The Truth About Building AI Startups Today

1. 0:00 – 0:41

   Intro

   1. JFJared Friedman0:00

      How would you differentiate between an idea that could be a great foundation for a billion-dollar company, and an idea that is likely to get run over by GPT-5?

   2. GTGarry Tan0:10

      Something that's boring might actually be an incredible business.

   3. JFJared Friedman0:13

      Yeah. Uh-huh.

   4. GTGarry Tan0:13

      But why is that?

   5. JFJared Friedman0:14

      Yeah. Let's talk about GPT wrappers.

   6. GTGarry Tan0:16

      (laughs)

   7. HTHarj Taggar0:16

      Are people worried about giving these datasets to OpenAI?

   8. DHDiana Hu0:20

      All these AI agents are passing the Turing test.

   9. GTGarry Tan0:23

      I mean, this is why I think the chat interface is wrong.

   10. JFJared Friedman0:25

       You wanna do something in AI, like, this is a good place to, like, look into.

   11. DHDiana Hu0:29

       Big generational companies are getting built as we speak.

   12. JFJared Friedman0:32

       Great startup ideas just lying on the ground, you'd, like, trip over them.

   13. HTHarj Taggar0:35

       This might actually be, like, a once-in-a-lifetime opportunity, and I, I think I actually agree.

   14. GTGarry Tan0:39

       What a time to be alive.

2. 0:41 – 1:53

   The Lightcone Podcast

   1. GTGarry Tan0:48

      Welcome to the very first episode of The Light Cone. I'm Garry. This is Jared, Harj, and Diana. And we're group partners at Y Combinator, and we get to work with some of the best founders in the world. Jared, why are we calling it The Light Cone?

   2. JFJared Friedman1:02

      Well, in special relativity, the light cone is the path that light takes from a flash of light. You can imagine a flash of light, and it spreads out in a cone shape. And in special relativity, you think about it spreading out in a cone, both in the future, but also in the past. And in this podcast, we are here in the present, but we are going to talk about both the past and future of technology. So, that's how we came up with the name.

   3. GTGarry Tan1:29

      And one of the things that we're all seeing is the encroachment of AI into almost every piece of, uh, society at this point. You know, every business transaction, every, uh, thing that we sort of use with computers, uh, suddenly a new burst of technology is sort of entering everything we're doing. And we're seeing it in the startups that we're funding, which is why we're so excited about

3. 1:53 – 3:34

   YC Recent Batch

   1. GTGarry Tan1:53

      it. I think, you know, what, what's the percentage of companies you've backed right now that have large language models embedded?

   2. JFJared Friedman1:59

      I think for summer '23, it was close to 50% of the batch. And that's pretty interesting. Like, I think a lot of people, like, see that number, and they think, "Oh, YC must have funded so many AI companies because we have this thesis about AI, and, like, it's just easier to get into YC if you're an AI company, because we just, like, love funding AI companies."

   3. GTGarry Tan2:20

      (laughs)

   4. JFJared Friedman2:20

      And it's funny to us because we know how that's not true, and yet that's probably what, like, 90... That's probably how 90-plus percent of people actually think YC works.

   5. HTHarj Taggar2:29

      Yeah.

   6. JFJared Friedman2:29

      How, how does, how does it, how does it actually work? Should we tell people, like, how it actually works?

   7. HTHarj Taggar2:33

      I mean, I actually think it's interesting. The smart founders apply to us with what they want to work on, and we fund the smart founders.

   8. JFJared Friedman2:38

      Yes.

   9. HTHarj Taggar2:39

      Like, irrespective of what they wanna work on, actually.

   10. JFJared Friedman2:40

       And, and exactly. And so the fact that half the batch is working on AI says something much more interesting than just the YC partners think AI is cool. It's an emergent phenomenon of what the strong, the smart founders wanna work on right now.

