified it, they would have built an app for a gaming console,

00:21:52.600 | they can, they should, they would, they do. Right? So that's going to get commoditized

00:21:57.800 | and broadly available. I think on the LLM side, I think we've talked about this.

00:22:02.120 | Everybody's converging on each other. In fact, there was an interesting

00:22:06.440 | article that was released that said that there was a handful of Google engineers that quit,

00:22:12.760 | because apparently Bard was actually learning on top of

00:22:19.480 | chat GPT, which they felt was either legal or unethical or something, right? So,

00:22:25.720 | so the point is, I think we've talked about this for a while, but all of these models will

00:22:30.120 | converge in the absence of highly unique data, right? What I've been calling these white truffles.

00:22:35.720 | So if you can hoard white truffles, your model will be better. Otherwise, your model will be

00:22:40.920 | the same as everybody else's model. And then you have the distribution endpoints of which there

00:22:46.440 | are many whose economic incentives are very high, right? So Facebook doesn't want to just sit around

00:22:51.880 | and have all this traffic go to chat GPT, they want to be able to enable Instagram users and

00:22:56.760 | WhatsApp users and Facebook users to interact through messenger or what have you. Obviously,

00:23:01.640 | Google has a, you know, many hundreds of billions of reasons to defend their territory. So I think

00:23:08.040 | all of this to me just means that these are really important use cases. As an investor,

00:23:14.920 | I think it's important to just stay a little patient. Because it's not clear to me that

00:23:20.200 | there are any natural blockers. But I do think that David's right that it's demonstrating a

00:23:24.920 | use case that's important. But it's still so early, we are six weeks in. Yeah, I tell you,

00:23:32.200 | I think there's a couple of great blockers here, or there's going to be an M&A bonanza for Silicon

00:23:36.600 | Valley. If you look at certain data sets, Reddit, stack overflow for programming, and Cora, these

00:23:45.000 | things are going to be worth a fortune and to be able to buy those or get exclusive licenses to

00:23:49.880 | those if you're maybe Google barred, or if you're a chat GPT, that could be a major difference maker

00:23:54.840 | Twitter's data set, obviously. And then you look at certain tools like Zapier. And if this and that

00:24:00.040 | they've spent a decade building the sort of meta API, that would be an incredible blocker. I think

00:24:07.400 | this is going to be like a balkanization of so many oil sources. Zapier already did it for free,

00:24:11.800 | they did a plugin for free. Yeah, exactly. I was just going to say, I don't think these are not

00:24:15.720 | blockers. I don't think this is the ace of hearts on the flush board. I don't think so. I think that

00:24:19.880 | these things are really interesting assets. They are definitely truffly in nature. But they may not

00:24:26.520 | be the you know, 10 pound white truffle from Alba that we're looking for. Yeah, no, but on the M&A

00:24:31.240 | side, don't you think this would be like incredible? No, but the only reason I say that again,

00:24:34.920 | is it is just so early, like I in the text, I mentioned this to you guys, I remember,

00:24:39.320 | and sacks and I were in the middle of this. We were both right at the beginning of social

00:24:44.920 | networking, sack started genie, I was in the middle of aim. And all of a sudden, we saw

00:24:49.880 | read start social net. Then we saw Friendster get started. Then we saw myspace get started. And you

00:24:56.920 | have to remember, when you look back now 20 years later, the winner was the seventh company, which

00:25:02.200 | was Facebook, not the first, not the second, it was the seventh, which started two and a half years

00:25:07.480 | properly after the entire web point to a phenomenon started. Yeah, same with search,

00:25:12.440 | by the way, where Google was probably 20. Exactly. To the scene. Yeah, excite like us.

00:25:17.800 | If you want to be a real student of business history, I'll just say something that's more

00:25:21.400 | meta, which is, if there's something that I've learned on the heels of this SVB fiasco,

00:25:27.400 | is that there is an enormous amount of negative perception of Silicon Valley, and frankly,

00:25:34.120 | a lot of disdain for VCs and prognosticating technologists, right. And I think that

00:25:40.280 | So you mean this podcast?

00:25:42.760 | I think we have to be very careful. Yeah. And I do think that we are an example of that,

00:25:47.560 | because we are the bright, shiny object of the people that were successful. And the broad makeup

00:25:52.760 | of America thinks that we're not nearly as smart as we all think we are. And after all of this money

00:25:58.280 | that's been burned in crypto land, and NF T's and all of this web three nonsense, to yet again,

00:26:03.880 | whip up the next hype cycle, I think doesn't serve us well. So I do think there's something

00:26:10.680 | very important here. But I think if we want to maintain reputational capital through this cycle,

00:26:18.600 | because government will get involved much faster in this cycle, I think it's important to just

00:26:23.800 | be methodical and thoughtful, iterate experiment, but it's too early to call it,

00:26:28.440 | I guess is what I would say.

00:26:29.560 | Yeah, it's definitely too early to call it. But,

00:26:31.240 | Saks, you're saying explicitly, you think this is bigger than the internet itself,

00:26:36.040 | bigger than mobile as a platform shift.

00:26:38.760 | It's definitely top three. And I think it might be the biggest ever. I think, look,

00:26:42.200 | I think things could certainly play out the way that Jamath is saying. However,

00:26:45.720 | I actually think that open AI has demonstrated now with these platform features, that it has a lead,

00:26:53.720 | a substantial lead. And I actually think that lead is likely to grow in the next year. And let me

00:26:58.520 | tell you why I think it's got a couple of assets here that are hard to replicate. So number one,

00:27:03.640 | user attention, I think they've now got, I would guess hundreds of millions of users. And this

00:27:08.200 | thing is caught on like wildfire, must have been beyond their wildest dream. I think it even

00:27:13.160 | surprised them how much this has taken off. It's really captured the public's imagination and

00:27:18.120 | people are discovering new use cases for it every day. If you are sort of the number two or number

00:27:24.360 | three or the seventh large language model to basically get deployed behind a chatbot, I just

00:27:30.040 | don't think you're going to get that kind of distribution because the novelty factor will

00:27:33.080 | have worn off and people will have already kind of learned to use chat GPT. So number one is the

00:27:39.480 | hundreds of millions of eyeballs. Number two is with this developer platform, I think we should

00:27:43.560 | describe a couple of other features of it. One of the problems with chat GPT, if you've used it,

00:27:49.160 | is that the training data ends in 2021. And so you very rapidly for many questions, get to a

00:27:55.080 | stopping point where it says, like, I don't know the answer to that because I don't have any

00:27:59.560 | information about the last two years. Well, one of the plugins that OpenAI has introduced itself

00:28:04.920 | is called the browsing plugin. And it allows chat GPT to go search the internet and not just run

00:28:11.640 | internet searches, but to run an internet search as if it were a human. So you ask chat GPT a

00:28:18.360 | question and it goes to find, it runs a search and then it scours through the list of 20 links

00:28:24.200 | and it doesn't stop until it finds a good answer. And then it comes back to you with just the

00:28:27.960 | answer. So it actually saves you the time of clicking through all those loops and it'll give

00:28:31.560 | you the browsing history to show you what it did. That's mind blowing. They also have a thing called

00:28:36.120 | a retrieval API, which allows developers to share proprietary knowledge bases with chat GPT. So if

00:28:43.000 | you have a company knowledge base or some other kind of content, you can share with chat GPT so

00:28:48.760 | that chat GPT can be aware of that. And there are some privacy concerns, but the company has said

00:28:53.160 | they're going to sandbox that data and protect it. As an example, I'm planning on writing a book on

00:28:59.400 | SAS using chat GPT, and I'm going to put together all the previous articles and talks I've done as

00:29:04.280 | a database so I can then work with that in chat GPT. So you're going to have more and more developers

00:29:09.960 | sharing information with chat GPT. You're going to have chat GPT able to update its training based

00:29:18.440 | on sort of the last two years, being able to search the internet. And I think that as those

00:29:22.680 | hundreds of millions of users use the product, and as developers keep sharing more and more of

00:29:26.840 | these data sets, the AI is going to get smarter and smarter. And then what's going to happen is

00:29:31.320 | both consumers and developers are going to want to use or build on the smartest API.

