Avlok Kohli on AngelList's role in venture and the evolving startup ecosystem

going on? What's going on, guys? How you doing? Uh, we're good. We're good. It's been a great week. Uh, a lot of lot of big news. Um, I'm not sure what you've been following more closely, AI or tariffs, but everyone's following one. Everybody's got something. Everybody's got something worming in their brain.

Uh but yeah, I mean could you just do a quick intro on you and and Angelist for anyone who doesn't know, I guess. Yeah. Ablock C of Angelist. Uh I was actually recruited in almost six years ago now. So I've been running the company for almost six years prior to that.

I'd started three companies sold to one of them was to Square. So I spent almost three years at Square as well preo to post IPO. Um uh I'm I'm curious. I mean, a bunch of stuff I want to get into. Uh, where should we even start?

I I think it's probably helpful for people to have like, you know, just sort of like the the refounding of Angelist because uh Angelist is a company that I feel like has just been in the timeline. Obviously, we built the show around X. Angelist I think like grew with Twitter in many ways.

It was sort of like the technology platform behind what was happening in the timeline. Um and in the same way that Twitter's evolved massively, Angelist has done the same. But would love that backstory before we get into everything else.

Yeah, the way to think about early Angelus was um there were a lot of u different experiments as the team was searching for product market fit. And so you know 2020 or 2010 to call it like 2018 2019 um sort of spawned three different businesses within Angelist.

It was the SPV business uh which is really the syndicate business. Uh there was talent which was uh to help startups find talent, startup talent and uh the third one was actually product hunt.

Angelus had bought product hunt early on and so by the time 2018 2019 came around uh there were kind of these three different businesses where they're really just connected uh with a thin thread of founders need to uh raise capital, they need to hire people and then they need to uh launch their product.

But outside of that, it was actually very hard to do all three of those under one roof just because the business models are different. Uh so at that moment what happened was kind of a splinter kind of a split. Uh and talent spun off on its own and then product hunt separated a bit more.

And then when I came in, I actually took the SPV business and I spun it out as its own company. And uh we got to work and we kind of took it from an SPV business to venture funds to we invented a whole new category of rolling funds. And then we got into rollup vehicles.

Uh which is actually how Jordy and I originally kind of got connected. Uh and then we got into startup products. And so today we're kind of a um you know the place to go to if you're going to start, scale, launch a venture fund. Uh we're now private equity.

We actually manage the scout funds for a lot of large firms as well. So, we're kind of an index now of what happens within venture. No, it's amazing.

Um, can you uh can you talk specifically about, you know, sort of like accelerating product velocity in Angelist because it just felt like you came in, you had this sort of warm-up period and then it just felt like every single quarter there was like a major new product launch and you sort of like awakened the beast, right?

like there was so much potential there. Uh if you were involved in the startup, you know, industry at all, you had invested in an SPV, maybe you were in a rolling fund, uh maybe you hired, maybe you got a job, maybe you launched a product.

So everybody was t touching sort of products in the ecosystem, but then you had to kind of like ramp up and basically say like, you know, we're going to kind of bring this crazy product velocity. Um so I'd love to hear like h how you did that and then how you're applying that now specifically.

There's, you know, a bunch of new opportunities. I think that are popping up. Yeah, I would say just at the core of Angelist, uh, Angelus has always held the founders at the top of the pedestal. Uh, and we've always believed that product leads everything. Product velocity leads everything.

And so when you have that in your DNA, the question you're always asking is what is the mix of people you actually want in the company. So we actually have a fairly high percentage of our team that are like ex-founders. And so what happens when you put a bunch of ex-founders uh in in a room?

What what are they going to do? They're going to look to uh be ambitious and how can you actually move into an adjacent opportunity? How can you go reinvent a whole new product category?

Uh and so a lot of the original thinking was hey let's actually get back to product innovation uh and let's get back to actually doing things that only we can do right and that's the one question we do ask and typically we don't get it right all the time but the one question we do ask is if we didn't exist in the world would this product exist and if it doesn't then okay great we should go do it but if it's going to exist without us we're probably not the best suited to go do that and so we do ask ourselves that uh quite quite often.