   11. DHDiana Hu2:54

       It's like, where do they think there's the high beta to build the largest company? And I think the most ambitious and smartest founders are going after this, because it's definitely... I think the exciting thing about right now with AI, I think it's, like, real. There's been a lot of waves for AI and multiple AI winters. But this one, actually, GPT-3.5 and then 4.0 blew out of the water a lot of task. And it impressed a lot of smart people. When a lot of smart people start paying attention and building in this current idea maze, I think big generational companies are getting built as we speak.

   12. HTHarj Taggar3:30

       One thing I'm seeing that's interesting is, I feel like a lot, um, a lot more

4. 3:34 – 4:44

   College Students and AI Startups

   1. HTHarj Taggar3:34

      founders are dropping out of college to start working on AI.

   2. DHDiana Hu3:36

      Because they don't want, there's a FOMO.

   3. HTHarj Taggar3:38

      Yeah. (laughs) Yeah, there's, like, an actual, like... And usually, it's so funny, my, my interview question is almost always like, "What's the rush? Like, why do you wanna drop out of college? Like, why don't you just, like, graduate? Because it makes a lot more sense to graduate and then do a startup?" Um, and the reply is usually like, "Well, like, this might actually be, like, a once-in-a-lifetime opportunity." And I, I think I actually agree.

   4. JFJared Friedman3:55

      And, and the other cool thing is that this is an opportunity where college students are particularly well... Like, young founders who are particularly well-positioned to work in it, because nobody has, like, like, there's no one walking around with, like, four years of LLM experience.

   5. HTHarj Taggar4:09

      Yeah, yeah.

   6. JFJared Friedman4:10

      So like, everyone is starting from the same playing field. And so if you can learn fast, you're gonna be at the same level as everybody else.

   7. HTHarj Taggar4:16

      That's right, and, and you know one, uh, an area I've seen that come to play is, like, developer tools for prompt engineering. I've been seeing, like, these sorts of tools are getting uptake. It's like, ability to, like, chain together different prompts and test your prompts and see, like, the second order effects. Um, and actually, a lot of college students are the people who are just, like, playing around with, like, prompting models and seeing what comes out. And it's a really easy startup idea for them to, like, just build the tools that they want. And like, the tools that they want are literally

5. 4:44 – 5:53

   AI Startup Success Factors

   1. HTHarj Taggar4:44

      setting, like, the standard for what every developer should want. Like, I know a lot of the headlines are all around, like, AGI and all of the fancy stuff, and then the really cool demos of, like, multimodal AI, like AI-generated video, and, and this kind of stuff. The stuff that I've seen in the batches that are actually taking off is a little bit more mundane. Like it's, um... I mean, I'd probably say a lot of it's sort of, like, workflow automation. Like, um, it's finding things where there was, like, a human doing some repetitive task, usually involved, like, searching for things or filling out forms, and then using, like, LLMs to replace that.

   2. JFJared Friedman5:18

      It feels very obvious to us, the people who work at YC, that this is an amazing opportunity. There's so many jobs in the world that are basically very mundane information processing, typically stuff that's, like, hidden in some back office somewhere, where there's somebody who's just, like, reading stuff and summarizing it, re-entering it from one system into a different system-

   3. HTHarj Taggar5:37

      Yeah.

   4. JFJared Friedman5:37

      ... in, like, a slightly different format. And it's such a perfect fit for LLMs. LLMs are, like, perfect for this job. And yet we actually don't get that many applications for people working on this. And there's a lot of founders out there who are searching for a great idea. So, if you're out there and you're looking for a great startup

6. 5:53 – 7:30

   Opportunities in Mundane AI Tasks

   1. JFJared Friedman5:53

      idea, and you want to do something in AI, like, this is a good place to, like, look into.

   2. HTHarj Taggar5:57

      I can give you an example. So, last batch, I had a company I worked with called Sweet Spot, and we funded them.Uh, uh, the idea was something about, like, food ordering from food trucks? Something, like, random. And they pivoted immediately, looking for a new idea, and the idea they found was, um, using LMs to automate searching for government contracts to bid on. And, um-

   3. JFJared Friedman6:20

      Oh my God, such a good idea.

