AI’s Future: Opportunity, Regulation & Limits with Vinny Lingham & Sunny Madra | E1729

0:00This Week in Startups is brought to you by Cashfly is a pure play CDN provider that makes CDN simple, effective, and secure. Deliver content faster than your competitors and get 10 terabytes free forever if you sign up at twist.cashfly.com. That's T-W-I-S-T dot C-A-C-H-E-F-L-Y dot com. House of Macadamias is the next big health trend. Get 20 % off your first purchase and a free bottle of cold-pressed macadamia oil at houseofmacadamias.com slash twist by using code twist20. And Notion is now offering eligible startups up to six months free for their plus plan, plus unlimited AI for that six months.

0:49Start saving your startup some money and apply for Notion for startups today at notion.com slash Jason. Welcome everybody back. two this week in startups ai and crypto are basically the two why nows that we've experienced in the last five years one of them perhaps fading here in the united states and the other one soaring with me to discuss it all on our weekly or every other week bi-weekly ai crypto roundtable sunny sundip madra the co-founder of definitive intelligence and vinnie lingham the co-founder of civic and weight room welcome back to the program boys thanks jake all right there has been a lot of discussion about the phases under which ai will be adopted ram alawalia the founder of lomita wealth management it's an alternative asset founded investment advisor he tweeted following prediction on ai one human leverage two operating leverage three management leverage you guys agree with this and did you see this tweet yeah i really like this breakdown because what it it and it's sort of like inverse and it builds on what we talked about the last podcast where we see the leverage in the engineers first then in the operating class and then finally in the management class and i think instinctually when people think about it they probably think it's the other way around but i i really like the way rom laid it out here uh finney what are your thoughts i agree i know rom uh he's a good guy he's very thoughtful about these things um you know he's very analytical as well so look i i think the the big danger is if we try and over analyze and sort of constrain um what is happening to you know like trying to create these categories and layers of what's happening where, quite frankly, it's just a wild, wild west.

2:53Who knows? This thing can go in a million different directions. It can upend all the norms that we, you know, management and company building and, you know, every other norm that we can think about. This is like true disruption on every scale. And Chimart said as much in the All In podcast last week, J.K.L., that you host. You know, he's having to rethink everything, and it's a humbling experience. yeah i i look at this and i think what we're seeing is smart people trying to build a mental model as to where this is going vinnie your point is well taken uh and shabath a great example he's one of the you know i would say one of the great strategic thinkers that we know and people are trying to build a mental model of what is happening right now and where this is going however the pace at which this is going and some of the dynamics as you're saying that are not typical they're unique to this moment in time they are in fact uh black swans perhaps i look at this and when i do my hey here's three phases which everybody seems to like to use the the three phase approach i'm not sure why it's not four or six mckinsey it's a mckinsey three yeah yeah we put them in circles i don't know odd number works better you can charge more money per hour when you do this mckinsey thing um i think about augmentation which he points out human leverage engineers 10xing but i put augmentation for salespeople sdrs uh writers everybody in between um and then automation where you're kind of taking some humans out of the process and then finally you know after augmentation and this sort of automation is you know replacement or deprecation in other words you just don't need certain functions anymore uh shortly after they're automated so some people might be watching the automations which we saw with auto gpt but maybe at a certain point you're like you know what i can trust this thing i don't need to have a human here just like we used to have humans as elevator operators at some point we felt confident enough to just take the human out of the elevator so So what is your thought hearing those two things, Sonny, about where this is going?

5:10Yeah, like I want to build on something that Vinny brought up last time. And I think, you know, we had a good discussion on it. And I think it really, you know, good on Vinny to bring it up. And a couple of days ago, you know, Samuel had a good Twitter post. Nick, I just shared it in our chat, right? which is actually a quote from Adam D 'Angelo, who's one of the smarter minds in Silicon Valley, ex-CTO of Facebook. It really aligns with what Vinny was saying last week and just building on that, which also aligns to that first point where we're seeing engineers who generally never felt this in technology.

5:55I really like the way he phrased it is that every week there's new releases, And then because of that, he has every day has spent hours to learn what's changed to also then, you know, figure out is what they're doing doesn't even make any sense. This is really, really important, I think, in terms of that first bullet. So I wanted to spend just a bit more time on that again, because, you know, last week we touched on it and it was, you know, we talked about what's happening, but to hear, you know, more people bringing that up is really important because that's a framework that I think every person that's building right now that's close to the AI ecosystem, and even if you're not, you should be thinking about that.

6:36yeah is your idea and what you're executing on even valid anymore because of something that's been released are you seeing that firsthand vinnie yourself yeah so i i don't think it's moving that quickly right now i think we still have a little bit of you know daylight uh between when that happens i'll be like when the when the machine becomes um you know self-aware singularity singularity then i think it becomes a little bit more um you know but more more more of a like i say a smaller gap between you know the the rate of change so the rate of change will go up a lot more right now it just feels fast for us but i look around you know the big companies aren't freaking out there's no huge m &a cycle right now in in ai they're not panicking just yet everyone's trying to like place their beds no one's really making money out of it either i mean there are a Yeah, that's, I think, the big catalyst for when things go parabolic, right?

7:41When startups form that start eating the big company's lunches at a rate that they cannot digest or handle, that's going to be when we have this tipping point where everyone goes. Revenue destruction is what you're referring to. Well, revenue destruction and or transference, right? So, and I'll use my company as an example just because, right? So we are a video conferencing platform. We're rolling out AI features. If people just start deciding to use us, and we haven't started charging in, but when we do, at a rate faster than Zoom or Hangouts with a higher rate of stickiness, and maybe we had a lower price, and people just move across, that's when these companies start to panic.

8:21And they go, oh, how do we compete with that? Well, it's AI driven. Well, what does that mean for us? Can we go and add AI? Well, no, because we have to worry about this and this and this existing customer and this revenue and how do we charge for it? and the machines of these big companies are just very, very slow. The larger the company, the slower it is for them to make change. This is why startups always have an advantage when they're building. We can just cut to the core and say, you know, all these features and legacy that they're busy supporting and customers are paying for it. We're not going to build that.