00:29:36.600 | Yeah, see, this is where it feeds on itself. I mean, yeah, I think there might be a

00:29:40.440 | I agree with much of what you're saying. But I do think somebody like Facebook, when they release

00:29:45.640 | their language model, which they're about to, is not going to allow chat GPT to have any access to

00:29:50.920 | the Facebook corpus of data. And then LinkedIn will do the same, they'll block any access to

00:29:55.640 | chat GPT to their data. And so then you might say, you know what, I'm doing something related

00:30:00.120 | to business and business contacts, I need to use the LinkedIn one. And they're just going to block

00:30:03.880 | other people's usage of and tell you, hey, you have to come to our interface and have a pro

00:30:07.400 | account on LinkedIn. And this all becomes little islands of data. And so I'm not sure that you may

00:30:12.760 | be right to call us too early to have a definitive opinion. But I would say I have to believe plugins

00:30:16.600 | are going to be promiscuous. Yes, exactly. plugins are the refutation of your idea does

00:30:20.680 | not have an API, Twitter turned off their API, people who are smart, data sets, Cora doesn't

00:30:25.560 | let people use its data. So I just picked three. Those are three incredible data sets that don't

00:30:30.840 | allow people and Craigslist doesn't. So people who are smart, do not allow API's into their data,

00:30:35.560 | they keep it for themselves. I think there were a lot of people when the app store rolled out

00:30:41.560 | that swore up and down, they'd never build a mobile app, because they didn't want to give

00:30:44.360 | Apple that kind of power that the internet was open, whereas the app store is closed and curated

00:30:49.080 | by Apple. And sure enough, they all at the end of the day had to roll out apps, even though in

00:30:53.640 | the case of Facebook, it definitely has made them vulnerable, because they're downstream of Apple.

00:30:58.200 | I mean, Apple now has enormous influence over Facebook's advertising revenue, because

00:31:03.400 | users have to go through Apple, they never had to do that before the internet. Nonetheless,

00:31:07.640 | Facebook felt compelled to release a mobile app, because they knew was existential for them,

00:31:11.880 | if they did it. And I believe that was happening. I don't think it's right analogy, the right

00:31:15.880 | analogy would be Google search, does Facebook does Craigslist allow their data to be indexed

00:31:21.400 | inside of Google search answers? No, right? They block that for a reason they and they will write

00:31:25.720 | a cease and desist letter. Fine. So so you know what, those guys will stay out of it. But look

00:31:29.320 | how much content Google search already has. And I think that chat GPT will start by eating a

00:31:35.160 | substantial portion of search, because again, you don't have to go through the 20 links,

00:31:38.360 | it just gives you the answer. It's going to eat a substantial portion of browser usage and app

00:31:43.800 | usage, because you're just going to tell chat GPT what you want to do. It will go book your plane

00:31:47.880 | ticket, it will go book your hotel room. Yeah, see, this is a play in this hold on the apps that

00:31:53.960 | want to play in this will benefit. So there'll be a powerful incentive for applications to get an

00:31:58.840 | advantage by participating. Let me finish my point. Yeah. And then eventually, they will be forced to

00:32:04.840 | do it not because they get an advantage, but because they're so competitively disadvantaged,

00:32:08.360 | if they don't participate in that ecosystem, I agree that they'll participate in it. But here's

00:32:12.760 | the thing, what's going to happen is Google is going to turn on bard and I've been playing with

00:32:16.200 | bard, it is 80% of chat GPT already. And then when they make bard a default, you know, little snippet

00:32:24.520 | on your Google search return page, or bard is built into YouTube, or Chrome, or Android,

00:32:29.240 | all the Play Store, they're going to roll right over chat GPT because they have billions of users

00:32:33.640 | already. So this advantage that you see today, I see that getting rolled real quick, because you'll

00:32:39.480 | be on YouTube. And on the top right hand side will be barred. And when you do a search, it's going to

00:32:43.400 | say, here are other sentences you could do, oh, you want to search Mr. Beast, when he's helped

00:32:47.800 | people or Mr. Beast when he's given away more money, or people have copied and been inspired

00:32:51.560 | by Mr. Beast, all that's going to occur inside of YouTube, and chat GPT is not going to have access

00:32:55.560 | to the YouTube corpus of data. And then when you do a search, it's going to be the same thing,

00:32:58.760 | it's going to be on the right hand side. And it's going to be playing just like it is in Bing,

00:33:01.960 | if you turn on your Android phone, they're going to make Google Assistant go right into bard.

00:33:06.600 | And Google Assistant is already used by hundreds of millions of people. So I think that Google

00:33:11.320 | will roll. I think they're going to roll chat GPT. I don't know who's going to win. But I'm looking

00:33:17.160 | at this saxapoo more reductively as a capitalist, which is what are people's incentives, because

00:33:23.320 | that's what they'll do. Google's incentive is to usurp chat GPT usage by inserting something

00:33:33.560 | inside of their existing distribution channels to suppress the ability for you to want to go

00:33:39.160 | to the app known as bundling. I think Facebook has that same incentive. Oddly, even though

00:33:45.720 | Microsoft is such a deep partner, I think certain assets of Microsoft have that incentive, you're

00:33:49.960 | talking collectively about five or $6 trillion of market cap, then when you add in Alexa, and

00:33:55.080 | Amazon and Siri and Apple, what is their incentive, I don't think their incentive is to let this

00:34:02.920 | happen. And I think if you look at the slack, Microsoft Teams example of even a better engineer

00:34:08.200 | product was excellent and widely deployed, even at hundreds of millions of users doesn't much

00:34:13.480 | matter when it's more cleverly distributed and priced. And so those things again, you may still

00:34:22.040 | be right, all I'm saying is, it's just so early to know. And as slow and lumbering as some of

00:34:28.840 | these big companies are, they are not so stupid as to kill their own golden goose and or defend it

00:34:35.000 | when threatened. So I think you just have to let let it see what happens. I want to finish the

00:34:38.680 | point on Google. And then we can move on to the bundling thing. Let me just make the counter

00:34:41.720 | argument, which is that I think Google is caught completely flat footed here, even though they

00:34:46.600 | shouldn't have been because they published the original paper on transformers in 2017. They

00:34:51.560 | should have seen where all this was going, but they didn't open AI, use that paper and commercialized

00:34:57.000 | it. And the proof of that is there was just a lawsuit a couple of days ago, or at least a claim

00:35:02.680 | by a former employee of Google who quit, because he said that they were using chat GPT to train

00:35:09.720 | their AI. So their AI is so far behind. They were violating the terms of use, hold on,

00:35:17.160 | they were violating the terms of use of open AI to train their own AI on chat GPT. That's not a

00:35:23.960 | good sign. That's not a good sign. I also think hold on, hold on, I'm just making the counter

00:35:29.640 | argument here. I mean, don't dismiss it out of hand, give me a chance to explain it. Moreover,

00:35:34.280 | chat GPT for which was just released a few weeks ago, we know that open AI had that they were using

00:35:41.640 | it internally for seven months. So the state of the art is not what we're using. It's what

00:35:47.240 | open AI has internally, they're obviously working now on chat GPT five. And so if you're saying

00:35:53.640 | that Bard is 80% of chat GPT four, well, I got news for you, it's probably 50% or 20% of chat

00:36:01.800 | GPT five. And who knows what the product roadmap is inside of open AI, I am sure that they've got

00:36:08.680 | 200 ideas for things they could do to make it better and low hanging fruit. But look, regardless,

00:36:14.280 | I think the pace of innovation here in development is going to speed up massively. I mean, there is

00:36:19.160 | going to be a flurry of activity. I agree, it's hard to know exactly how it's going to play out.

00:36:24.040 | But I think this idea that oh, it's a foregone conclusion, these big companies are just going

00:36:28.200 | to catch up with open AI, I think that there's a strong counter argument. That's not the case.

00:36:32.520 | I'm making a very specific argument. It's not a foregone conclusion where all the value will

00:36:36.360 | get captured. Just like in any of these major tidal waves. If you make the bets too early,

00:36:43.160 | you typically don't make all the money. And it tends to be the case. And it has been in the past,

00:36:48.520 | at least with these transformative moves. It's sort of in the early third of the cycle,

00:36:53.880 | is where the real opportunities to make the tons of money emerge. And there's a lot of folks that

00:36:59.080 | show you a path and then just don't necessarily capture the value. I'm not saying that that's

00:37:03.160 | going to be the case here. All I'm saying is, if history is a guide, all of these other big waves

00:37:09.000 | have shown that fact pattern. And so I'm very excited and I'm paying attention. But I'm just

00:37:14.760 | being circumspect with this idea that you know, having been in the middle of these couple of

00:37:18.920 | waves before it, I made all the money by waiting a couple years. I don't know if that's going to

00:37:22.760 | be true this time around. But right, that's sort of my posture right now.