   4. GTGarry Tan6:21

      Wow. (laughs)

   5. HTHarj Taggar6:21

      Yeah. And submitting the proposals.

   6. JFJared Friedman6:22

      That sounds so boring. What could be more boring than searching through, like, a list of all the government contracts? (laughs)

   7. HTHarj Taggar6:28

      Yeah. Uh, you know, how they found it? Is, um, the, the, like, exploring startup ideas, and then they realized one of their friends, his job was to work for one of these, like, government contractors, and his whole day was just spent, like, refreshing this government website, um, to, like, find things and then submitting proposals. And they're like, "What? Like, that's... Like, exactly that, that's so boring." (laughs) Like-

   8. JFJared Friedman6:47

      (laughs) Like, "Wouldn't you like a tool that did this for you?"

   9. HTHarj Taggar6:49

      Yeah.

   10. JFJared Friedman6:50

       (laughs)

   11. HTHarj Taggar6:50

       And they launched and, like, pretty much straight out of the gate, got like, um, a pretty decent amount of traction, because they're, like, opening up, um, the people who would actually do it. Like, it becomes easier to, like, find government contracts to bid on when it's all automated away, and, like, software does it for you.

   12. GTGarry Tan7:07

       So, you know, obviously, we all know that, you know, something that's boring is actually kind of awesome-

   13. JFJared Friedman7:14

       Yes.

   14. GTGarry Tan7:14

       ... but why is that?

   15. HTHarj Taggar7:15

       (laughs)

   16. GTGarry Tan7:15

       That's like, you know, just off the bat, you know, we have a sense that something that's boring might actually be an incredible business.

   17. JFJared Friedman7:22

       There's an old PG essay where he talks about this, and he says, um, he, he quotes a phrase, "Where there's muck, there's brass."

   18. HTHarj Taggar7:28

       (laughs) Nice.

   19. JFJared Friedman7:29

       It's like,

7. 7:30 – 9:36

   Beware of "Tarpit Ideas"

   1. JFJared Friedman7:30

      it's, as far as this, it's almost like Old English. Do you wanna explain it, Harj?

   2. HTHarj Taggar7:33

      Just means, like, uh, you can-

   3. JFJared Friedman7:34

      It's like-

   4. HTHarj Taggar7:34

      ... find treasure in surprising places.

   5. JFJared Friedman7:36

      Yeah.

   6. GTGarry Tan7:36

      And, and I think the cool thing is you have to go deep and vertical and solve a very concrete problem. Like, some of the problems with... Let's maybe talk about AI tarpets.

   7. JFJared Friedman7:47

      What a tarpet idea is, is it's an idea that from the outside looks really shiny and attractive, it looks like a great startup idea, and so lots of founders go and they start working on it. And then you realize once you're in it that it's actually not a good startup idea, but, but by the time you're there, you're, like, stuck in it. And so it just attracts founder after founder, and they just get stuck in the tarpet idea. And we see this a lot at YC because we see all these applications, and so it's really obvious to us when, like, 500 people apply to a YC batch with the same idea, but they don't know that 499 other founders are also stuck in the same tarpet.

   8. HTHarj Taggar8:21

      What's tricky, I think, about tarpet ideas for AI is, like, we know something's a tarpet idea in hindsight once, like, enough people have been stuck in it. So, with AI so new, we don't know yet. So I, I have a couple that I'm actually, like, keen to get your all's thoughts on. Um, a very common one is AI co-pilot. So it's like, "Hey, I'm gonna make it easy for, um, people to, like, build an AI co-pilot for their product or, or service." It's this really unusual type of phenomenon where there's so much interest from potential customers to, like, want a co-pilot that it's actually quite easy to start getting, like, inbound leads if you pitch this, and it've, it's even easy to get people to pay you money upfront. But what's really hard is to get them to actually, like, use the co-pilot, because they don't actually know what they want it for. Like, they just heard that AI co-pilots might be changing the future of software, so we should have an AI co-pilot, but they don't actually know what their customers will use it for.