8:47We're not going to offer that. That doesn't make sense anymore because AI takes care of all these things. That's a disruption. I think that's coming. I think when you get to specific examples, it really opens people's eyes. In other words, there's Zoom out there. there is um otter a lot of people got into video conferencing maybe transcripts early and now a new entrant with only i don't know how many people have a weight room but i'm assuming maybe a dozen people or something you know a company with 1200 people you know now there's an offering with 12 people and if they you can charge it you have a different cost basis than they do yeah well exactly like for us because we we cut off video at about 10 now it was he changing it around but like at 10 you know let's say there's 100 people on on a call everything else is audio we're only paying for 10 video streams whereas with zoom everyone has their cameras on and you know they're busy they're busy feeding the you know the video cost i mean the video cost they have to eat or the customer's paying for it right so yeah just as an example right our model is different and that's always that's not what makes us i think different necessarily i think um for us we're focusing on the ai components because we've we've both out a feature roadmap that we're going to you know in the next 60 days probably rolling out four to six features which they can't do in the same amount of time this is not possible uh we're a team of less than 20 people and so the feet it's the feature so right now i call it feature stuffing okay like you got to take products and services that have been built and you're going up against the incumbents you have to stuff it with features way way faster this is what zuckerberg did with facebook in the early days by the way product velocity Exactly.

10:27Product velocity. How many sprints do we have in the next 60 days, 90 days? How many features can we roll out? We've done the two-by-two matrix, and we're looking at all the quick wins, the low-hanging fruit, stuff that we can roll out quickly. So it's not so much about rethinking everything. And I get what Adam D 'Angelo is saying. But I mean, when you're solving problems that already exist, and you're just doing it in a better way, I think you can overthink things a bit too much and focus on all the rate. Like focusing on the rate of change itself means that you're not going to be focusing on building product.

10:56And I always make, you know, whether it's wrong or right, I'd rather focus on product right now than looking at all the things that change. Even though I agree that there could be some redundancies, I'd rather roll out 10 features and two become redundant than roll out two that are really well, because I think it's actually it's a feature race right now. Listen, everybody, when it comes to the blocking and tackling of running your startup, you don't need to reinvent the wheel. Well, CDNs, aka content delivery networks, are the place where startups can really overcomplicate things. You don't need custom authentications or custom codes.

11:29No. If you're a startup, you need to just check out Cashfly. It's a pure play CDN. And CDNs are literally all they do. So they're the best in the world at it. They've been doing it for over two decades. That's 20 years. Cashfly makes CDN simple, effective, and secure. Let me say that one more time. Simple, secure effective effective secure simple simple effective secure you eventually are going to outgrow the smaller ones that are trying to give you too many discounts you want the best of breed so don't burn your startup credits using a cdn at one of those larger players let cashfly handle it all for you they will help you deliver your content faster than your competitors you have to be fast if you want to compete whether it's videos your mobile app games content sass the faster you go with your delivery the more people use your products so go check out cashfly and twist listeners are going to get 10 terabytes free forever when you sign up at twist.cashfly.com that's twist.c-a-c-h-e-f-l-y.com 10 terabytes are waiting for you for free stop what you're doing pause the podcast go to twist.cashfly.com and get your 10 terabytes for free forever yesterday i brought our accelerator and founder university back to in person i went to the city of san francisco and i was in the fi die uh and i had our accelerator class at fenwick and west over in the fi die and then i had the founding university over at wilson cincini's office and 100 people showed up for that and then seven people in the accelerator and the big topic of discussion was i built this using chat gpt or some other language model and And now 4.0 kind of includes it.

13:16And then are they going to just, are users going to be able to just go use ChatGPT4 instead of using mine? And we had a really thoughtful discussion. And I said, you know, this is exactly, Sonny, the moment that happened in app development, and you had a very famous app development company in Canada with hundreds of developers. If you built something that was of low value and easy to replicate, like a flashlight app, like a calculator app, uh those things were 99 cents people bought them there were you know there was a lighter app i mean people had all kinds of silly apps that they built and they were able to make you know a hundred thousand dollars or 250 thousand dollars but shortly thereafter you had to make something thoughtful uh if you want it and something subscription that people could kind of work on and so i told folks there was one um that was making a virtual travel agent and i said you know there's things that chat gpt4 can do but there's this entire feature set that they're never going to get to and customizations uh that you can do here and i don't want to say them out loud here and give people a roadmap but is that not a good analogy or is it a flawed analogy sonny the one i gave about apps in the app store i like it but i think there's one massive difference here and again, this topic's come up a few times, which is there is an element of having access to unique data when it comes to building something in and around AI that's very important.

14:45And that was one thing that you didn't have to think about if you're building an app. And so there's like the feature angle, but you could build something that wasn't quite the flashlight or something like that. But this whole notion that you have to have access to unique data, that's a new layer in the thinking around how you can differentiate yourself. Because what I refer to OpenAI as is the apex aggregator. So if it's anything that can be aggregated, it will be like the top of the food chain, right? The T-Rex, the great white, whatever you want to call it, right? And so I think - Like Google indexed the world's information.

15:25sure but in in but what google always promised to and it continues to still do um and you know this really well but you know you were impacted by this in mahalo right which was yeah they promised at least drive traffic to you uh when open ai becomes the apex aggregator they necessarily can um they don't necessarily have to drive traffic to you they can connect to your api and they can get whatever data they need or perform whatever action they need without ever having sent someone to your service. And so unless you have something that's really defensible, you're going to be at risk. I didn't see that in mobile.

16:05And that kind of defensibility is much harder to achieve than it was in mobile. So let me just ask you, Sunny, what do you think about the you know the difference so open ai yes apex um you know what do you call apex aggregator or aggregator yeah impacts language model yeah fine so so but they're not the only one there's there's you know there's a half a dozen of these out there that are actually pretty sophisticated maybe not at the same level maybe not the same development team when companies are busy building out um and google's got one as well which is actually pretty good etc like how agnostic do you think developers should be so in other words like when you're building features like we're building features it's it's transcription it's summaries etc those features are agnostic to the model we use so if we find that you know google makes it a better product we can switch across to it so like how do you think about that well i think the ultimate pattern right now at least that seems like it's going to play out is that you'll have like sort of a main, let's say LLM that you interact with.