00:37:25.720 | So you look, you obviously have a point because we're only four months in. So how can we know

00:37:28.680 | where this is going to be in five years. So you could be right to your point, Saks. I think it's

00:37:33.000 | clear. And this is, you know, big ups to the open AI team, that they will be one of the top two or

00:37:38.520 | three players. Absolutely. We all agree on that, which is extraordinary in itself. And the top four

00:37:43.720 | players freeberg are obviously going to be Microsoft OpenAI, we'll call that like, whatever

00:37:49.720 | that little, you know, pairing, and then Google, Facebook. And then we haven't talked about Apple,

00:37:56.120 | but obviously, Apple is not going to take this sitting down. And hopefully, they'll get in gear

00:38:00.760 | and have Siri, you know, make it to the next level, or they'll just put her out to pasture.

00:38:05.000 | If you were to look at those four, and we're sitting here a year from now, who has the best

00:38:11.080 | product offering? Who has the biggest user base? Just take a minute to think about that. Because

00:38:15.880 | you were at Google. And we all know, the word on the street is, it's the return of the kings.

00:38:22.360 | Larry and Sergey are super engaged by all reports, every back channel, everybody I talked to was

00:38:27.480 | saying that their every day, they're obsessed with Google's legacy now and making this happen.

00:38:32.680 | So what can you tell us in terms of who you think a year or two from now, we'll have the biggest

00:38:38.600 | user base and be the most innovative amongst that quartet, or maybe you think there's other players

00:38:44.440 | who will emerge? The advantage that OpenAI has, which is the advantage that any call it emerging,

00:38:53.320 | you know, advantage competitor has, outside is, yeah, outsider is that the incumbents are

00:39:01.080 | handicapped by their current scale. Much of the consideration set that Google has had in deciding

00:39:07.560 | what features and tools to launch with respect to AI over the last couple of years, has been driven

00:39:12.440 | fundamentally by a concern about public policy and public reaction. And I know this from speaking to

00:39:18.280 | folks there that are close enough to kind of indicate like, Google has been so targeted has

00:39:26.120 | been such the point of attack by governments around the world with respect to their scale and

00:39:31.400 | monopoly and monopolistic kind of behavior. Some people have framed it privacy concerns, you know,

00:39:39.240 | etc, etc. The fines in the EU are extraordinary, that so much of what goes on at Google today,

00:39:45.160 | is can I get approval to do this? And so many people have felt so frustrated that they can't

00:39:49.960 | actually unleash the toolkit that Google has built. And so they've been harnessed and focused

00:39:54.680 | on these internal capabilities. I think I mentioned this in the past, but things like,

00:39:58.440 | what's the right video to show on YouTube to keep people engaged? What's the right ad to show to

00:40:04.040 | increase click through rates, etc, etc. versus building great consumer products for fear of the

00:40:09.080 | backlash that would arise, and governments coming down on them and ultimately, and ultimately

00:40:13.560 | attacking the revenue and the core revenue stream. And this is no different than any other kind of

00:40:19.080 | innovative dilemma. You know, any other business of scale in any other industry historically

00:40:23.400 | ultimately gets disrupted, because their job at that point is to protect their cash flow and their

00:40:28.360 | revenue stream and their balance and their assets, not to disrupt themselves, especially as a public

00:40:34.040 | company, especially under the scrutiny and the watchful eye of governments and regulators.

00:40:38.200 | So I think Google has, in aggregate, probably good competitive talent, if not better talent

00:40:44.840 | than open AI and others. Google has arguably the best corpus of data upon which to train

00:40:51.080 | the best capabilities, the best toolkit, the best hardware issues, the lowest cost

00:40:55.400 | for running these sorts of models, the lowest cost for serving them, etc, etc. So frankly,

00:41:00.360 | they're way behind the battle is theirs to lose, if they are willing to disrupt themselves. And

00:41:05.800 | this is the moment that Larry and Sergey should wield those founders shares that they have.

00:41:10.360 | And they should wield the comments that they wrote in that founders letter, that they will always

00:41:14.040 | make the right decision for the long term for this company, even if it means taking a cost in

00:41:18.200 | the short term and disrupting themselves. This is the moment to prove that those founders shares

00:41:23.320 | were worth, you know, the negotiation to get there. And, and I think that it is going to require

00:41:29.080 | a real degree of scrutiny, a real degree of regulatory uncertainty, a real degree of

00:41:34.040 | challenge by governments, and public policy people, and perhaps even a revenue hit in the

00:41:38.520 | near term to realize the opportunity, but I do think that they're better equipped to win if they

00:41:43.160 | chose to. Well said,

00:41:45.000 | well, really well said. I think the founders share insight is particularly interesting,

00:41:48.760 | Saks. The fact that by the way, sorry for those who did nothing with them.

00:41:52.600 | Gotcha. Yeah, no, no, I was just gonna say the exact same thing. It's like, if they don't use

00:41:56.360 | it now, what would it take? And when? Yeah, this is good. Yet another, yet another case of the

00:42:03.160 | emperor has no clothes, just a power grab by Silicon Valley execs, which was meaningless,

00:42:07.960 | because if in this moment, you don't wield that power, and break that company into bits as you

00:42:12.840 | need to, what was the point of having it? They need to come in and say, we're going to give

00:42:18.360 | barred results to 10% of users and ask them to get feedback on it. Because who has worse

00:42:22.920 | queries than just one point I want to make there for revert, who has more reinforcement learning

00:42:29.240 | than Google, that search box is everywhere. And people write after question and Gmail,

00:42:35.000 | and Google Docs, etc, etc. I mean, they have so many people asking questions. And YouTube might

00:42:40.600 | be the the transcripts of YouTube, every video and the image of every video bananas and the comments

00:42:46.200 | under it, you know, the comments under the video, you have the transcript of what happened in this

00:42:50.440 | video. And then what was the question and answer underneath it? Let me make the counterpoint,

00:42:55.160 | please, to my own point, like, look at how Gerstner came after Zuck. So Zuck had his point

00:42:59.880 | of view, his strongly held belief that AR VR was the future of the platform. That's what he wanted

00:43:05.000 | to bet into. That's what he wanted to lean into. It's what he wanted to build the company against.

00:43:08.520 | He did it. And then the financial analysts and the investors came at him and said,

00:43:13.560 | this is a waste of money, focus on making money, you have a responsibility to shareholders,

00:43:17.880 | F those founders shares, you don't deserve that 10x voting right, or whatever the framing might

00:43:21.880 | have been to get him to say, you know what I acquiesce, I'm giving it up. And I think that

00:43:25.880 | we should also think about what's going to happen on the other side, Google is a trillion plus dollar

00:43:29.320 | market cap company. Their shares are owned by every public endowment, public pension fund,

00:43:34.680 | institutional investor owns Google in their portfolio. So the backlash against Google

00:43:38.840 | making a hard bet like this, and potentially destroying billions of dollars of cash flow

00:43:43.800 | in the process every year will not be easy to do that the same sorts of letters that Gerstner at

00:43:48.680 | all and obviously, we love Gerstner. And, you know, we can all defend him all day long. At

00:43:53.000 | Zuck is what might may end up happening with with alphabet if they did choose to go this path.

00:43:58.040 | Saks, what do you think here about the founder share specifically Google's chances

00:44:02.600 | of disrupting themselves and, you know, just putting this into every product

00:44:07.560 | and shoving it down users throats and catching up?

00:44:10.600 | Well, I mean, with all due respect, Larry and Sergey, I mean, they've been on the beach a long

00:44:16.120 | time. This reminds me of Apollo Creed coming out of retirement in rocket four.

00:44:22.440 | A lot of shape, a lot of shape.

00:44:24.920 | A lot of a lot of fanfare.

00:44:29.880 | They could be a little out of shape.

00:44:31.160 | Sam Altman may not look like Ivan Drago, but but this this is one shrewd character.

00:44:35.480 | This is one shrewd character. I mean,

00:44:37.400 | Altman is fit.

00:44:38.200 | He's fit.

00:44:38.840 | He's been in the arena.