   9. GTGarry Tan9:20

      I think for me, and maybe I just have a, uh, a mental block around chat interfaces, but I've never been that big a fan of chat, because it puts so much of the emphasis on the user knowing how to speak to a computer. And,

8. 9:36 – 10:30

   AI Integration into UIs

   1. GTGarry Tan9:36

      you know, while in the next five or 10 years I think we will all get far more used to using it that way, um, I think the, the low-hanging fruit right now is just using the large language model to actually do the sort of knowledge work that a human being could do, and then package it into the UI that, you know, may, whether it's a mobile app or a web app that is just familiar, like sort of what people use to do their work right now. And it's, you know, basically, the LLM is better used as sort of this, like... I don't, I mean, it's almost like, you know, this thing that's sprinkled in that, you know, it, it, the software suddenly does something really powerful (laughs) , but you don't have to change the way you would wanna use the software as it is.

   2. JFJared Friedman10:21

      It's sort of like a, an example of a phenomenon that, like, I, I think we have seen in the past when, like, some technology gets really hot and all of a sudden ev- like,

9. 10:30 – 11:54

   Avoiding the "Checkbox Mentality"

   1. JFJared Friedman10:30

      all these companies are, like, they're being asked by people, like, "What's our AI strategy?" And they're like, "Oh, well, we better get an AI strategy." Or, like, with crypto, there was like, oh, everybody needed a blockchain strategy. And even before that, it was like, everybody needed a mobile strategy. For a moment in time, it's, like, easy to sell them something that, like, placates their desire to check some box-

   2. HTHarj Taggar10:49

      Yeah.

   3. JFJared Friedman10:49

      ... but in the end, you've gotta actually make it successful for them. Like, otherwise it's not gonna stick.

   4. HTHarj Taggar10:56

      I agree.

   5. JFJared Friedman10:57

      And so, like, per- perhaps with this AI co-pilot thing, like, maybe it's too early to call. Like, perhaps they actually will find product market fit, maybe with something that's not a chatbot UI. Like, they'll, like, keep iterating on the UI until they find something that's an AI co-pilot people actually want, or maybe it's just gonna, like, fizzle. Just, like, turns out most people don't need an AI co-pilot.

   6. HTHarj Taggar11:13

      Some of the advice I've been giving those, those specific companies is the, another old PG essay about if you, if you're trying to sell technology to someone and they're not buying-

   7. GTGarry Tan11:24

      Mm-hmm.

   8. HTHarj Taggar11:24

      ... like, see if you can just build a competitor. And so it's like, hey, if you're trying to sell, like, um, a fintech company a co-pilot, and they're not buying it, well, like, if you are convinced they should have a co-pilot-

   9. GTGarry Tan11:36

      (laughs)

   10. HTHarj Taggar11:36

       ... like, why don't you just, like, build the company with the co-pilot as the main experience and see if you can out-compete them or not?

   11. JFJared Friedman11:43

       I like that.

   12. GTGarry Tan11:43

       That, I like that. I think getting people to focus on the use case. I think the problem is the whole thing with, um, kind of the gold rush. People selling more the shovels and the tools. And even

10. 11:54 – 13:45

    Focus on Genuine Use Cases

    1. GTGarry Tan11:54

       then, in this case, it is a bit of that, but a lot of people aren't digging gold yet.

    2. HTHarj Taggar11:58

       Yeah.

    3. GTGarry Tan11:58

       Like, the reality is, uh, this is such a new technology, and even the end applications that apply AI, the reality is they're so early they don't have product market fit, so it's sort a bit of a, the blind leading the blind in here. It's like-

    4. HTHarj Taggar12:12

       Yeah.