17:16And that LLM will have a series of agents that it interacts with. Some of those may be LLMs themselves, and some of those agents just maybe more traditional, just pieces of software that interact with an API. And I think the interesting thing about OpenAI right now is that they've made that jump before anyone. So they have this plugin ecosystem, they have a whole bunch of developers submitting to it. My guess is they'll have thousands when they launch, right? So it won't be like the App Store. It'll be even App Store day one, where it's just a few apps when it came out. And I think what's quickly going to happen is we're going to see a disruption.

17:53So kind of to answer your point, Vinny, it is a little bit of a time to pick a winner and try to align to that. Now you have to have some flexibility in your approach. But if you look at where things are at today, there is a distinct difference, right? And we know in our group chat, we were testing a quick example of one of these open source ones. And, and J. Cal, you know, you had done, I did the same question that you had put in the other ones a few weeks ago, right around, hey, give me a list of companies. And it's just not as good at it as OpenAI is. If you love snacking like I do, you got to find the perfect snack.

18:31It's not easy. you want one that's going to remove cravings that fit your dietary goals regulates your blood glucose levels and gotta be something you look forward to eating and you know what i love i love macadamia nuts what a treat they got so many health benefits i'll give you the highlights right here they are the only nut that is rich in omega-7s that's linked to natural collagen production reducing inflammation stabilizing your glucose levels and healthy fat metabolism and the folks my friends at house of macadamias are obsessed with making the highest quality macadamia products available chocolate dipped macadamia nuts are my favorite but i also sometimes i get that coconut craving you ever get that they got a chocolate coconut macadamia bar that is amazing and it hits the spot it's not going to spike your glucose it's something healthy but it's a treat it's a treat and they just launched an amazing cold press macadamia oil with a buttery flavor and a high smoking point perfect for drizzling over meals or cooking use the code twist 20 get 20 off and for a limited time they're giving a free bottle of premium cold pressed extra virgin macadamia oil with any purchase go to house of macadamias.com slash twist and use the code twist 20 house of macadamias.com slash twist and use that code twist 20 so they know jay cal sent you i do think there's an opportunity here though for a new business model uh back in the day uh in france they had a little communication system before the internet where you could look up restaurants and it was called minitel and minitel uh if you remember charged by the minute and it was an information service and you and a compu serve prodigy and originally aol they charge for the minute for access to data i think you know with the lawsuits that are coming with reddit twitter quora etc uh getty bloomberg and others new york times i'm sure we'll be right behind them saying to people like chat gpt4 open ai hey you're gonna have to pay us to use that data a new business model emerges that will make everybody a winner except for advertisers so if you build and i'll just take the travel instance and i want to uh have a travel service that's built on chat GPT for well when I hit the plugins and I hit Yelp or I hit travel and leisure magazine or a blog about travel whatever it happens to be in order to use that data in my model I pay them a fraction of a penny then that means we could invest in content and somebody like the New York Times somebody like Bloomberg even a blogger if their data was used in some way they can and specifically through an a through an api call i'm sorry with the plug-in call it's super easy to give attribution this could become the new model for the internet and it would be as if you know re micro payments were built into the browser right it was just something brave's been trying to do and crypto tried to do and we should have had micro payments but this could be an incredible new model and of course advertisers did you see my tweet jay no i didn't is it parallel thinking yeah yeah yeah i literally said uh i'll post a link here first but um a couple days ago i mean perhaps it's obvious but uh yeah we just haven't had this business model ever licensing was something that existed before the internet if you wanted the ap story or an ap photo in your newspaper you signed a licensing agreement and that kind of went out the window i mean blogs did have to pay in some cases to license photos and and the big ones we have like a we have a a you know a monopolistic version of that in apple news right when you pay for apple news as part of your apple one you get access to a whole bunch of these things because apple is yes you know further paying those publications right you're kind of getting an aggregator but what i'm saying more is you need micro payments though that's the thing so so how do we how is micro payments it's like machines aren't going to pay with wires and bank transfers and so it has to be a digital currency is it going to be a central bank digital currency is it going to be bitcoin is it going to be lightning is going to be you have your credit card on file the user wouldn't be required to pay it the service provider would so the service provider background would be paying no no no no it's not gonna work because this is this is the problem because micropayments or visa massacre there's like a 30 cents charge minimum per transaction you can't do micropayments when when machines are doing millions of hundreds of queries across different vendors for different people it doesn't work no no but for a new startup to buy a hundred to buy you know five hundred dollars a month in credits like i pay now i have my credit card on file at open ai to use the sandbox and my whole team has access and as they do stuff it just starts building that's all i'm saying reconcile at the end of the month so that's what i'm saying is the new startups yeah that want to leverage this they would be paying as they go just like they play for cloud storage they would be paying for access to to pools of data and we could build an infrastructure behind the scenes which had gpt4 your your credit card on there that bought a thousand credits you know a thousand dollars they would just handle the micro payments in the background uh at every month reconcile them so yeah okay so that's assuming that when when the the they that open ai would have deals with all these data providers for example so nick you pull up my tweet quickly so i put this out a week ago and you know everyone's been going on about web3 being um you know you know well web3 like being crypto maybe it's not crypto it's ai like the next evolution of the web is actually ai and not crypto crypto may be a sub segment of it like the payments rails or whatever but the web3 isn't about decentralization and web 3 isn't about these things which you know i guess we've created this whole notion we think it is maybe it's just the web evolves and instead of using google search using chat gpt using i think this is right i think i think this is right most people we've had three web 3 so far the semantic web people claimed was web 3 yeah crypto they claimed it and now ai is claiming it and and actually i think some people in the vr space were kind of trying to claim web 3.0 yeah but i think maybe third or fourth time is a charm here go ahead sonny you had something to say well let's combine let's combine a couple of ideas here and this kind of brings a podcast together so i think someone both of you guys know is grimes yeah right and yesterday grimes put a tweet out maybe nick can find it's quite easy um that said she is willing to split royalty 50 50 with anyone that does an ai um like a remix or whatever the right term would be called a ai so like even a new track you could create a track with her voice you could write the lyrics you could do the beats whatever this is a tweet exactly yeah so this is it right here and she says i'll split the royalties 50 50 which anyone that uses her voice and the voice will be you know through one of these ai tools right that allow you know we've seen the drake one and a bunch of other ones now recently so in order for this to pull off now vinnie i'm going to throw it back to you right what is the best way for her to pull this off leveraging so obviously there's an ai aspect of this which does the voice but you know how does this all reconcile you know in terms of payment and and tracking and everything else so because i think you'll have a really good answer here okay so interesting i'm actually just messaging my friend right now um because he is a very famous dj and he's going to do something with this um and i don't want to say who he is yet but he he looked at this tweet and And he was like, I was chatting to him yesterday.