00:44:40.440 | Yeah, he's, you know, he's a multi time founder who sat at the top of YC and got to see everything

00:44:44.600 | that worked. Yep. And got to see all the research and he's been plugging away at this for what,

00:44:51.240 | like, years. So there's a there's a big I just think there's a big gap to catch up on. Now,

00:44:57.320 | Google has all the resources in the world. And they've got a lot of proprietary assets,

00:45:01.880 | too. And they've got all the incentive in the world. So do I think that Google will be

00:45:05.320 | one of the top four players in AI? Absolutely. But this idea that is going to come in steamroll

00:45:10.760 | open AI, I have a prediction. I got a prediction. But then next year, Larry and Sergey take the

00:45:15.560 | title of co CEOs. And then they do a demo day where the two of them get on stage.

00:45:20.520 | They actually do the demos of these products. If that happens, that fictional quantification,

00:45:26.440 | that's it. That's listen, and base us to the run for president. Those are my two predictions.

00:45:30.520 | I'm taking love riddance. Can you imagine if Larry freeberg? Where are the chances of Larry

00:45:34.760 | and Sergey taking co CEO slots? That's prediction one and then prediction two,

00:45:39.000 | where are the chances of them running the next Google IO, where they get on stage and they walk

00:45:44.200 | people through all the products that they shepherded and that they have a vested interest in that

00:45:48.840 | there they want to demo. There is an institutional problem at Google at the top level, which does

00:45:54.760 | need to be solved, which is this position of constantly being in defense against the scrutiny,

00:46:02.040 | again, of regulators and public policy folks and, you know, all these different groups that are

00:46:07.240 | against Google. And so as a result, the kind of cultural seasoning, particularly the executive

00:46:13.800 | and the board level has been one of like, you know, protect the nest, don't overreach, don't

00:46:19.720 | overstep. And it's a real, you know, I think one for the for the business school books or whatever,

00:46:25.800 | ultimately is what they end up doing about it. Because now is, you know, the time when

00:46:30.200 | that defensive posture is really kind of putting the entire business at risk.

00:46:34.040 | The same thing happened to Microsoft. Remember in the late 90s? That's right,

00:46:36.920 | when they got crushed by that antitrust lawsuit, that can be very defensive.

00:46:41.320 | Well, that can know but that consent decree,

00:46:43.000 | they put up they had a wartime CEO come in, Balmer came in. And, you know, followed by kind

00:46:48.200 | of an innovative guy who could kind of continue to build. And I think that there may be a moment

00:46:53.720 | here. I look, I love Sundar. He's a great guy, great CEO, because sooner or not, I don't forget,

00:46:58.760 | if I ever told you this, he and I started at Google on the same day. We're both in the same

00:47:01.800 | nuclear class, we were the freaking hat on the TGIF day and on stage. He was a product manager,

00:47:07.160 | and now he runs the company. But I think the question is like, whether it's the CEO or the

00:47:13.320 | broader whole kind of executive org or the board, a degree of disruption necessary to shift that

00:47:18.440 | cultural seasoning is so necessary right now, for them to have a shot at this. And similar to what

00:47:22.760 | you just said, sacks, like you're gonna need a bomber type moment to kind of, you know,

00:47:27.400 | reinvigorate that business. And by the way, I'll tell you such a moment, I think that

00:47:31.480 | it's an important port point when bomber took over during that period after gates,

00:47:37.240 | when they were on their heels, he basically just focused on revenue and paying dividends

00:47:42.680 | and stock buybacks, and the stock went sideways. And he missed mobile. And now,

00:47:47.560 | yeah, it's a good point.

00:47:48.840 | Jay K. You're forgetting one big thing, which is that that was also because he had to operate

00:47:53.480 | under a consent decree to the DOJ. Exactly. So the product managers of Microsoft were replaced

00:47:57.880 | with lawyers from the Department of Justice, and you had to get their sign off before you

00:48:01.560 | could ship anything. So we have to remember that those things probably slowed Microsoft down as

00:48:06.920 | well. And the great thing that Satya had was a blank slate and the removal of that consent decree.

00:48:14.120 | So he was able to do everything that just made a lot of sense. And he's executed flawlessly.

00:48:19.720 | I think the problem at Google is not Sundar or Larry or Sergey. I think it's more in the deep

00:48:28.520 | bowels of middle management of that company, which is that there's just far too many people

00:48:33.080 | that probably have an opinion. And their opinion is not shrouded in survival. Their opinion is

00:48:39.960 | shrouded in elite language around what is the moral and ethical implications of this and where

00:48:46.040 | has this been properly tested on the diaspora of 19 different ethnic tribes of the Amazon.

00:48:53.000 | That's the kind of decision making that is a nice to have when you are the second or third

00:48:58.920 | most valuable technology company in the world. But you have to be able to pause that kind of

00:49:05.080 | thinking and instead get into wartime survival mode. And it's very hard. So it doesn't almost

00:49:11.000 | matter. At this point, what Sundar wants, the real question is, what is the capability of middle

00:49:17.720 | management to either do it or get out of the way. And I think that in all of these big companies

00:49:23.080 | that struggle, what you really see is an inability for middle management to get out of the way or

00:49:28.680 | frankly, just you need somebody to then fire them. And if you look at folks who get their groove back,

00:49:34.840 | let's see what Facebook does. What are they targeting? They're targeting middle management.

00:49:40.360 | If you look at what Elon does in the companies that he owns, there is virtually no middle management.

00:49:46.840 | It's like get out of the way, build product, build product and ship it. Yeah. And what is

00:49:53.640 | the core truth? And so if failure is there in front of you, and if David is right, that you

00:49:58.200 | have 200 million users come out of nowhere, who are voting every day with their time and attention

00:50:03.720 | to use an app, and that doesn't create a five alarm fire where you get middle management out

00:50:09.080 | of the way and you are the senior most people talking to the people doing the work and shipping

00:50:15.000 | things every day you're you are toast. You are toast. A lot of people are starting to think

00:50:20.760 | we're moving a little bit too fast. When it comes to open AI is incredible performance, which I GPT

00:50:27.480 | for the plugins and all this and so the future of life Institute, which was formed in 2015. It's a

00:50:33.560 | nonprofit that's focused on de risking major technology like AI. They did a petition titled

00:50:41.240 | pause giant AI experiments and open letter, a bunch of computer scientists sign this letter.

00:50:47.000 | And the letter quote says we must ask ourselves should we let machines flood our information

00:50:54.840 | channels with propaganda and untruth? Should we automate away all the jobs including the

00:50:59.880 | fulfilling ones? Should we develop non human minds that might eventually outnumber outsmart

00:51:04.440 | obsolete and replace us? Should we risk loss of control of our civilization, a number of notable

00:51:09.640 | tech leaders like Elon, Steve Wozniak, and a handful of deep mind researchers have signed it.

00:51:15.320 | What do you guys think of the latter? Are we going to slow down or not? And then we can ask the

00:51:18.840 | question generally, how close are we getting to AGI, which is what everybody's scared of,

00:51:22.680 | is that these agents start working with each other in the background to do things that are

00:51:27.560 | against human interest. I know it sounds like science fiction. But there is a theory that when

00:51:32.840 | these AI is start operating on their own, like we explained in the previous sort of segment here with

00:51:38.920 | plugins, and they make agents that are operating based on feedback from each other, could they get

00:51:45.560 | out of control and be mischievous and then work against human interest? Or what do you think sex?

00:51:49.880 | I think there's a difference between what could happen in the short term and then what could

00:51:53.560 | happen in the long term. I think in the short term, everything we're seeing right now is

00:51:58.120 | very positive. And let me just give you an example. There was a really interesting

00:52:02.600 | tweet storm about a guy who wrote about how Chad GPT saved his dog. And do you guys see this?

00:52:09.160 | This is one of the really mind blowing ones to me, use cases. So his dog was sick, took him to a vet,

00:52:16.680 | vet prescribed some medication, three days later, dog still sick, in fact, even worse.

00:52:20.600 | So the owner of the pet just literally copied and pasted the lab result for the blood test for the

00:52:27.720 | dog with all the lab values into chat GPT and said, what could this be like, what's your

00:52:33.240 | likely diagnosis? Chat GPT gave three possible answers, three illnesses. The first one was what

00:52:40.200 | the vet basically diagnosed with. So that wasn't it. The second one was excluded by another test.

00:52:45.480 | So he then went to a second vet and said, listen, I think my dog has the third one.