Can you talk about AI uh in the business how you guys are leveraging it to just uh you know you're in uh in you know investing in a company is very simple on Angelist like I'm a part of a scout program it's very seamless you know painless process uh but then under the hood there's you're dealing with different entities and you're dealing with y there's uh the code is like the easy part like the law is really the hard part and the challenge.

So I'm I'm curious how you're applying AI in the business to make you know internal team members more efficient to make uh and it's good good for the world if you can sort of like accelerate capital formation investment and all these things it it can accelerate the world. So um curious about that.

Yeah thank you Jordy for recognizing that there's it's like an iceberg product right there's so much underneath the hood. Yeah. Um, you know, the initial investment is actually just the beginning of that uh relationship with that company on behalf of the fund.

And so for context for others, um, when an investment is made from a fund into a company, that's typically a uh it can be like a 10ear 15 year hold period. Uh, and that's just because of how illquid venture typically is. These companies need time to mature.

And so what ends up happening is over the quarters, months, quarters, and years, there's all sorts of activity that can happen within that investment.

all sorts of activity that can happen within the fund and there are there are real like legal repercussions financial repercussions if you get it wrong and the way we're using AI is these are all the back office functions and so typically humans would manage all the different workflows what we've started looking at is how do you take all these different workflows and then how can you actually have um AI agents starting to take on some of those workflows now we have to be careful because the uh uh you know the uh when you use chatbt and you get every time you ask a question you get different answers that's beautiful that's product market fit right like write me a poem or write me many different poems awesome you don't want that when it comes to your finances right it's like hey what's you know what's the share price you don't want 10 different share prices you want one uh and so we're we are uh pretty bullish on taking that and and having it automate huge portions of our back office and that's already starting to happen uh and we think there there are ways that we can actually then have uh a lot of the uh folks that are doing some of the back office work move up the stack in terms of the type of judgment they apply on all the different workflows there.

So that's like one piece. Uh the second uh that I'm personally extremely excited about is we launched a front-facing product uh called um uh on on it's our intelligence product. Um and uh it's in beta. It's still new. Uh so we're not ready to like fully talk about it with the world.

But um what we're doing there is we're actually creating a um uh an agent to go look at all private market data. So we have access to a lot of unique aggregated anonymized angelist data and we're actually looking at building partnerships with many other providers.

So for example, you can ask a question like who left OpenAI in the last month to start a new company. That can help you with deal sourcing and you can imagine it scales to many more companies.

Uh we're also looking at uh how do we help you get ready for your briefings uh for pitch com uh pitch meetings uh while you're yeah while you're actually like in the middle of a pitch post pitch great take the transcript and we can help you uh do a lot of the necessary re research market map um uh the full analysis so give you time back and we think of this as we want to build a co-GP we want to help you make more money right we can help you find better deals yeah what about on the founder side I could imagine that you know there's all these new benchmarks for fastest company to 100 million ARR uh valuations are all over the place are you thinking about building any products for entrepreneurs they upload their deck and then they say oh well like you should probably expect you know 10 on 80 pre in this you know to be kind of in the fairway maybe go out there and get more but here's at least our take on it 100% we're we're already in that world you know classic line the future future is is here.

It's just unevenly distributed.

So that that's basically Yeah, it's funny how how entrepreneurs try to understand how to price their rounds is really they talk to a handful of investors who just reference like the last two or three rounds that they saw get done that were maybe not even in the same category and then it's just like completely guessing and I think it's you know super like you guys will be able to pretty quickly I imagine give a pretty precise and say like here's where we predict your round is going to get priced and sort of like building that feedback loop is be cursor for dogs.

But what does that really work? How do you uh It feels like we're at this amazing time right now. Uh you know, some people don't think it's amazing. I think it's great of of venture maturing and then you see like the convergence between private equity and venture.

You guys launched some uh products more geared specifically for private equity, but um how how do you see that line sort of like blurring over time? Uh and I guess like h how how are you kind of um adjusting your your product roadmap for that reality?

Yeah, the the way to think about the way we're approaching product is we have the kind of vertically integrated part of the product where you come in one you know one-stop shop you can launch scale venture fund uh which also allows us to uh manage scout funds for some of the largest firms and then we actually have our software that's getting unbundled and folks want to adopt that.