26:18And he basically is going to take Grimes' voice and remix it into whatever, some DJ track, and then do the 50-50 royalty with her. And he's a big name DJ. He's like, maybe not tier one, but definitely tier two. And so I thought that would be really, really interesting. And I think that's what it needs. It needs another high profile. like i don't think that some guy in a basement is going to remix grimes and create this amazing track to often maybe you know maybe if enough people do it it'll happen totally possible yeah it's totally possible oh no if a thousand songs come out five will be sure sure sure sure sure i mean who's produced a thousand tracks in a year there's going to be a thousand grimes tracks out next month yeah um 900 are going to be terrible exactly 100 of the 100 that are left they're going to be good greater bangers but i'd be betting on the i'd be betting on the dj with a big name with an existing with an existing audience to go it's permissionless collaboration exactly so that's my point like someone i know an artist in the basement might make a good track but no one's gonna know about it if there's a thousand others will be drawn up and those grimes retweets it but maybe maybe or she adds it to her catalog on spotify like if she uploads it by distro kid to her spotify then that would then she would be managing the royalties and it would be official it would be on her spotify and the you know whatever they call those playlists like this is grimes this is the rolling stones like she could actually feature it if she approves it if it hits that sort of level but you were asking yeah sunny the question of how should these royalties be done and so she answered that she said she wants to do contracts smart contracts well yeah well that's that's kind of what i'm saying is a smart contracts for that right so when you create it so that no one else can rip yours off right i think that's where it kind of intersects back with the blockchain and someone says this smart contract this token this is the version that i created and here you go right okay exactly this is the tweet uh talking with my team if you register music with us we can collect and pay out royalties direct to anyone who uses ai grimes vocals using smart contracts the future really is now this is so cool working on a way to register and easily access vocals through elf.tech um so yeah there's gonna i mean brave new world all right you hear me raving about notion all the time we run our company on it and now notion lets you leverage really powerful artificial intelligence my guy presch works for me he takes everything we do when we do a tactical talk at foundry diversity or we have an internal meeting and he'll take the transcript right we cut and paste the transcript into notion but then he will highlighted.

28:58And then he uses the magic of Notion to create bullet points and a summary. And this is incredible. What would have taken him an hour or two of work to summarize a meeting? It's done in seconds. I mean, you're starting on third base. Right now, great news for all of you. Notion is offering eligible startups up to six months free of their plus plan, plus unlimited AI for six months. Notion is going to help your startup. You're going to save money because you can consolidate many of the different tools you're using. It's going to make you super efficient. All the infos in one place. Notion is becoming more powerful by the day and you need to get in on the action.

29:35Notion is now offering eligible startups up to six months free for their plus plan. Plus again, unlimited AI for six months through the Notion for Startups program. You start saving your startups some money and apply for Notion for Startups today at notion.com slash jason when you use our link you're supporting the show start saving your startup some money apply for notion.com slash jason at the same time our bestie chamath uh wrote a blog post and uh in his blog post he basically it's basically a rehash of um what he said on the all in pockets nothing actually new here but he basically says these are the main arguments against regulating ai and why he disagreed with them it's premature to regulate something that barely exists and he writes mark these down as famous last words while it's admittedly early in the life of ai and generative pre-trained transformers progress is compounding at a rate that's measured in days let's take that first one do you agree where do you stand on this chamat's positions or that it's premature yeah i'd love to hear regulation or free markets right now got to pick one i'm i'm for free markets free markets vinnie i'm the same i think i think it's premature trying to regulate this right now because all you're going to do is you're going to stifle you're going to stifle the the range of potential end results that we could have it's better to like run all the experiments see what works see what doesn't work regulate the stuff that doesn't work than to prevent, then they put regulation in and prevent the good stuff from ever coming in the first place.

31:17Because some of the good stuff may be really good for society. That's in my take. And if I can just add to what Vinny just said, like, you know, some of these, the comments are interesting, but, you know, AI in different forms have been around for a very long time used by a number of companies, whether it's drug companies to create, you know, compounds, right? You know, DeepMind years ago, you know, solved one of those protein folding problems. and there wasn't a problem. I think the issue that we have today and where I struggle with terms of Vinny's point there is like, what is the scope of it?

31:48Are we now saying this applies to machine learning, right? Because that's been around for a long time. Does it apply only to generative AI? Does it only apply to a generative AI that comes out of things that leverage transformer technology? Where do you bound it? Right. There's a definition that what is self-driving as but one example is self-driving adaptive cruise control lane guidance or full self-driving they're having that discussion there as well and you know my position on full self-driving is if it's a beta and you agree that you have to be paying attention and they have a camera on the inside that's watching your eyes and it shakes like on my full self-driving beta i am now at four strikes out of five and i am super vigilant and i i got all four strikes for the same reason which is looking down at my not my phone but my navigation and it just i didn't realize how sensitive it is oh no actually a couple times was also because the steering wheel so there is self-regulation that needs to occur i think self-regulation is the way to go here so i'm gonna well well let me let's just add to that jerry i i think when you say self-regulation it doesn't mean that the industry didn't regulate itself necessarily.