    5. DHDiana Hu12:13

       What? Do I even know what the pattern is for co-pilot? I mean, it sounds cool just to join the cool kid club of, "We're doing AI and we're gonna check mark." So, I think that's a danger for a lot of these, uh, startup is, like, it seems that they're getting traction, as you mentioned, but then when you- we poke them closer, is anyone actually using you? What are the actual use case? And then the founders come back and they stare blank at us like, "Oh." But look at all the sign-up. Look at all the revenue. But then they're not really using your product.

    6. GTGarry Tan12:42

       I mean, we're seeing even the second order effects, right? So, a bunch of us are funding, uh, dev tools companies that sell to AI companies, and they're selling tooling, but then they might s- you know, they might sell a enterprise contract to someone who also upstream has a Fortune 100 that said that they'd pay $100,000 a year for that contract. And then six to nine months later that, you know, Fortune 100 went back to the incumbent, uh, you know, some other leading, you know, IBM, Salesforce, like, s- something like that, um, because they ended up adding large language model technology to what they- they were doing and people just switched back. And suddenly the dev tool company suddenly realizes, "Oh, I had five contracts, but three of them went away because my customer actually lost their customer." So, it's actually, like, sort of remarkable how fast this is evolving, you know, right now in 2024.

    7. HTHarj Taggar13:36

       Uh, a specific type of idea I'm curious to get thoughts on here as well is, um, offering, like, fine-tuning open source models, sort of as a, as

11. 13:45 – 15:20

    Fine-Tuning Open-Source Models

    1. HTHarj Taggar13:45

       a service broadly. Like, that's a very popular idea, I think, over the course of 2023. Here's what I've seen. So like, why do people want, like, why is there any demand for a fine-tuned, like, open source model at all? Um, it tends to be initially, I think, the big driver was cost. Like, OpenAI, like ChatGPT was expensive and people wanted a, um, cheaper version of it. And so I think it was very easy to get customers with the pitch of, "Hey, like, we can fine-tune an open source model and it's just going to be much cheaper." Uh, what I think a bunch of the companies in space are seeing is that, like, that's not enough to keep the customers, especially because, like, OpenAI, like, the cost of all of the models is just going down.

    2. DHDiana Hu14:27

       And that's gonna keep happening with the- OpenAI has a plan for all of those, so there's something more that all these fine-tuning companies need to do.

    3. HTHarj Taggar14:35

       Yeah.

    4. JFJared Friedman14:35

       It has to be better, not just cheaper.

    5. DHDiana Hu14:37

       I think whereas exactly that, where I think is having more legs is when these companies need to customize it to private data sets. So, you have the open general big foundation model, but then you have to tune it up to specific data sets that, for example, a healthcare or fintech can give out- can give out and they don't have the team of, um, experts to do it. So, I think the one company that, uh, I think Brad worked with was Credl that kind of was doing that.

    6. HTHarj Taggar15:07

       What are you seeing about, like, so the concern around data privacy is another big reason. Like, are you seeing that as being enough? Like, are people worried about giving these data sets to OpenAI?

    7. GTGarry Tan15:19

       It's really interesting.

12. 15:20 – 17:09

    Data Privacy Concerns

    1. GTGarry Tan15:20

       I mean, whenever you have something so new like this, it's actually, um, sort of resets the clock on the competitive landscape again. So, you know, you almost can expect all the same things will happen again, um, you know, just as 10, 15 years ago, cloud was brand new, and then you had cloud cybersecurity and CrowdStrike and all these companies sort of come out. Um, you know, we're seeing the first wave of cybersecurity companies. You know, like PromptArmor. So they sort of wrap your API calls and, uh, what they actually have figured out is that for a lot of large language models, if you do any sort of fine-tuning or training with private data, you can actually just speak to the model and get it to spit out your private data again. (laughs) And they have a solution that stops it. So, it's so interesting because, you know, it's entirely possible. You know, they're basically creating a new industry again, um, of cybersecurity for LLMs, sort of in the same way that cloud opened up that space and created cybersecurity for the cloud.