00:52:50.200 | And vet prescribed something and sure enough, dog is cured, saved. So that's really mind blowing

00:52:57.880 | that even though chat GPT hasn't been specifically optimized, as far as we know, for lab results,

00:53:04.120 | it could figure this out. The reason I'm mentioning this is it gives you a sense of

00:53:07.400 | the potential here to cure disease to, you know, like I could see major medical breakthroughs

00:53:14.760 | based on the AI in the next five or 10 years. Now, the question is, like, what happens in

00:53:19.400 | the long term, you know, as the AI gets smarter and smarter, and we are kind of getting

00:53:23.960 | into the realm of science fiction, but here would be the scenario is you're on

00:53:27.960 | chat GPT 10, or 20, or whatever it is, or maybe some other companies AI. And the developers asked

00:53:35.800 | the AI, hey, how could you make yourself better? Now do it? Which is a question we asked chat GPT

00:53:41.640 | all the time in different contexts. And so chat GPT will already have the ability to write perfect

00:53:47.400 | code by that point. I think, you know, code writing is one of the I think of its superpowers already.

00:53:52.040 | So it gives itself the ability to rewrite its code to auto update it to recursively make itself

00:53:57.880 | better. I mean, at that point, isn't that like a speciation event doesn't that very quickly lead

00:54:04.200 | to the singularity if the AI has the capability to rewrite its own code to make itself better?

00:54:11.160 | And you know, won't it very quickly write billions of versions of itself.

00:54:14.200 | And, you know, it's very hard to predict what that future looks like. Now, I also don't know

00:54:19.320 | how far away we are from that that could be 10 years, 20 years, 30 years, whatever.

00:54:22.680 | But I think it's a question worth asking for sure.

00:54:26.600 | Is it worth slowing down, though, sex? Should we be pausing because, based on what you said,

00:54:31.880 | you know, I think you've framed it properly. When these things hit a certain point, and they start

00:54:36.200 | reinforcing their own learning with each other, they can go at infinite speed, right? This is not

00:54:42.120 | comparable to human speed, they could be firing off millions, billions of different I think you're

00:54:47.400 | right scenarios. We're definitely now on this fuck around, find out curve. Yeah. And so there's only

00:54:54.600 | one way to really find out which is somebody is going to push the boundaries, the competitive

00:55:00.120 | dynamics will get the better of some startup, they'll do something that people will look back

00:55:06.120 | on and say, Whoa, that was a little that was a bridge too far. So yeah, we're just a matter of

00:55:11.320 | time. Yeah, I think we're not going to slow down. I actually think it's going the other way. I think

00:55:16.120 | things are going to speed up. And the reason they're going to speed up is because the one

00:55:20.280 | thing Silicon Valley is really good at is taking advantage of a platform shift. And so when you

00:55:25.320 | think about like all the VCs, and all the founders, you know, everyone accuses us of being lemmings.

00:55:31.480 | And so when there's like kind of like a fake platform shift, or people kind of glom on to

00:55:37.480 | something that ends up not being real, everyone's kind of got egg on their faces. But the flip side

00:55:42.120 | of that is that when the platform shift is real, Silicon Valley is really good at throwing money

00:55:47.560 | at it, the talent knows how to go after it. And they keep making it better and better. And so

00:55:53.720 | that's the dynamic we're in right now. You look at 70% of the last YC class was ready all AI

00:55:58.120 | startups for the next one probably 95%. So I think that we're on a path here where the pace of

00:56:04.360 | innovation is actually going to speed up. Companies are going to compete with each other, they're

00:56:08.520 | going to seek to invent new capabilities. And I think the results are going to all be incredibly

00:56:13.960 | positive for some period of time, like you know, the vet example, we're going to cure

00:56:18.840 | illnesses, we're going to solve major problems are positive, then we invest more, we trust more.

00:56:25.320 | But the paradox of that as Chamath is pointing out, Friedberg is if we trust it more, we invest

00:56:29.720 | more than some person in a free market is going to say, you know what, I need to be

00:56:34.360 | chat GPT. Therefore, I'm going to take the rails off this thing, I'm going to let it go

00:56:38.440 | faster, and take off some constraints, because I need to win. And I'm so far behind.

00:56:43.080 | How do you feel about that scenario that sort of Chamath and

00:56:47.400 | Saks T. Dep Friedberg, I think there's like, GPT three, I think ran on 700 gigs. Is that right?

00:56:57.320 | Does anyone know what GPT four runs on? It's got to be on some number that's, you know, not too

00:57:03.800 | not not many multiples of that. But look, someone could make a copy of this thing and fork it and

00:57:12.840 | develop an entirely new model. I think that's what's incredible about software and digital

00:57:20.600 | technology and also kind of, you know, means that it's very hard to contain. Similar to like what

00:57:26.760 | we've seen in biology ever since biology got digitized through DNA sequencing, and the ability

00:57:32.920 | to kind of express molecules through gene editing. You know, you can't control or contain the ability

00:57:39.720 | to do gene editing work at all. Because everyone knows the code. Everyone can make CRISPR cast

00:57:46.280 | molecules, everyone can make gene editing systems in any lab anywhere. Once it was out, it was out.

00:57:51.480 | And now there's hundreds of variants for doing gene editing, many of which are much improved

00:57:57.240 | over CRISPR cast nine, I use that as an analogy, because it was this breakthrough technology that

00:58:01.480 | allowed us to precisely, specifically edit genomes. And that allowed us to engineer biology and do

00:58:06.120 | these incredible things where biology effectively became software. And remember, CRISPR cast nine

00:58:11.160 | gave us effectively a word processing type tool find and replace. And the tooling that's evolved

00:58:18.280 | from that is much better. So whatever is underlying, whatever the parameters are for GPT for

00:58:26.280 | whatever that model is, if a close enough replicant of that model exists, or a copy of that model is

00:58:32.680 | made, and then new training data and new evolutions can be done separately, you could see many, many

00:58:37.960 | variants kind of emerge from here. And I think this is a good echoing of chumas point, we don't

00:58:41.960 | know what's ultimately going to win. Is there enough of a network effect in the plugin model,

00:58:46.440 | as Saks pointed out, to really give open AI the sustaining competitive advantage? I'm not sure.

00:58:51.560 | The model runs on 700 gigs. That's less data than, you know, fits on my iPhone. So you know,

00:58:58.200 | I could take that model, I could take the parameters of that model. And I could create

00:59:01.720 | an entirely new version, I could fork it, and I could do something entirely new with it.

00:59:04.680 | So I don't think you can contain it. I don't think that this idea that we can put in place,

00:59:10.200 | some regulatory constraints and say, it's illegal to do this, or, you know, try and,

00:59:17.080 | you know, create IP around it, or protections around it is realistic at this state,

00:59:21.480 | the power of the tool is so extraordinary, the extendability of the tools are so extraordinary.

00:59:25.960 | So the economic and the various incentives are there for, you know, other models to emerge,

00:59:32.520 | and whether they're directly copied from someone hacking into open AI servers and making a copy of

00:59:37.080 | that model, or whether they're, you know, open sourced, or whether that someone generates

00:59:40.760 | something that's 95% is good, and then it forks and a whole new class of models emerge.

00:59:45.720 | I think this is like, it's as Saks pointed out, highlighting the kind of economic, market

00:59:52.200 | uprooting, social uprooting potential, and many models will start to kind of come to market.

00:59:57.800 | What do we think the impact of white collar jobs getting annihilated by this technology,

01:00:02.120 | if that in fact, I want to say one thing on this. Yeah, look, I just share one example here. So

01:00:06.840 | here's a Reddit post that I was made aware of earlier this week. I lost everything that made

01:00:11.960 | me love my job through mid journey overnight. I am employed as a 3d artist and a small games

01:00:16.840 | company of 10 people. Our team is two people who basically explains he says since mid journey

01:00:22.520 | version five came out, he's not an artist anymore, nor a 3d artist. All they do is prompting

01:00:27.800 | photoshopping and implementing good looking pictures. And he basically says, this happened

01:00:33.080 | overnight, and he had no choices. Boss also had no choice. He says I am now able to create rig

01:00:38.120 | and animate a character that spit out from MJ mid journey in two to three days before it took

01:00:43.640 | us several weeks and 3d. The difference is that he cares about his you know, job and for his boss,

01:00:48.440 | it's just a huge time money saver. He's no longer making art and the person who was number two in

01:00:53.160 | the organization who didn't make as good content as him is now embracing this technology because

01:00:58.600 | it carries favor with his boss. And he ends basically saying, getting a job in the game

01:01:04.280 | industry is already hard, but leaving a company and a nice team because AI took my job feels very

01:01:09.560 | dystopian. I doubt it would be better in a different company. Also, I am between grief

01:01:14.520 | and anger and I am sorry for using my gosh, your art fellow artists. This is yet another reason

01:01:20.200 | that figma really needs to close this acquisition from Adobe. I mean, it's like, the value of these

01:01:27.080 | apps are just getting gutted. If you take a workflow management tool for things like design

01:01:33.480 | and imagery, and you reduce it by an order of 90% it's like what is that app experience worth?