32:59It also means regulation by enforcement too. So what I mean by that is I'm actually very interested to see what happens when all these artists and people who provide data to these AI tools go and sue and go and run lawsuits and run it through the legal process. Which is fine. That's how we got DMCA. That's how we got a lot of the sort of foundational laws for the internet was based upon litigation. so let's test let's test our assertions of freedom let's test our assertion of creativity and ip and ownership etc and let's see where the where the courts lie because it's pretty pointless i have a counter to that i have a counter to that okay go ahead i understand your position which is there are existing laws you're taking there are existing laws position people break the law it's on them and that is completely true so if a new type of gun comes out and it shoots 100 bullets a minute and you know and you know instead of one bullet you know if he shoots 100 bullets a second versus one we were like hey well the person did the murdering but then we did say you know what we're going to have some limits on magazine size in certain states we're going to have waiting periods in certain states we actually did look at some of the nuance and machine guns automatics were not allowed right and so when we look at ai i would encourage people to think well what is the equivalent of the automatic weapon versus a weapon that is you know you have to pull each time in order a semi-automatic standard rifle or just a rifle standard rifle etc and the magazine size so you know that is one way to look at it another way to look at it is the mpaa hey people are taking their families to the movies what's a way for us to let people know hey somebody's going to be murdered in this way in a comic book way or in a horror sense where they're literally going to be decapitated and it's going to be blood everywhere one's going to cause nightmares and one is going to be a comic book killing where somebody turns into a pile of dust and you don't see anything right these were common sense things some of them became laws other ones were done by the industry i think the people who are providing these services will need to start thinking about um people being registered to use these services and audits being done on what they're doing.

35:08Now, audits are being done right now on what you do on AWS, if you want to steal movies and put it on an AWS cloud, you will get caught, right? They're fingerprinting stuff. There's a fingerprinting system for child porn on AWS, right there. I don't know what that's called. But Google has and they've sort of look for horrible things. I think here, the AI company should start thinking sooner rather than later about the auto GPT is because I could see a scenario where somebody sets an auto gbt up and i brought this example up on uh on all in um where it would start hacking people in a very personalized way so instead of just you know asking people to log into gmail it could be let me find your friend graph let me write personal notes based on tweets that you shared with somebody and automate a million phishing attacks as but one example how do we have the platforms think about that sunny and self-regulation you heard vinnie's let the laws handle it you heard me give some examples of self-regulation where where would you uh land after hearing those two descriptions well you know i i again just put my technologist hat back on and i go look anytime these things happen you have to fight the technology with more technology.

36:29So, I think for every AI that will, you know, potentially exist to do something that's harmful, there'll be AIs that are there to protect you. And we have those now, right? Those are the things that run in our spam filters and, you know, on our phone. You know, I don't know if, you know, Apple has something that runs on your phone, even without your permission for child pornography, right? That they report if, you know you take those pictures and they end up in your uh photos which i think they they were the scanning of people's personal clouds they stopped doing that people felt that was too intrusive but if they put it in the cloud maybe they did i'm not sure where they wound up they do it on device yeah but but coming back to that jaco that's technology that is using ai to you know combat bad things that people are doing and so i just come back to there'll be a set of companies that are out there combating the auto gpts that are trying to hack people and and you know write fake reviews to sites and so i think it's just going to be even a larger technology battle on both sides which we've seen so for every spam originator there were there is there's multiple companies that fight spam there's multiple companies that provide firewalls right like there's i mean you probably even had them as sponsors at some point jay cow right they're very standard in airports and stuff like that.

37:52And so here's sort of just the way the technology cycle works. Let's talk about another argument that Chamath heard, and that he rebuts in his blog post government doesn't have the credibility or track record to effectively regulate technology. He says experience suggests the contrary in a well regulated industry like drugs and pharmaceuticals, for example, the Food and Drug Administration, they maintain a tight hold over what can be sold in order to ensure safety of customers so do we believe the government analogy here for the fda is analogous to 100 million people using chat gpt i think there's a scale issue here what do you think vinnie yeah i know no no good son no i was gonna say i think it's a scale issue i think the government can properly regulate anything they want to it's just a matter of can they keep up with the pace of the you know the innovation i think i think here's the problem jake we're at the bottom of a very very steep slope that's approaching for for the amount of change and disruption and whatever else that's gonna happen ai and when regulators start trying to do stuff now or like the government starts looking at it now what they by the time they like take stock of what it is today and try and process it and figure it out and create a law this has changed so much in the next six months 12 months they have to start from scratch again.

39:11So that's another good argument from my side, why I think it's just, government just needs to sit back. I don't think they can do much. The other thing is, we're also hitting computational constraints. I mean, all the big cloud providers are hitting the wall with GPUs, running the models, the training models, et cetera. That's an extraordinary point you're making, Vinny. When's the last time we saw technical constraints in our industry? it was dial-up speeds yes um storage costs and storage costs and storage costs right so those were the last two massive hurdles and then before that obviously there were some cpu gpu issues of running a movie on your laptop right it would yeah so we had cpu gpu issues in the 80s and 90s we had bandwidth issues in the 90s those went away in 2000 uh we had storage issues up until dropbox which would be 20 2005 2010 we saw it yeah yeah youtube like we youtube and dropbox work and box were companies that were built i mean it was free or close to cheap so for me it was um the speed of connection then it was the cost you remember like it was very expensive uh and slow in the beginning um and then the storage like just streaming watching youtube etc even the even like netflix they were spending so much money on at one point on losing money back in the days on the cost of just streaming right now it's now it costs nothing to stream like you can download i like i download 80 gigs for a modern warfare so here we are again no problem here we are again with a constraint so if we were to look at a big constraint which is these nvidia what do they call them g100s or something well not just you are just like a100s yeah it's not it's not just a100 sorry it's not just those chips it's any it's any like we have a shortage of gpu gpu compute power across generative ai uh in ai there's lots of different chips my point is like not just ai chips we just have a gpu shortage that's emerging worldwide if something has to happen to tsmc taiwan semiconductors okay if if china and america go to war with over taiwan or whatever if any of that stuff happens in the next 12 to 18 to 24 months we're going to be screwed we don't have enough compute power we're not going to be able to make pictures of david sachs as batman yeah i'll give you the other side of that though okay you know i started my career in semiconductors and so the the cycle in semiconductors only two years so at any given point to get to the latest and greatest technology it's only ever two well you have to have a fab up right you can't you You can't not have a fab.