So some of the largest um private equity firms actually adopt our uh digital subscriptions product or uh some others can adopt our banking product. We actually we run banking underneath the hood. We've actually built a whole banking infrastructure which is what allows for a lot of the smooth operations.

Uh and so you know as we're seeing the market evolve um our roadmap is effectively evolving to continue to keep taking on larger and larger funds for this vertically integrated product and then for the largest ones where there we can support them for any of their needs and that's actually working quite well.

Uh in terms of what we're seeing in the market today we are seeing a bifurcation.

Um, so what's interesting is if you look at year-over-year uh within our data, we're actually seeing uh capital flows increase Q1 2024, Q1205 by almost a double in and into just venture funds and like we talk about SPVS and all of that, but like just venture funds alone, capital flows are almost have almost doubled.

So there's definitely a bounce back. There's more optimism, right, that's coming in. Um but it is being um it's sort of concentrating into uh a certain subset of managers. Um so you kind of have uh you know call it zero to let's say 100 million. You kind of have a divide and then you have like the mega firms, right?

And then the mega firm strategies are actually also bifurcating, right? They all have kind of their own their own view on things uh and what they're going to go after. So it's a very different market than 2019 2020.

Uh, and we're not seeing as much of, you know, because what happened in 2021, 2022 was you had crossover firms really come in, right? Private equity firms, public firms come in to venture. We don't see that anymore. I think we at least we're not seeing too much of it.

Maybe like here and there, but generally it's the venture firms that are scaling up and they're now starting to push up into different asset classes. How do you think of um the job of GPS over the next 10 years? I I've you know, everybody's, you know, been running the analysis of like how safe their their job is, right?

Like if you're a writer right now and your job is to summarize information and you know, re or repurpose information and just put it out there, uh you got to be pretty worried.

I think investors in general are pretty safe because people want to give money to one person right now and have them be sort of responsible for returning that money and and hopefully a lot more in the future.

Uh and venture is about this combination of not just picking which is like can be very analysis driven of the market and the product but also the people.

Um, but then the big thing in venture is, you know, I expect AI to not necessarily dominate venture so quickly because likability and access are like such big parts of it, right?

It's like, you know, somebody wants Avlock to angel invest in their company because you're the CEO of Angelist and like an AI could be like, you know, um, better at finding companies, but that doesn't mean they're going to sort of like win the allocation.

So, uh, is your thesis, you know, let's give investors AI tools to be able to be better investors and make more money, um, or do you see a world in the future where people are setting up, you know, sort of a fund on Angelist that's entirely, you know, the GP is effectively, you know, um, a machine itself?

Yeah, it's a good question. I don't know who said this, but I picked this up from some podcast. When a founder picks an investor, they're doing it because they believe that person is going to increase their probability of success. Yeah.

And in order to increase the probability of success, um you really looking for a partner who can help you solve any like number of problems that can come up. Some of it's brand to help with recruiting, some of it's like actual like operational help. And and it could be many more of those, right?

But you need to partner with someone so you can actually have a reasonable chance of success with the company. So I don't think that's ever going to come from an AI at least today at least in in the form that we know of it today.

So as we think about um the the role of the GP we think it's going to continue to stay the same. Uh I think what will get challenging is I I do think there will be more capital that floods in because even at the earliest stages you know folks are talking about how valuations are decreasing.

Yes, at the later stages at the earlier stages it's increasing. So preede seed series A there's just a uh a huge amount of capital looking to get in and the actual root problem is that the the startups don't need that much capital at the earliest stages.

So what you have is a supply demand mismatch and so the only way the valuations go is up, right? Yeah. And so the the access question and problem is only going to get worse and worse. And so we still think the the human the GP is always going to be in the loop.

So the way we think about the tooling that we're building is uh is they're meant to amplify the investor. They're meant to help them make better decisions, but it can't ever help you be become likable to the founder or help you get access to the deal, right? that's going to be incredibly hard.