    2. HTHarj Taggar16:18

       Yeah, I definitely think that whole world of controlling, within an enterprise in particular, like, controlling who has access to, like, which LLM has access to, like, what data and who has permissions is, like, a really ripe space for building interesting software.

    3. DHDiana Hu16:31

       I think the other exciting area that a lot of dev tools are getting built is getting more, this is, like, a step further fine-tuning, but more purpose train models that are smaller. So take, uh, for instance, uh, LLaMA and getting those to run locally in machines for inference. And when you customize them, train on a specific domain and target data is gonna perform better than the general model. The general model was kind of train on all of the human language for all of the task, but if you wanted to build, like, the best, let's say, um, language

13. 17:09 – 18:36

    Purpose-Trained AI Models

    1. DHDiana Hu17:09

       model for parsing SQL queries, you would then parse very specifically just the set for SQL query. And I think some of those that are interesting companies that we fund is like Ollama that you funded that's trying to make the development process for running all of these locally a lot faster. And I think we're also funding some of these that are custom for coding. The thing that was surprised learning from some of the startups that are building, um, coder type of, uh, co-pilots, which I think is, is a use case that's working out, making a lot of the workflow for programming a lot faster is kind of like auto-complete and co-pilot type of thing. They're training on older models of, uh, GPT. They don't even need the newest one. And then I asked like, "Why is that?" And even for, like, one of the companies we funded last batch, Metalware for hardware, they're not using the state-of-the-art model. Like, the older GPT... I forget which one was, like, the older 2.5 or 3, was sufficient in actually creating good enough results because the vocabulary f- for a specific domain for hardware or software is a lot smaller than the human language. So, it's this other world where the open model that's...... customized, I think it's gonna win and compete versus the big one for specific domains. So, lots of companies with this.

    2. GTGarry Tan18:25

       Yeah, that's what, uh, Tobi Lütke from, uh, Shopify actually still dabbles with the stuff. I think he actually built the, uh, internal co-pilot for, uh, Shopify.

    3. JFJared Friedman18:35

       Oh, cool.

    4. GTGarry Tan18:35

       And what he was

14. 18:36 – 19:45

    AI Models for Prototyping

    1. GTGarry Tan18:36

       saying is the best way to use whatever, GPT-4 or the, you know, latest closed source models that are most expensive and have the most parameters, uh, just think of it as a prototyping tool. Anything you do with those prompts, you can get your own model to do (laughs) with a little bit more training.

    2. DHDiana Hu18:53

       It's kind of like, uh, when people build hardware. You have the analogy of, uh, prototyping with FPGAs-

    3. JFJared Friedman18:58

       Mm-hmm.

    4. DHDiana Hu18:58

       ... which are very expensive, right? And then when you have the right architecture for hardware, then you do the circuit path and actually do the custom SoC.

    5. JFJared Friedman19:08

       Yeah.

    6. DHDiana Hu19:08

       So, right now for some of these tasks, the large language model is sort of like your FPGA, your whatever, GPT-4, and then when you customize it, you do like, the super efficient one, coding path for, I don't know, Shopify, for coding assistance and hardware, software, et cetera. That becomes your SoC that you train and customize, which is cool. I think that pattern's emerging.

    7. JFJared Friedman19:29

       It's like, as I hear you talk about that, Diana, what's intui-... I just think, it's just, like, so many different startups that could be built. It just feels like we've never had this moment. At least I don't feel like I've, I've never experienced a moment where there's just so many potential startup ideas to be built, like, all at once.

15. 19:45 – 20:54

    Surge in Startup Ideas

    1. JFJared Friedman19:45

       Yeah, there, there absolutely has been and we, we definitely saw this in the last batch with all the pivoting companies.