41:56But those things have existed in the US for a long time and they can fire those back up, right? And so to get to fab capacity of the latest technology, it's always just two years away. And the companies that provide all the technology underneath to make that happen are all, you know, mostly American companies as well, like applied materials and whatnot, right? And so I think you can't fight it right away. But given that this rhetoric has been happening for the last year to six months, I think we're only, you know, six to 12 months away from having, you know, start having start of capacity ourselves.

42:34Well, Amazon, I think I saw was actually building their own competitive chip to the A100. Well, and Elon at Tesla is building his own platform as well. So, you know, when you have constraint, it leads to even more innovation and more players going after this. i've never seen you want to talk about things moving quickly i mean when did we ever see the government push through a regulation and a spending bill at the scale of the chips act they are in full-scale panic mode over the possibility that we could have chips be constrained because of this taiwan situation and one wonders if the pace of ai combined with uh you know the availability of chips is actually pushing this issue onto a different timeline we're seeing a geopolitical and just letting this sink in for a second we're seeing in geopolitics the possibility that a war over a constrained chipset for ai could be the next world war i mean i don't want to sound like hysterical it's very possible in fact i would actually say that it's more likely than not because what's going to happen in the next so it goes back to my revenue destruction transference argument the moment we start seeing revenue moving to startups from traditional companies and then these companies which are at scale have to implement ai and need the gpu compute power and they don't have it like this is where the choke point comes in right right now the people using ai by and large and tell me telling you i know you're more on the ground in san francisco these days than i am like it's not it's a it's a it's a cottage industry of startups and small enterprises it's not the big guys right no it's a it's a it's everywhere like i'll find it sure like uh you know morgan stanley just uh put a little presentation out earlier today i'll find it and send it in the chat in a second um it's it's all over the place i think are they deploying at scale my point is are they deploying like inference for example um you know with these models like are they are they running like are they running it at a high enough capacity or utilization that it's going to cause strain on the gpu infrastructure that they've got i don't think so i don't think it's there yet oh i mean i think we we saw it with azure i mean they when p they're limiting chat gpt4 searches if you have a paid account to 30 every 25 yeah so i mean they're if they're throttling it there's a reason and i think the reason is this system these systems are being pushed to the brink because of so much innovation but we'll catch up i mean this is happening this is not the first what i'm saying is what i'm saying is like the corporate america hasn't fully adopted ai yet we're talking about that's the crazy part maybe maybe one percent maybe if that the rest are all sitting on the sidelines and the moment they need to where's the compute power going to come from we don't have enough gpus out there i told everybody in all my companies uh or the two companies i'm running to get on chat gpt4 to expense the 20 a month and i invited everybody to get on the play around at my investment company i had gave every one of them access to my you know credit card account back thing and i said i want everybody to do every job you have start with a chat gpt4 sir and then go and back and do your job and then report back with screenshots and And I'm deadly serious about it.

46:00If anybody on my teams, because they listen to the pod, if you're not using chat GPT for every day and testing it on everything you do, you're not going to last at a company I work for period full stop. That would be like showing up to work without a computer. You show up to work for me without a chat GPT for window open. You don't work for me because that's how much I believe this is going to transform stuff. And as, but one example, uh, you know, we're raising this fourth launch fund. And I was like, I wonder if there's, lp and institutions i don't i don't know and we're having this angel summit i was like let me invite 20 lps to come to the summit to balance out all the vcs who are there and they like to meet vcs and vcs of course like to meet lps so i start pulling up lists and um i start doing a chat gpt i put it in a table and then i open up my window over at my sandbox and i say hey for this foundation i've never heard of can you give me the twitter url and the linkedin url of anybody who has the word partner uh you know in their thing and it of course knows linkedin i guess because microsoft and chat gpt and open ai are so connected and they own linkedin and then i got the table i was like this is a research project that i would have spent thousands of dollars and weeks on and it's done instantly now i don't have auto gpt and i'm not a developer yet with and one of the guys from open ai told me they'll put my account on the plugins but i'm going to plug this in to my gmail or zapier or something and i'm just going to send a chat gpt message that says dear blank this is my ai it's jason calacanis this is my ai coming from my real address i asked my ai to tell me people who might be interested in coming to the angel summit i see you're connected to these other speakers who are already coming here if you're interested in coming let me know and if you hit reply you'll actually get the real jason this is jason ai how hard is that last piece to do sunny because i'm not a developer you know what jaco we should just do as an experiment we can sit down for three hours and do it three hours would be a stretch right it's all doable right now all right i'll come to your i'll come to your office on saturday i'll buy you lunch um it'd be cool to hang out i mean i i was just thinking i want to be a developer again that's how inspired i was by of this i was like if i do this that's literally a full-time job at my company i don't have to hire well i was thinking of hiring somebody to do relations i don't need somebody to do relations if i've got this ai out there doing this for me it's 10x jcal that's what you get it's a 10x jcal networking yeah again this is this is why i think the benefit is going to accrue to um the startup world at a much higher rate because because if you have we have i think eight or ten engineers if i can 10x that i've got a hundred percent effective team right but you precisely but a company with 10 000 engineers what do you do with 100 000 engineers you can't do this you know i'm saying there's they start laying 10 000 people off exactly what you start doing which is why zuckerberg just did i did he do a third riff did i read that correctly he did like he did 10 10 and then another four four yeah wow 24 000 and a hiring freeze he's getting fit and people are see this is the thing about the sky is falling um you know people saying we're saying as technologists the sky is falling and that we're overreacting when people say we're overreacting and i explained to you the story of running a venture firm and that i was going to hire a human being to meet with lps for me potential lps and invite them to events and try to build relationships with them.

49:43And I, in an afternoon as a non-developer, figured out, hey, I can get halfway there and Sonny as a developer or any other developer hired could get me the other three hours there. And I 10x myself, I don't need to hire that person. Anybody hearing my voice and hearing that example should be thinking about their own job and how much code it would take to rewrite and automate their job. I mean, just speaking of code rewrites, you realize you can give an entire code base to OpenAI and say, rewrite it in a different language or clean up the code and write it better, remove all the bugs, scan for vulnerabilities.