01:01:40.200 | And how could you replicate it if you were a big company that already has distribution? That's one

01:01:47.240 | comment. But what I would tell you Jason to answer the white collar question is I think there are a

01:01:51.560 | handful of companies you need to look at exclusively because they will be the first ones to

01:01:56.520 | really figure out how to displace human labor. And that is TCS. So Tata Consulting Services,

01:02:03.800 | Accenture, Cognizant, these are all the folks that do coding for hire work at scale. I think

01:02:10.920 | Accenture has something like 750,000 employees. So the incentive to sort of squeeze op ex to

01:02:18.280 | create better utilization rates to increase profitability is quite obvious. It always has

01:02:24.280 | been they will be the first people to figure out how to use these tools at scale. Before the law

01:02:29.240 | firms or the accounting firms or any of those folks, even sort of try to figure out how to

01:02:33.320 | displace white collar labor, I think is going to be the coding jobs and it's going to be the coding

01:02:36.760 | for hire jobs that companies like Accenture and TCS. So those business processing do for other

01:02:42.360 | people, developer kind of folks, they're going to need half as many people 25% as many people,

01:02:48.120 | we're going to find out the efficient frontier. Yeah.

01:02:50.440 | I see it a different way. I mean, this argument that productivity leads to job loss has been made

01:02:56.200 | for hundreds of years. And it's always been refuted. When you make human beings more productive,

01:03:01.240 | it leads to more prosperity, more wealth, growth, more growth. And so yeah, it's easy to think about

01:03:07.080 | in a narrow way, the jobs are going to be displaced. But but why would that be? It's

01:03:11.000 | because you're giving leverage to other human beings to get more done. And some of those human

01:03:16.040 | beings, really anybody with a good idea is now going to be able to create a startup much more

01:03:20.840 | easily. So you're going to see a huge explosion in creativity in startup creation, new companies,

01:03:27.400 | new jobs. Imagine think about the case of you know, Zuckerberg founding Facebook at Harvard,

01:03:33.480 | he wrote the first version himself, maybe with a couple of friends, that project happened and

01:03:38.520 | turned into a giant company because he was able to self execute his idea without needing to raise

01:03:44.360 | venture capital or even recruit employees. Even really before forming a company, anyone with a

01:03:50.520 | good idea to be able to do that soon, you're going to be able to use these AI tools, they truly

01:03:54.280 | will be no code, you'll be able to create an app or a website just by speaking to some AI program

01:03:59.640 | in a natural language way. So more flowers will bloom more startups, more projects. Now,

01:04:04.440 | it will create I think a lot of dislocation. But for every testimonial that is like the one that

01:04:11.880 | you showed, which I think is, I'd say a little bit overly dramatic, I have seen 10 or 100

01:04:17.960 | testimonials from coders on Twitter or other blogs, talking about the power that these new

01:04:24.280 | tools give them. They are like this makes me a 10x engineer, right? And, and especially these

01:04:29.320 | like junior engineers who are right out of school, who don't have 20 years of coding history,

01:04:33.400 | they get superpowers right away. Like it makes them so much better.

01:04:37.160 | It's a proper response.

01:04:38.200 | Let me give you a response to that guy. So so and using sexist point, that guy saying what used to

01:04:45.160 | take me weeks, I can now do in two to three days. And I feel like my work is gone. And that's because

01:04:50.680 | he's thinking in terms of his output being static. And if he thinks about his output being dynamic,

01:04:56.360 | he can now in the matter of three weeks, instead of making one character, he cannot make a character

01:05:01.960 | every two days. So he can make 30 characters in three weeks. That's an alternative way for him

01:05:07.400 | to think about what this tooling does for him and his business, the number of video games will go up

01:05:12.360 | by 10x, or 100x or 1000x. The number of movies and videos that can be rendered in computers can go up

01:05:18.600 | by 10x or 100x or 1000x. This is why I really believe strongly that in some period of time,

01:05:25.000 | we will all have our own movie or our own video game ultimately generated for us on the fly.

01:05:29.880 | Based on our particular interests, there will certainly be shared culture, shared themes,

01:05:34.040 | you know, shared morality, shared things that that tie all these things together.

01:05:39.480 | And that will become the shared experience. But in terms of like us all consuming the same content,

01:05:43.960 | it will really like you with YouTube and Tick Tock, we're all consuming different stuff all the time.

01:05:48.680 | And this will enable an acceleration of that evolution and personalization. I'll also highlight,

01:05:54.120 | you know, back in the day, one human had to farm a farm by hand. And we eventually got the tool of

01:05:59.960 | a hoe and we can put in the ground and make, you know, make stuff faster. And then we got a plow.

01:06:05.400 | And then we got a tractor. And today, agricultural farm equipment allows one farmer to farm over 10,000

01:06:13.240 | acres, you go to Western Australia, it's incredible. These guys have 24 row planters

01:06:16.360 | and harvesters. And it's completely changed the game. So the unit of output per farmer

01:06:21.640 | is now literally millions of times what it was just 150 years ago.

01:06:25.480 | And in that case, Freeberg, nobody wants to do backbreaking labor in the fields. And everybody

01:06:30.280 | wants that cheaper food. But in this case, let me just read you one quote that I didn't read in the

01:06:34.520 | original reading of this. He says, I want to make art that isn't the result of scraped internet

01:06:39.960 | content from artists that were not asked. And so I think that's part of this is that it's bespoke

01:06:44.920 | art. But the one question I have for sacks was, sacks, you we started this conversation, we're

01:06:51.720 | saying, hey, this is different than anything in terms of efficiency that came before it. This is,

01:06:55.640 | I'm going to put some words in my theory, but this is like a step function, more efficient. So

01:07:00.520 | to the argument of, hey, efficiency has always resulted in, you know, more ideas, and we've

01:07:07.640 | found something to do with people's time, is this time different, potentially,

01:07:11.240 | because this is so much more powerful. This isn't just like a spellchecker.

01:07:14.360 | I would say differently, I think, and I agree with what J. Cal is saying, because I think that

01:07:18.360 | the thing that technology has never done is tried to displace human judgment. It's allowed us to

01:07:29.480 | replace physical exertion of energy, but it has always preserved humans injecting our judgment.

01:07:37.000 | And I think this is the first time where we're being challenged with autonomous systems

01:07:42.440 | that has some level of judgment. Now we can say, and it's true,

01:07:47.560 | again, we're four months in that that judgment isn't so great. But eventually,

01:07:53.320 | and because of the pace of innovation, eventually is probably not that far away.

01:07:58.760 | To judgment will become perfect. I'll give you a totally different example. You know,

01:08:02.520 | how many pilots are there in the world? Will we, at some point in the next 10 years, want

01:08:09.320 | folks to actually manually take off and land? Or will we want precision guided instrumentation and

01:08:16.680 | computers and sensors that can guarantee a pitch perfect landing every single time in all kinds of

01:08:21.800 | weather conditions so that now planes can even have 50 x the number of sensors with a computer

01:08:28.520 | that can then process it and act accordingly. Just a random example that isn't even thought

01:08:32.840 | of when we talk about sort of where AI is going to rear its head. I think that this judgment idea

01:08:38.120 | is an important one to figure out. Because this is the first time I've seen something that is

01:08:43.160 | bumping up against our ability to have judgment. And what this person was talking about in this

01:08:47.240 | mid journey example is his judgment has been usurped. Yes. Yeah, I would disagree.

01:08:53.320 | I don't know. Yeah, let me just let me just make one point on this. So, you know, an image is a

01:08:59.720 | matrix of, you know, data that's rendered on a screen and as pixels and those pixels are different

01:09:05.480 | colors. And, you know, that's what an image is. Is it or is it is it the judgment of the creator?