    2. DHDiana Hu19:49

       Oh, yes. (laughs)

    3. JFJared Friedman19:51

       People don't always realize this, but like, many of the companies get into YC, within a month after we fund them, they're looking for a new idea 'cause the old thing didn't work or they lost interest in it or something, and it's normally, like, not actually that easy to find a great startup idea for a team to work on, but man, was it easy last summer. It was just there was, like, great startup ideas just lying on the ground.

    4. GTGarry Tan20:09

       (laughs)

    5. JFJared Friedman20:09

       You'd like trip over them. Yeah.

    6. DHDiana Hu20:11

       That was the fast... I think you actually had a tweet about it. I was p-... went pretty, uh, viral that talked about this is the batch, the batch ever in your whole career working at YC where founders got to good ideas the fastest ever.

    7. JFJared Friedman20:24

       And Hart has been here even l- even longer.

    8. GTGarry Tan20:26

       (laughs)

    9. JFJared Friedman20:26

       (laughs) Yeah, no, it definitely feels unique. I've never had so many successful pivots.

    10. DHDiana Hu20:30

        Yeah.

    11. JFJared Friedman20:30

        And Gary, to your point about the ChatGBT wrapper, I think back, like, I feel like that meme really came out, like, just about a year ago- Yeah, let's talk about GPT wrappers. (laughs) Yeah, like G- like-... I feel like the first sort of group of ideas I saw in the batch were all sort of generative AI ideas built on shop- top of ChatGBT, so it was stuff like, "Hey, like, automate your marketing copy or automate, like, your creative content," or something

16. 20:54 – 22:02

    The "GPT Wrapper" Term

    1. JFJared Friedman20:54

       like that. And that term got thrown out. "Oh, these things are all just like wrappers on top of ChatGBT, and, um, OpenAI is going to, like, take all of that. You're just gonna build all these things and then we're gonna release their app store and, like, it's just gonna take all the value and these things will die."

    2. GTGarry Tan21:08

       All of SaaS software is just MySQL wrappers.

    3. JFJared Friedman21:11

       Exactly.

    4. DHDiana Hu21:12

       (laughs)

    5. GTGarry Tan21:13

       (laughs)

    6. JFJared Friedman21:13

       I think this is a great analogy, that you can think about any SaaS product as basically a database wrapper. Like, you could imagine, like, negging any SaaS product 'cause, like, the first version of a SaaS product, it's basically just a crud app, and s- just like you took, like, MySQL and then you, like, built, like, a website on top of it. And I think people are gonna look back on this term, GPT wra- GPT wrapper, like, similarly to how t- we think of, like, how we would look at the term database wrapper, which just seems, like, silly.

    7. GTGarry Tan21:40

       I mean, this is why I think the chat interface is wrong. Like, I actually think there is value accrued to really great UX, like, good copy, good, um, you know, interaction design, information hierarchy, uh, you know, being able to approach a product and say, like, "This is the job to be done," and for users to come in, just sort of naturally understand what to do.

17. 22:02 – 23:39

    Importance of UX

    1. GTGarry Tan22:02

       Like, there is a craft to building software that is timeless and that sort of transcends whether or not you're using a large language model. And so, you know, the- that- that I think is what I mean by, you know, these things are not... You know, SaaS software is not, uh, a MySQL wrapper.

    2. JFJared Friedman22:19

       Well, here-here be a question that I'd be interested in, in, in everyone's thoughts on. Suppose you're a new founder and you really want to build a big company and you want to do something on top of LMs. How would you differentiate between an idea that could be a great foundation for a billion-dollar company and an idea that is likely to get run over by GPT-5 and is probably, like, not a good starting point?

    3. DHDiana Hu22:45

       I think if a founder's working on something too general and not solving a specific need for a user they can actually go talk to in a use case... So, I- I worry about the ones that are too generic and building... going after some kind of abstract-

    4. JFJared Friedman23:04

       Like if-

    5. DHDiana Hu23:04

       ... it will solve all the things.