50:23This is what makes a 10x developer. You can just basically, with very little coding knowledge and experience, as a junior developer, you can do the stuff that senior developers do because this thing is doing it all for you. It's quite amazing. You know how many times, like, I mean, microservice architecture has changed the way we do development over the years. But 20 years ago, migrating from one stack to another was, it would take months to rewrite the code. You could, you know, now it's just like, we open up an API and the system's talked to each other. But remember those days, Sonny? It was insane.

50:50Like, using JTME and, yeah, like, whatever. Like, whatever stack you were using, it was a joke. Have you guys used ShareGPT where you can kind of share your thread? Share your prompts, yeah. You share your prompts and the whole thing. i find that super fascinating um and so i basically shared it on twitter and i asked people like what they would change and i had so many different developers contact me like yeah i could take this to the next level for you and i'm like wow this is super powerful sunny you uh also shared with me in the group chat a link to an open source chat gpt competitor you referenced it earlier what's the name of it and how close is it to why should we care how close is it to chat gpt 3.5 or 4?

51:34Yeah. I'm not in a position to kind of benchmark it because it literally just released a few hours ago. So Hugging Face is actually like a platform where they collect lots of different machine learning models and kind of beyond just AI and the way we've been thinking about it the last few months. And so they've launched something, I believe it's called like Hugging Face GPT or something like a chat um let me pull up the exact name and why this is important it goes back to some of the origins around what people have been talking about with respect to safety and transparency so it's an open source model everything is open and so we can see how it's been trained we can see how it operates we can see the code for it and you know go goes back to your friend Elon and why he named open AI, open AI was meant to be open for the safety and benefit of humans.

52:31Right. And I had shared a tweet a few minutes ago with, with Nick, if you want to pull up that one, where it kind of intersects this part of the conversation or previous one, where he goes on about like, people just don't understand sort of the power of, and this where I kind of a little bit of a different view from Vinny is that the AIs are already powerful because folks like elon who've been thinking about this a lot longer than any of us are already saying it and so um and he's like you know right on like i've seen a few technologies develop but none with this level of risk and you know smart humans have trouble imagining um something that's vastly smarter than themselves and i think he's gotten to the point and maybe you know jake how you can comment here where he believes that these things are already smarter than us i was having these conversations with him 15 years ago um when uh super intelligence that book came out uh and sam harris myself and elon were talking about it over dinner and sam harris was new to ai um but elon was absolutely convinced because what people don't know is elon was on the board of and a major investor in deep mind he begged the deep mind company to not sell to google to remain independent he offered them the funding and then he wound up giving sam altman the funding uh to try to make this open source version to try to get the world to see what was happening and make it open source and of course um sam decided to pull the old switcheroo and took up what was the mission to be non-profit and open to for-profit and closed that is a story that remains a mystery i think to the world as to what happened there i am i have my concerns uh if i'm being totally honest that open ai is not sharing the source code that to me is very much a tell they said this technology is too dangerous to not have it be open source then they said it's too dangerous to make it open source what does that mean what does that even mean does that mean they have discovered something that is really truly scary or they want to maintain a lead and make a huge profit off of it you know you you could take the cynical sam and the team just want to make a ton of money or you could take the virtuous and believe them hey it's too dangerous for people to have what do you guys think i i i am struggling with this and i know sam i mean we're not besties but i know sam relatively well what's the actual motivation here is it capitalism or is it that they know something that we all don't and they are scared to share it the source code what do you think occam's razor i would side with the i would side with yeah you know occam's razor is like most obvious answer and the most obvious answer in this case is maybe musk's razor which is it's what he's saying is that it's too dangerous and that's why i would you know for someone that uh like you said has been around it a long time uh really devotes his life to you know expanding humanity um through lots of different ventures this is one of the more important ones because there will be no reason to be an interplanetary species if uh the you know some form of agi ensures that we are no longer a species before we can get to the you know the interplanetary stuff so i would just side with him i'd go with the most obvious answer which is side with him he's been looking at it thinking about it a long time that's my my i i mean i'll have a slightly different take jaco i think that okay okay you know occam's razor uh would basically you know for me it would point to the fact that there's possibly you know ideological um decisions that are made that they don't want to be public so you know it's it's the majority of the team decides that you know we don't want people to know that we're doing this or that and that's the simplest explanation is that they don't want you to see what's doing.

56:29I don't think it's necessarily, oh my gosh, we've created a monster. I think it's just like, we just don't want anyone to see the way we make decisions and... Therefore, they can make more money or have a more defensible operation. For whatever the reason is, it's just like we just don't want, you know, whether it's capitalistic... I wish they would tell us. I think this is where Sam needs to be a little bit more honest with everybody because he's been on this press store anyway. So just tell us, what is the story between the nonprofit, for-profit, flip? Maybe he's concerned about liability. Going back to the Chimant's letter, this is where the government could step in.