01:09:11.000 | Well, no, I'm just saying an image in general. So like when Adobe Photoshop and digital photography

01:09:15.240 | arose, photographers were like, this is, you know, BS, why are you digitizing photography

01:09:20.200 | was analog and beautiful before. And then what digital photography allowed is the photographer

01:09:25.080 | to do editing and to do work that was creative beyond what was possible with just a natural

01:09:30.920 | photograph taken through a camera. And they're arguably different art forms, but it was a new

01:09:34.600 | kind of art form that emerged through digital photography. And then in the early 90s, there

01:09:38.840 | was a plugin suite called Kai's power tools that came out in Adobe Photoshop. And it was a third

01:09:44.920 | party plugin set, you would you would buy it and then it would work on Photoshop and it did things

01:09:49.080 | like motion blur, sharpening, pixelation, all these interesting kind of like features. And

01:09:54.680 | prior to those tools coming out, the judgment of the digital artist, the digital photographer was

01:09:59.480 | to go in and do pixel by pixel changes on the image to make that pixel to make that image look

01:10:04.600 | blurry, or to make it look sharper, or to make it look like it had some really interesting motion

01:10:08.840 | feature. And the Kai's power tools created this instant toolkit where in a few seconds, you

01:10:14.200 | created a blur on the image. And that was an incredible toolkit. But a lot of digital artists

01:10:19.160 | said, this is automating my work. What is my point now? Why am I here? And the same happened in

01:10:24.280 | animation when three when you know, CGI came around, and animators were no longer animating

01:10:28.760 | cells by hand. And in every point in this evolution, there was a feeling of loss initially,

01:10:33.880 | but then the evolution of a whole new art form emerged, and an evolution of a whole new area of

01:10:38.040 | human creative expression emerged. And I think we don't yet know what that's going to look like.

01:10:42.280 | Do you think you think the the level of judgment that AI offers you is the same as the level of

01:10:49.160 | judgment that Kai power tools offered? Yeah, look, I mean, I think that the person making

01:10:52.600 | the judgment or the decision about which pixel to change into what color felt like, you know,

01:10:56.600 | I have control. And I think it's ultimately like, I just told her I disagree with you. I mean,

01:11:00.840 | I think that this is a magnitude different number. It's more of a magnitude. Yeah,

01:11:05.480 | it's still love. It's on you. You don't look nice. You and I have sat in spreadsheets. And

01:11:11.800 | we've I'm generally happy with this idea. So I'll give you a different example.

01:11:14.920 | Today, we use radiologists and pathologists to identify cancers. Yep. There are closed loop

01:11:22.600 | systems. We have one right now that's in front of the FDA. That is a total closed loop system

01:11:27.240 | that will not need any human input. So I don't know what those folks do.

01:11:33.000 | Except what I can tell you is that we can get cancer detection, basically down to a 0% error

01:11:40.600 | rate. That is not possible with human intervention. That is judgment. Right? So I just think it's

01:11:48.040 | important to really acknowledge that this is happening at a level that it's never happened

01:11:52.200 | before. You may be right that there's some amazing job for that radiologist or pathologist to do in

01:11:58.040 | the future. I don't know offhand what that is. But these are closed loop systems now,

01:12:04.360 | that think for themselves and self improve. I get it. But I think that there there is an

01:12:08.760 | unfathomable set of things that emerge. We did not have the concept of Instagram influencers,

01:12:14.360 | we did not have the concept of personal trainers, we did not have the concept of like,

01:12:19.800 | all these new jobs that have emerged in the past couple of decades, that people enjoy doing that

01:12:25.000 | they can make money doing that is a greater kind of experience and level of fulfillment for those

01:12:29.800 | that choose and have the freedom to do it than what they were having to do before, when they

01:12:34.200 | had to work just to make money. What do you think that radiologist or pathologist wants to do?

01:12:38.760 | Be a trainer or Pilates instructor? No, I think that's gonna look like

01:12:42.600 | all right, Sam. Yeah. You have any thoughts on this? As we wrap this topic? It's obviously a

01:12:48.120 | lot of passion coming out. Yeah. Elimination of white collar jobs in a massive way.

01:12:52.760 | I think that this is a short term versus long term thing. In the short term, I see the benefits

01:12:57.240 | of AI being very positive, because I don't think it's in most cases, wiping out human jobs is

01:13:03.240 | making them way more productive. You still need the developer. It says that there are five times

01:13:07.960 | or 10x more productive. But I don't think we're at the point in the short term, we're gonna be

01:13:12.600 | able to eliminate that role entirely. What I've seen in basically every startup I've ever been

01:13:17.800 | a part of is that the limiting factor on progress is always engineering bandwidth. That is always

01:13:23.800 | the thing that you wish you had more of. Totally.

01:13:26.760 | It's the product roadmap is always the most competed on thing inside the organization.

01:13:31.000 | Everyone's trying to get their project prioritized, because there's just never

01:13:35.080 | enough engineering bandwidth. It's really the lifeblood of the company. If you make

01:13:39.240 | the developers more productive, it maybe just accelerates the product roadmap.

01:13:42.680 | I don't think in the short term that what's going to happen is these companies are going to look to

01:13:47.800 | cut all their developers because one or two of them can do 10x the work. I think that they're

01:13:52.360 | going to try and accelerate their product roadmaps. Now, again, you have this long term concern that

01:13:58.280 | maybe you don't need developers at all at some point. But I think that the benefits of developing

01:14:03.640 | this technology are so great in the short to midterm that we're going down that path no matter

01:14:07.720 | what. And we're just gonna have to find out what that long term really looks like. And maybe

01:14:11.480 | will look very different. I mean, once we get past the short term,

01:14:16.520 | we may have a different long term view. I think in this narrow vertical, I 100%

01:14:21.560 | agree with you. Look, I think that AI is going to eliminate unit testing, it has already done.

01:14:26.600 | So it's going to eliminate most forms of coding, the engineers that you have, all of them will now

01:14:31.560 | become 10x engineers. So with fewer of them, or with the same number, you'll be able to do as much

01:14:37.480 | or more than you could have before. That's a wonderful thing. And all I'm saying on that

01:14:40.840 | specific narrow vertical is you'll see it first rear its head in companies like Accenture and TCS

01:14:46.040 | because and cognizant because they have an immediate incentive to use this tooling to drive

01:14:52.760 | efficiency and profitability that's rewarded by shareholders. It'll be less visible in other

01:14:58.040 | companies. So but what I am saying, though, is that you have to think about the impact

01:15:02.280 | on the end markets for a second. And I think that AI does something that other technology layers

01:15:10.280 | have never done before, which is supplant human judgment in a closed loop manner. And I just think

01:15:15.960 | it's worth appreciating that there are many systems and many jobs that reply that rely on human

01:15:21.320 | judgment. Where we deal with error bars, and an error rate that a computer will just destroy and

01:15:30.840 | blow out of the water. And we will have to ask ourselves, should this class of job exist with

01:15:36.760 | its inherent error rate? Or should it get replaced fully by a computer which has no error rate? And

01:15:41.960 | I think that's an important question that's worth putting on the table.

01:15:44.440 | Okay, so let's wrap here. I just have my final thought on it is like, you're going to see

01:15:49.080 | entire jobs, categories of jobs go away. We've seen this before phone operators, travel agents,

01:15:54.760 | copy editors, illustrators, logo designers, accountants, sales development reps, I'm seeing

01:15:58.840 | a lot of these job functions in the modern world, like phone operators previously, I think these

01:16:04.920 | could wholesale just go away. And they would just be done by AI. And I think it's going to happen

01:16:08.840 | in a very short period of time. And so it's gonna be about who can transition. And some people might

01:16:13.160 | not be able to make the transition. And that's going to be pain and suffering. And it's going

01:16:15.720 | to be in the white collar ranks. And those people have more influence. So I think this is could

01:16:19.960 | lead to some societal disturbance. I'm going to learn Pilates and be an influencer.