Um, and so we think that it's about enabling them. So really there's going to be more leverage to the best investors. So I actually think the power law is going to get even more concentrated. Um, because even the tools that we're building are going to help investors understand very quickly the lay of the land, right?

So one thing that's actually very hard today is when you're listening to a founder, you're doing a pitch meeting, you generally like you don't quite know, okay, who are all the other competitors? Who are the other founders of those competitors? Mhm. With one click, we're just going to make it easy.

And it won't be just SOS. I'm sure there are other tools that will come out of like one click, you got full layer of the land. And so you have full visibility on what's going on with this company uh what this founder wants to do. And so I think leverage will increase uh to the best investors. Makes sense.

Uh I don't know if you have a hard stop, but I did have one more question around how your thinking around stable coins has evolved. You guys were very quick to adopt stable coins. I think it was in 2021 or early 2022 that you uh rolled them out.

That was probably from user demand being like, I have a lot of money on chain and I want to put it into startups.

Um, now I imagine uh I'm I'm a long-term believer in the power and value of stable coins, but I never in using Angelist at any point in the last two years was like, you know, I want to fund this investment via stablecoin.

And so I'm curious how how you think about it broadly and are you using them behind the scenes at all or you just you you said you talked a little bit about your banking infrastructure.

Um yeah and I would assume that that's still based on traditional fiat rails just because beauty of venture is like sometimes you want to close an investment quickly but we have same day wires and those like work pretty well.

Uh and then you're not like trading in and out of assets like super rapidly or anything like that. Um, yep. And so, but but anyways, I'm curious to hear um how you're thinking about stable coins broadly. Yeah.

So, we originally added stable coins uh because of user demand and it was like extremely aggressive in terms of like demand, you know, and I was like fine, let me go let me go figure this out.

And at the time, we actually looked at I mean all the who's who providers and none of them could really quite fit the use case we had. Um, and so we eventually actually worked with a a startup and um ended up building out exactly what we needed and and now they're actually doing pretty well and they're scaling out.

Um, was that uh was that layer was that layer 2? Yeah, they renamed themselves to rail r a l. io. I'm a I'm an angel as well. I remember that saying like we got we got a lock, you know. That's awesome. Yeah.

Um, so what's interesting about stable coins on uh Angelist is when the the biggest use actually comes from the biggest pain from when LPS fund and the biggest pain is when you're trying to move money internationally.

And so we actually see a um a huge amount of demand from international LPs trying to move money into the US because if you've ever tried to tried to send a Swift, God bless you if you have, it's insane.

um it literally takes days and and you just don't know where the money is and sometimes bank will just hold on to it and they won't even tell you because it's got stuck somewhere in the middle. With stable coins, you're able it's one hop. You're able to get the money right into the US.

It goes through all the same compliance laws and everything. Um but we see a huge amount of demand for that. So it's not necessarily in the fund side to companies because you're right, there's not that much trading in and out, but it's more on the LP side into the fund. Yeah. Well, yeah, that's very cool.

Uh last uh last question I have for you and then I we will let you leave. I know we're I know we're 5 minutes over. I'm just very curious. Have you seen um looking at some of these sectors that are uh especially overheated or potentially in sort of bubble territory, right?

People are, you know, you we could argue whether AI is a bubble or defense tech is a bubble. Have you seen any sort of near-term slowdown in defense tech investing this year?

Um, and and I'm sure you don't have the data in front of you, but but I'm curious that feels like one that, you know, is still very hyped, but it's potentially investors like sort of have their bets now and and they're letting, you know, want to let their kind of bets play out.

Um, no, I was actually just talking about this with uh someone else the other day. Uh, it is still going on with defense tech. I think we're actually just entering a uh sort of a golden age uh for the US wanting to basically uplevel all of the all the technology in in the military and the Navy.

Um so I'm actually not seeing it slow down. Uh I agree with you that there is it feels like it's a bubble, but the beauty of uh financial markets in general is you'll just get bubbles and then you'll get a few great companies that will come out of it. Yeah.

But I haven't seen that particular bubble sort of like wind down yet. Uh we're still seeing some pretty active heavy heavy investing there. Very cool. Makes a ton of sense. Fantastic. Well, we'd love to have you on to uh be our our private market data. I mean, you have all the data. Yeah.