57:05They could have a hearing, and it could be open or closed, and they can ask. Yeah. I think this is so acute in terms of job disruption. I don't want to say destruction, but job disruption. Or competitiveness, and chat, GP, OpenAI's lead. that the government asking a couple of questions um if open ai is not going to be transparent about it i think it's not a bad idea and i'm not saying that because i'm kind of trying to cause trouble for them but it's just too weird that they were open source and now they're closed no it's again they were for they were non-profit and they're for profit it's just i've never seen that in the history of our industry if let's just assume okay for argument's sake this was started in in florida and it was a trump organization and they were close source you know and what what would you assume if it was that for profit i think trump only cares about profits yeah do you think that well let's say it was fox news as well let's say so let's say fox news started an ai company honestly i think fox news is the reason fox news picks a side is because they realize that you're going to get more ratings it's better to appeal to one side and get ready knowing what i know about media having spent my career in it i i know the exact cynicism of fox news because they're a liberal when it comes to the movies and tv shows they're making and then when it comes to news they are incredibly conservative because that was a great mode that's my point to make money it's conservative versus liberal like so so do you think the majority of people who work at open ai are liberals or conservatives well obviously liberals but i think a bunch of them are libertarians too but you know so in that they're capitalists well in that vein in that vein the question i guess is are there is there a liberal um you know agenda no i don't think this has to do with politics i think there's a profit motive here maybe it's either a profit motive or there's something in the code to your point that is either concerning or worth protecting got to be one of those two things but anyway we're going to find out i just wanted to show this example um here's the share gpt i did it's got 3300 views i i shared it on my twitter as well as internally but if you take a look at the share gpt i did you can pull it up boys uh producer nick and uh you'll see it basically gave me the linkedins you know and you can see you know i'm playing with it i'm asking it to add more if you scroll down you'll see um uh you know me putting stuff in a table and then asking it for more and more endowments and doing the linkedin but then when i went to the sandbox and i started playing with that so here's the playground image you know stuff is getting this is when i think about you know like a research position or what people's entry-level jobs are you know if senior level people can do this and i just said please give me the urls of everyone with the word partner in their title at the ford foundation that's a big endowment uh please provide url for the linkedin page profile page on their website and the twitter profile put that in a table and sure enough like here's a couple of people who work there and i got everything i need this is a this research that i'm demonstrating here is weeks of research and i haven't glued this together to put it into a database or make automated emails yet or dm them or at mention them i mean there are so much more i do here if i just knew how to write a little bit of code you know i want to i want to code coach once you teach me how to write code what is it replit would i do this in replit is that the company or the platform i would use to be honest sure yes you can and uh it's a wonderful but like once you get plugins enabled you can just do this all within the the uh the plugin architecture kind of chat gpt uh you know interface because it can write code now and it can run it for you right there yeah i uh a senior level person at open ai told me they would send it to me so i gotta just i don't know i mean we gotta log out and log back in and see if oh i have plugins thank you to my friends for moving me up the list i have plugins okay well i know what my afternoon is going to be like uh if you're working my companies and you're not using this every day you know you might not be working for me for much longer because you got to level up everybody i'm i'm looking at this and i gotta say having come from you know the editorial background at inside dot com i'm trying to find a developer because i did summaries of newsletter stories that we do like you know we kind of do bullet pointed news kind of summaries and i would say the chat gpt four ones were not great uh but they were within 70 and then i was like well compounding that means it's going to be 110 this year and uh so that doesn't mean i don't think we need the writers what i want the writers to do is do a level of curation and insight on top of what chat gpt does and so i'm just it's got me inspired to think about so many different ways to you know take the newsletter business take the investment business and just level it up crazy jake i'll think about this right mahalo yeah human power search but all of chat you and chat gpt could recreate all of mahalo it's crazy i had 100 people working there that's what i'm here in right you and just sitting in front of a prom could recreate an entire company from not that long ago yep it's bonkers and you know when i talked to sergey about mahalo and you know like this tiny competition we might be at some point for google he said he really liked what we were doing and marissa also liked it and um so did megan smith they just thought eventually this would be done by ai and i guess they were right uh eventually it would be and here we are 15 years later 15 years what were the top five mahalo pages it turned out that i put yeah yeah it's a great question it was very weird we had started to do some how-to content so we were just we were trying everything we did coupon pages those were the highest monetizing so we had said like that beth and beyond coupon right and then we gave strategies for getting the coupons pages that had them how to get them and then just you know some links and some content then we did pages on how to cook salmon and at one point how to cook salmon we were the number one result in google it turned out a lot of people are typing that into google and um but it didn't monetize then i started making videos because people just wanted the answer of how to cook salmon and so we just started adding more answers to the page and um you know i it was a very good lesson for me because we built a q a system we had built a wiki system before you know after wikipedia but before quora and um you know the real issue was the business model we were dependent on google for search traffic and then when google saw that we were answering people's questions they cut our throats and uh you know i think that's the antitrust stuff that you know a lot of us had to deal with you know e-how all the different how-to sites wonder whatever what was the site about.com that the new york times oh yeah we all basically got cut off at the knees because google saw us as a threat and matt cuts lied to my face we don't have partners and then i had the other side of the business saying we are partners because we're making 10 million dollars in google and said's together you know when you start making 10 20 30 000 a day in google and sense you can be sure you got a pretty good partnership going and people were having me over to hang out at google and then matt cuts wouldn't talk to me and so i emailed sergey and i talked to sergey and i talked to marissa and i talked to megan smith and they're like i wish we could help you but you know the algorithm people they don't allow anybody to talk to them i'm like you don't allow anybody to talk to them and they basically wouldn't give you an answer so hey it's a good lesson for everybody you gotta you can't have a dependency on any one company whether it's google or stripe or you know look at buzzfeed and facebook yeah you know buzzfeed is like i i don't even know if that company is going to be here next year like they're making two or three hundred and rev 200 300 million dollars i think the company's worth 100 million dollars now they just laid off the entire news team yeah as soon as they didn't they just shut down or what they shut down buzzfeed news but they still have like the listicle stuff i think still got it but they were an seo they weren't seo driven they were social driven so oh you you know you want to share this on facebook and they would siphon traffic off for facebook until facebook was like you know what this content should live on facebook not your website you know so just do not be dependent on anybody and i think that's i think why i think this is so inspiring for founders is that you're going to be able my belief i think we'll end on this one is we're going to see a hundred times the number of startups yep that's what i think is going to happen i think this is exactly like when y combinator and cloud computing angel list open angel forum my syndicate that moment in time many more people could build something because you only needed five people and a million dollars to build something because of cloud computing and then it became well you only needed like five people and 150 000 i think now you need two people and you know 25 000 yeah you know or 250 000 which means like a very niche website or a very niche service you know uh an app for snowboarders right we use the the app slopes right it's a i think it's probably a couple person team well there could be a version just for snowboarders but nobody's taking the time to build that because it's only going to make five million dollars in revenue but if one person can build it and do customer support and everything that's like a pretty cool business and maybe 10 people can make 10 apps instead of 10 people making one app so i think that we're going to see like a long tail of startups and incredible software products because of this i I think this is going to create a revolution in startups.

1:07:20I couldn't agree with you more. I think it's already happening. It's already happening. Yeah. The one thing that needs to flow is the revenues, right? So I keep hopping on that because otherwise it's just a, this is an expensive business. I mean, the cost of running these models, the GPU costs of running this stuff, even the open AI costs, it's not cheap. If startups is producing stuff that just moves, you know, bits and bytes from one place to another, and there's no money in the other end of it, it can be disrupted, but we have to figure out the revenue models first. All right, listen, you guys are amazing.