01:16:25.160 | That's it. But I do agree with sacks that the software development backlog, if this is what

01:16:29.000 | you're saying is so great, that I don't think we'll see it in software development for a decade

01:16:33.160 | or two. There's just so much software that still needs to be made. All right, last week, we talked

01:16:36.200 | about Tick Tock. And this first bipartisan hearing we've seen in a long time. And people actually,

01:16:42.360 | I think framing correctly exactly how dangerous it is, in my opinion, to have Tick Tock in the

01:16:46.280 | United States. And of course, then we get the great disappointment of the actual bill,

01:16:52.520 | the restrict act was proposed by Senator Mark Warner, Democrat Virginia, on March 7. The problem

01:16:59.000 | with it is, is it seems like it's poorly worded, that there will be civil penalties and criminal

01:17:04.600 | penalties to Americans for breaking the law and using software that's been banned. And many people

01:17:12.280 | said, you know, this probably is just bad language. I have a question. Yeah, does it does

01:17:17.640 | this apply to incognito mode? Because if it doesn't, I don't know, it is not. Yes. They're

01:17:25.400 | saying they're saying that you can get, you know, you can get fined or 20 years in jail, whatever it

01:17:30.440 | is, for using a VPN VPN to Tick Tock. Freeberg, what are your thoughts on it? Look, I think this

01:17:36.200 | is a real threat to the open internet. I'm really concerned about the language that's been used

01:17:42.200 | that basically speaks to protecting the safety and security of the American people

01:17:46.040 | by actively monitoring network traffic, and making decisions about what network traffic

01:17:51.480 | is and isn't allowed to be transmitted across the open internet. It's the first time that I

01:17:56.120 | think in the United States, we are seeing like a real threat and a real set of behaviors from our

01:18:00.600 | government that looks and feels a lot like what goes on in China and elsewhere, where they operate

01:18:05.960 | with a closed internet and internet that's controlled, monitored, observed, tracked, and,

01:18:11.640 | and gates are decided by some set of administrators on what is and isn't appropriate. And the language

01:18:17.160 | is always the same. It's for safety and security of the people. The entire purpose of the internet

01:18:22.120 | is that it did not have bounds that it did not have governments that it did not have controls

01:18:26.120 | that it did not have systems that are politically and economically influenced that the architecture

01:18:31.160 | of the internet was and always would be open. The protocols are open, the transmission of data on

01:18:35.880 | that network would be open. And as a result, all people around the world would have access to

01:18:40.600 | information of their choosing, and it allowed ultimate freedom of choice. You know, this,

01:18:45.880 | this kind of is the first of what I'm concerned, creates a precedent that ultimately leads to a

01:18:50.520 | very slippery slope, saying that tick tock cannot make money in the US by charging advertisers or

01:18:55.560 | managing commerce flows is one thing. That's where the government can and should and could,

01:18:59.720 | if they chose to have a role. But I think going in and observing tracking internet traffic and

01:19:04.840 | making decisions about what is and isn't appropriate for people, I think, is one of the

01:19:10.280 | things that we all should be most concerned about what's going on right now. There is no end in sight

01:19:15.240 | to this. If you allow this to happen in the first time, you know, VPNs, virtual private networks,

01:19:21.880 | allow you to anonymously access internet traffic and and access internet traffic via remote

01:19:29.240 | destinations. So, so that the ultimate consumption of content that you're using can't be tracked and

01:19:36.520 | monitored by local agencies or ISPs. And I think that saying that that can now be restricted,

01:19:42.680 | takes away all ability to have true privacy and all rights to privacy on the open internet.

01:19:48.120 | So I'd love to talk about this more. Unfortunately, I got to run.

01:19:51.400 | Great. So this is a super threat to me. And I think this is something we should be super,

01:19:56.680 | super concerned about. And that the entire community of technology, internet, and anyone

01:20:01.560 | that wants to have, you know, freedom of choice, steps up and says, this is totally inappropriate

01:20:06.520 | and overreach. Yeah, there are other ways to manage stuff like this.

01:20:09.080 | Feels like complete overreach, Sax.

01:20:11.080 | Yeah, I agree with this.

01:20:12.760 | Intentional overreach or poorly written or somewhere in between? What do you think?

01:20:15.880 | Both. I think both. I think this is the biggest bait and switch that Washington,

01:20:20.360 | the central government has ever tried to pull on us. Everybody thinks that they're just trying to

01:20:24.600 | ban TikTok from operating in the US. And if that's all they did, then I think the bill would be

01:20:29.400 | supported by most Americans. But that's not what they're doing. They're not restricting TikTok,

01:20:33.560 | they're restricting us.

01:20:34.680 | That's not the goal here. Yeah. What a bait and switch.

01:20:38.360 | It's a huge bait and switch. And so just so you know, what the act provides is that a US citizen

01:20:42.920 | using a VPN to access TikTok could theoretically be subjected to a maximum penalty of 1 million

01:20:49.320 | in fines or 20 years in prison or both. Now, they'll say, Mark Warner, the sponsor of legislation,

01:20:57.000 | will swear up and down. That's not the intent. But the problem is that the language of the bill

01:21:01.160 | is so vague that some clever prosecutor may want to pursue this theory one day.

01:21:05.720 | And that needs to be stopped. Also, there's another problem with the bill, which is,

01:21:10.520 | you think this is just about TikTok. It's not. What they do is, it says here, I guess they don't

01:21:17.240 | want to mention TikTok by name. So they're trying to create a category of threatening application.

01:21:22.360 | But because it is a category, it's very, very broad. So the bill states that it covers any

01:21:29.320 | transaction, transaction, not just an app, in which an entity described in subparagraph B has

01:21:35.560 | any interest. And then entities described in subparagraph B are, quote, a foreign adversary,

01:21:40.680 | an entity subject to the jurisdiction of or organizing the laws of a foreign adversary,

01:21:45.240 | an entity owned, directed, or controlled by either of these. And then it gives the executive branch

01:21:50.520 | the power to name a foreign adversary, any foreign government regime that one of the cabinet

01:21:58.040 | secretaries defines, without any vote of Congress. So this is giving sweeping powers to the executive

01:22:04.920 | branch to declare foreign companies to be enemies. It feels like the plot of the prequels in Star

01:22:11.240 | Wars. Well, like T-Power, here we go. You know, we criticize China for having a great firewall.

01:22:16.760 | What do you think this is? Yeah, I mean, this should obviously have nothing to do with the

01:22:21.960 | American consumer and everything to do with a foreign adversary collecting data of Americans

01:22:26.120 | at scale. This could be written in a much simpler way. Yeah, you know what it should be? It should

01:22:30.520 | be one sentence, which is that app stores are prohibited from allowing TikTok to be an app

01:22:36.760 | in their store. That's what they do in India. That's it. Case closed. Game over. I think India

01:22:40.760 | is doing okay, right? They block like 100 Chinese apps, and I think their society is still functioning.

01:22:45.080 | So, you know, all due respect to AOC, you know, like the idea that 150 million Americans are

01:22:51.080 | going to suffer because they can't be tracked by the CCP is kind of nuts. This is going to give

01:22:55.480 | sweeping powers to the security state to surveil us, to prosecute us, to limit our internet usage.

01:23:05.080 | This is basically the biggest power grab and bait and switch they've ever tried to pull on us.

01:23:09.880 | And again, if they really were concerned about TikTok, it's one sentence.

01:23:13.560 | Yeah, we were done. All right, everybody. It's been an amazing episode for the

01:23:18.040 | Sultan of Science, David Freeburg, the Rain Man himself, David Sachs, and the dictator,

01:23:22.840 | Chamath Palihapitiya. I am the world's greatest moderator, and we will see you next time. Bye-bye.

01:23:28.600 | [Music]

01:23:30.600 | [Music]

01:23:33.080 | [Music]

01:23:35.720 | [Music]

01:23:37.960 | [Music]

01:23:40.120 | [Music]

01:23:42.120 | [Music]

01:23:44.120 | Let yourself, let yourself be in all of them

01:23:46.120 | Let yourself, let yourself be in all of them

01:23:48.120 | Besties are gone

01:23:52.120 | There's my dog taking a genisine ride

01:23:54.120 | I noticed your driveway sex

01:23:56.120 | Oh man

01:23:58.120 | My avatars will meet me at light

01:24:00.120 | We should all just get a room and have just one big "U"Georgie

01:24:02.120 | because they're all just useless

01:24:04.120 | It's like sexual tension

01:24:06.120 | that we just need to release somehow

01:24:08.120 | What you're about-

01:24:10.120 | What you're about-

01:24:12.120 | B

01:24:14.120 | We need to get merch

01:24:16.120 | Besties are gone

01:24:18.120 | I'm going all in

01:24:20.120 | I'm going all in

01:24:22.120 | I'm going all in

01:24:24.120 | ♪ I'm falling ♪

01:24:26.620 | [BLANK_AUDIO]

E122: Is AI the next great computing platform? ChatGPT vs. Google, containing AGI & RESTRICT Act

Chapters