Pull it out of you one way or another. This is great. Thanks so much. Great. Thanks for joining. Likewise. Good to catch up. Have a great weekend. Bye. Um let's move on to some timeline. We got some uh massive news yesterday from Anderol. Uh they launched the Seabed Century.

Uh Anderl writes uh we must fortify autonomous subc dominance of the US and its allies. Uh Seabed Century is their new AI enabled mobile undersea sensor node network designed for persistent monitoring and real-time coms.

So I imagine this would be valuable if you had an undersea cable that you didn't want to get cut or blown up. Exactly. There were Yeah, there were some very funny uh post about I mean uh Polymer contextualized it very well by just saying like what we do for the sensor towers on land.

We now have a product that does that underwater. Um, but he said that it can uh detect uh uh like other submarines and boats, but also biological uh things. So I think it can even track like whales, which is very very cool.

Um, and so if we domesticate the whales, get them working for us, turn them into a defense tech weapons, a weapons platform, uh, and is going to be on top of it. And interestingly, if you zoom in on this, uh, video very closely, you'll see that they're partnered with Sonerdine, uh, which is a very old company.

They've been in business for uh 50 years. Um and they uh and and they they have a number of case studies for who they work with. Energy, ocean science, defense, carbon capture, etc. Uh and I'm sure there'll be a case study on this product as well. It feels like as we get more autonomous underwater vehicles.

You have people like Chris Hamdon working on stuff like this. Tons of exciting companies in the space. The risk to pipeline and cable sabotage. Totally just feels like if again if you can send a $10,000 drone to blow up a pipeline that can cause billions trillions of of economic damage.

Um that's just going to be a huge problem. Uh you know we've been talking about like the maybe the anderol of X is anderol and uh but this was something that I haven't even seen startups working on. This seems like an idea that you only get if you're as deeply entrenched as anderol.

Um, and uh, yeah, if you're going on if you're going down into the deep sea and you can afford to come up with products that aren't being purchased yet, right?

Like basically create markets and you have um, uh, once you're at a scale that Anderoll is at and have those customer relationships, I think it's a lot easier to do that. But if you're focused on subc dominant diving, what watch should you have on your wrist? Probably submariner. And where you going to get it? Bezel.

Go to getbbezzle. com. shop over 24,500 luxury watches fully authenticated in house by Bezel's team of experts. You can get a dive watch added to your collection. You need a dress watch, you need a sports watch, you need a a dive watch. Tariffs are hitting Switzerland, so get in while you can.

But uh so retail prices will go up. Um uh I expect prices on bezel to go up as well, but probably slower and less uh systematically. So download the app, start building your collection, your wish list, and find something beautiful to add. Uh speaking of tariffs, uh George H.

Hots posted from the Tiny Corp account uh a big long post about uh how the tariffs are affecting his business, and it's it's really it's really like a frustrating story. Uh he says, you know, we talked to Chris Power Adrian. He's obviously a beneficiary of the tariffs.

Uh the tiny box is uh is is much less of a beneficiary uh potentially harmed. And he breaks it down. He says, "To date, we have man manufactured all tiny boxes in America. However, we buy parts from abroad. There is no way to buy Americanmade GPU or motherboard at this time, and there won't be for a long time.

If these tariffs stand as is, we would have negative margins on tiny boxes. Our motherboard manufacturer has already reached out and tried to get us to pay the tariffs on things we already agreed to deliver on the delivery price for, but I don't blame them. Their margins probably go negative with the tariffs, too.

I sort of doubt we'll be getting our 5090s at the price we agreed upon either. And if that's true, the whole thing is really out the window. And even more stupidly, there's a restricted list of countries you can ship 5090s to.

So, I'm worried uh I'm not so sure we could move manufacturing of the green V2 their product, and the product may just be cancelled. I'm not going to spend my time figuring out weird loopholes and incentives to reexport and FTZ and maybe try and ek out a small profit after all the administrative cost.

Tariffs are regulations. When difficulty of business go of doing business goes up, many people only marginally making money just stop doing business. The US has an ease of business, ease of doing business score ranking of six. Hong Kong's ranking is three.

If we manufacture here in Hong Kong, we have free trade and can continue our policy of selling everywhere and passing the tariffs on to the buyer. Uh European Union people have been dealing with us for a long time. Now, US people will too.

if we can't get 5090s because of short-sighted US export regulations, we'll have to ship, we'll have to switch to something that we can get, maybe a different graphics card. So, very frustrating.

Um, and and and an interesting real life case study, not just a pundit kind of saying, "Oh, I think the tariffs are bad for economic reasons or they're great for economic reasons or or, you know, it's some part of some grand 5D chess or whatever. " This is somebody who's really trying to build a business on the ground.

clearly, you know, one of the greatest programmers of all time and and and a fantastic uh just I don't even know how to describe him, like business person, developer. He's kind of everything. Um but George H. Hot.

Hot breaking it down about why uh the tariffs are actually leading him to leave America and and focus on Hong Kong. Uh and and you know, who knows how that will play out, but uh very frustrating story, but interesting to hear him uh him break it all down.

And of course, if the tariffs are moving the markets, you're going to want to be on public. com. uh multiasset investing, industryleading yields, and guess what, John? They're trusted by millions. They're trusted by millions. Investing for those who take trusted by us. Go to public. com.

Thank you to public for supporting the show. Um and uh and in relation to the uh uh to the to the market turmoil, uh Skooks had a funny post here. After the 911, after 911, the stock market lost 1. 4 trillion inflation adjusted. Today, the stock market lost$2 trillion. So these tariffs are like 1. 42 911s. Yikes.

Rough 5k likes. Uh very popular. Um what else should we cover here? Um just uh just FYI, the the map of states that are renamed for countries with similar GDP really puts America's dominance in place. Just the state of New York is the same size as Canada. Just California is the size of India.

Just Texas is the size of Brazil. Just just Florida is the size of Indonesia. mobed. Uh George Hos is over in Hong Kong. That's the size of Indiana. Yeah, America uh stays undefeated. I mean, this is 2019. We'll see where it goes.

A lot of these countries are growing and some of these states might be shrinking, but uh I'm still long America even with all the crazy tariffs. Um and you know, no better way to get your message across in America than some out of home advertising on ad. com.

If you want to break through the noise, you got to go out of home. Out of home advertising made easy and measurable. Say goodbye to the headaches of out of home advertising. Only AdQuick combines technology, out of home expertise, and data to enable efficient, seamless ad buying across the globe.

Um, should we talk about what else is interesting? Uh, Vtorio had a funny meme, the virgin tariffs warrior versus the Chad, we'll see what happens lmao. Yeah, I feel like, uh, this obviously resonated. Uh, it's our our approach has just been, you know, this is happening. We'll see what happens.

Uh there's a lot of negative impacts and then there's people like Chris Powers. Yeah. That are benefiting from it. Um you know, overall it's much less easy to be in the Chad camp if you are George Hots and Tiny Corp and your your product is being um you know impacted or or you know the business entirely.

So, we'll we'll we all have to see what happens, but um definitely feeling that for entrepreneurs this week that are just dealing with uh immediate repercussions from it and uncertainty for sure. Should we close out with uh this thread from carried no interest? Let's do it.

Uh so carried no interest, he's been on the show before. Uh he says it's time to coin some new AI software terms. He's employing Kugan's law. He says inference to impact I2I is the first and inference risk quotient IRQ is the second. And he defines these. This says the first that stands out to me is inference to impact.

What does this mean? It's the amount of time it takes within a new software product from hitting an LLM API to getting customer utility. I've noticed a distribution in these new startups. Let's start with cursor. Cursor has a very low eye to eye inference to impact.

As soon as you start ripping the application, you are hitting an LLM and getting results. The I2I is immediate. This is a good thing. Now, the opposite is AI SDRs. You hit a bunch of LLMs and send some outreach. Much longer I2I on this product bad. It's kind of the the the iteration loop.

A low I2I is good on a bunch of levels. Your customers are seeing immediate magic. You can onboard users faster. Sales cycles should be shorter. When you can shorten the amount of time between LLM and utility, this is objectively very good. But it's time for another one. The inference risk quotient, the IRQ.

The IRQ describes the amount of risk created for the customer from a series of LLM calls. Uh for some AI first software companies introduce very little risk to a customer by calling an LLM. Others a good amount. Let's describe it. Cursor works well in for this example as well.

Um, cursor's IRQ, the the risk quotient is theoretically low. Cursor generates code, that code goes into a div, a a diff. Uh, it should be tested and most errors caught as it progresses through staging environments. Solid IRQ. What about the inverse? Let's go back to the AIS SDR example.

Your AI SDR starts ripping emails out the door. Some of them are bad. They maybe embarrass your marketing department. Oof. Probably a medium IRQ. What would be a high IRQ, an inference risk quotient? Let's think about Harvey. Harvey is an AI software for lawyers to analyze whatever lawyers analyze all day.

In my opinion, this would be high IRQ. If the AI misses something important or mclassifies, you could have potential legal damages. There's probably one more item, one more term in here. LMAO, IURQ, inference utility to risk quotient. This would be the amount of risk relative to utility.

cursors utility relative to risk is so high. Your staging environment should catch bugs and your output is much higher. Fantastic framework. Frameworks uh very on point. I almost always say carry no interest real name every time. So eventually you're going to get docs buddy.

Uh not on purpose, but uh this reminds me of what Avlock was bringing up. If if you're asking um Angelist via an LLM what the share price is on a specific position, investment, etc. , it it has to be right. Right.

Because you're going to go make you know different financial decisions based on that data and like they they they need to be a u you know accurate source of records on these type of things and people just aren't going to tolerate um you know hallucinations in that environment.

So uh I think this this very well sums up why we're seeing the adoption uh you know cursors adoption curve versus we don't we see these sort of AI SDRs that are um you know getting customer traction but it seems like the churn is much higher and the utility is much lower.

I haven't seen people you know it was it was great having from rocks on. Yeah. Sounds like people are getting a lot of utility out of that, but it's almost more of a CRM type tool than purely um an outbound, you know, engine. And so um we haven't seen anybody raving about their AI BDR SDR yet.

That's also could be that if it's working so well, you don't want Yeah. Don't alpha, I guess. Yeah. Don't want to. Well, I want to do a couple more. Uh Andre Carpathy had an interesting post here. He says, "Let's take AI predictions from blog posts, podcasts, and tweets and move them to betting markets.

are state-of-the-art and truth. Uh, obviously we are sponsored here by Poly Market and I'd love to see more AI markets on Poly Market. I'm sure we'll be working on spinning some of those up. Uh, Andre Karpathy continues to say, "Uh, my struggle has been coming up with good, concrete, resolvable predicates.

This is always the tough thing with Polymarket is that, you know, it needs to have a clear resolution. It needs to be not too far out, not more than a year. You know, uh, you don't want your money just sitting there.

" Uh, ideally predicates related to industry metrics and macroeconomics, eg naively, one might think the GDP, but I'm not so sure that works great. Egc productivity paradox. I also think eval are not amazing predicates because we see over and over that they are incomplete and hackable and and and saturated often.

Um, and I thought this was interesting because he's he's close to kind of my my thesis about the uh uh artificial economic intelligence just maybe instead of tracking towards you know IQ or how does this benchmark against a human or the uh the the touring test we just want to say how much economically valuable work is being done by LLMs and AI agents and diffusion models etc like of the GPU view cycles that we we have we have strict data on capex and inference cost and how much energy is going into these data centers how much how much economic value is being produced and once that hits 10% that's probably some sort of tipping point once that hits 50% like the robots have kind of won and that could be good or it could be bad um but but that is true like okay the robots the AIs are producing more economic value than all of human all of humanity combined when we hit that 50% GDP generated by AI threshold.

That feels like singularity territory to me. That feels like a fundamentally different society. That feels like UDI or something. 2027. Yeah. Yeah. I I would take the under on on AI generating um over 50% of GDP by 2027. But yes, the productivity paradox is is tricky.

uh Tyler Cowan has tried to formulate this saying that like you know the AI doomers they should express their pdooms in the form of longdated puts and say well if you really think that this is going to go poorly you should imagine that there will be economic turmoil and you should be betting on that and profiting off of that and if you're not then maybe you're just yapping and so he's going to be on the show in a few weeks and and and we'll have to dig into more about how we can concretize these bets and think about uh how we make predictions about AI progress.

Um, there's humanity's last exam. There's ARC AGI and there's all these interesting evals and uh and tests, but they're not scratching the itch for me in the same way when some new model goes viral. I care more about the Studio Giblly moments than the okay Google's Gemini 2.

5 is two points higher on MMLU or some some exam uh or hacking AP bio even I I care le also something interesting which is if a lab has a truly uh worldchanging uh discovery or innovation they should not tell the world and not release it and just leverage it to the absolute max uh internally uh which again goes back to people's frustration with open AI okay if you are at the frontier and you've felt like it should be you know open sourced uh and now it's closed um you know how how do how do the incentives and everything change um let's close out with Nick uh talking more about AI alignment he says broadly I think a AI alignment people are maxed out in smart and low in wisdom.

I like this because it's goes back to the uh when you build a character in an RPG, you have int and whiz, intelligence and wisdom, and they're slightly different. Um and uh and I think as we like as we try and define what makes a human truly successful. It's not always just put all the points into intelligence.

Charisma obviously matters. Yeah. strength and like strength as a sense of like your grit, your grind, your ability to keep continue working. Charisma obviously super valuable for coordination, bringing people together. There's this agency, drive, lots, talking about RZ. Yeah, RZ, all these different things.

And and yes, we might be seeing uh intelligence max out and get to these super high IQ models, but are they going to be super agentic, super creative, super wiz, super driven by wisdom? Um, and this comment is obviously about the AI alignment people. Um, he says, "Not a comment on anything in particular.

I've noticed myself saying it and thought it would be worth writing down, but obviously uh it's it's like perfectly timed with the AI 2027 thing. " Um, says, "Lots of room for people from other fields to contribute wisdom learned throughout history, even if they can't do a math olympiad or whatever.

I don't think being able to do highle math helps that much. " I guess one specific comment I'll make is that I'm quite excited about EMTT Shear's new company, Softmax. I think he's thinking about things in interesting ways.

And so we'll have to dig into EMTT Shear's new company and how he's thinking about this because he's been uh very outspoken on AI alignment.

um but also a very successful entrepreneur and has built up probably a very deep trove of wisdom running Twitch and and building a real business that has to interface with the realities of the economy and doesn't live he's not a pure academic but he engages with the AI alignment debate in at the same level as academics in my opinion and so I'm I'm excited to see that but uh it is it is it is interesting obviously Sean was talking about this like AI 2027 maybe it's just fanfiction uh I like fanfiction I have fun reading that thought provoking and and I think there is utility in that.

Even if you're just telling a sci-fi story, I'm here for it. I I'll read it all day. So I I would say yeah, we want we want people doing that work. Yeah, I would say more of that, but also you do need to you you can't lose the plot. Um we have to announce a specific milestone, which is that we just hit four hours.

Oh, our first 4 hour show. Uh we're clearly addicted. Um, and I just have one now, you know that we've been talking about AI AI predictions. I have one prediction to make, which is that Monday at 11:00 a. m. Pacific, 2:00 p. m. Eastern, we will be back sitting in these chairs. Yeah. Ready to go again. Yeah.

And I can't wait. And I mean, what's your prediction for TBPN 2027? I think we'll be here. We'll be right here. We'll be right here doing the same thing. Doing the same thing from the front lines of the battle scarred Terminator apocalypse. We will be shooting humanoid robots with with microphones in our other hands.

James Bond's back, baby. We will We will never surrender. We'll never surrender. We will uh keep doing this until the the AI uh the bioweapon that the AIs release. Uh you know, just maybe we'll be podcasting from a hermetically sealed biocure facility. Yeah.

so that no one can no one gets in, no one gets out, but we're always live streaming and then sleeping on our eight sleeps and then waking back up and doing it again. Um, anyway, have a fantastic weekend everyone. Thank you for tuning in. Uh, we appreciate you all. Yeah. And uh