    6. JFJared Friedman23:06

       Yeah, if it's like, "Hey, like, throw your data in here and we'll do, like, automations on top of it, like, for everything," that's probably hard to compete with whatever one of the foundational models might offer. But if it's like, "Hey, we are... Give us, like, your sales log data and we'll, like, um, spit back, like, suggested next actions that you can... like, for salespeople to make them better at sales," that's probably gonna work better.

    7. DHDiana Hu23:34

       Or, "Give us all your compliance checklist to pass HIPAA compliance and process that."

18. 23:39 – 25:03

    Focus on Specific Problems

    1. JFJared Friedman23:39

       Yeah.

    2. DHDiana Hu23:39

       It's like, that's very specific and lots of business logic. Or, "Give us all of your data for p- processing government forms," right?

    3. JFJared Friedman23:48

       Yeah.

    4. DHDiana Hu23:49

       So it's a lot of custom business logic. So, the same thing with the SaaS era. A lot of the applications and how you build applications in there, there's always the separation of business logic and they crowd in a lot of architectures for these app, and a lot of the value of the company is accrued on that business logic that is so custom per company. And there's a whole pattern of, uh, programming patterns on how people separate those.

    5. HTHarj Taggar24:13

       Yeah, I guess as this all goes multimodal, this is going to get really interesting. So early days, but yeah, we've seen companies work on voice AI apps to be like a sales rep. And I think, um, it's an interesting example of the kinds of ideas that might be possible now with AI, is where you take something like a Salesforce, and you try and reimagine like what would Salesforce do if it were started today with all the power of AI. Well, it'd almost certainly do more than just be like a CRM, right? Like, it would ma- like, it would find who your leads might be. Like maybe now it can make the calls for you. It could like set them up. Like maybe it goes all the way to start, like, implementing s- like, the first version of the product for them. Like, I think it's just like the scope of software you can build with AI now is so big. I think that's another good way to find ideas. Like, look at software today

19. 25:03 – 26:58

    AI Agents and Open-Source AI

    1. HTHarj Taggar25:03

       and reimagine it with the power of AI today.

    2. DHDiana Hu25:06

       Which should fund the number of companies that effectively are AI voice agents for small businesses, because they receive... I don't know, if you're like a flower shop or a AC repair man in the middle of, um, the US, there's a lot of calls for you to schedule and you don't have a lot of stuff automated. And there's these YC companies that are using... they're building these AI voice agents to basically be their receptionists.

    3. HTHarj Taggar25:31

       I know one of our partners, Paul Bouheyd, is quite worried about this actually. He's worried about there's going to be a world of just sort of like all these AI agents that are out trying to do malicious things, and that we're going to need like our own like good defensive AI agents out there, making sure we don't get scammed out of all of our money. (laughs)

    4. GTGarry Tan25:50

       I mean, this is actually why I'm so, uh, an advocate for open source AI, because these things are sort of real considerations. Um, you know, can you imagine there only being one hyper dominant AGI, and it's totally closed source, it's owned by one company, and, uh, you know, it's only available to the highest bidder. And, uh, you know, imagine you being a, you know, someone who just had to go to the doctor, and, uh, on the other end of it is, uh, some health insurance company that, uh, you know, bought the, bought access and blocked it out from an- everyone else. And, you know, you getting on the phone, you're not able to sort of navigate or go against this sort of, you know, impenetrable AGI that is able to sort of get around anything that, you know, your side might throw at it. Like, we actually want, you know, some form of actually equity at the AI level. (laughs) Like, we actually want, uh, you know, not merely the biggest companies to own the most capable AIs. We want all consumers to be able to have, from the bottom up, uh, the same access to that same technology. And that's, uh, you know, the best insurance against tyranny.

20. 26:58 – 30:15

    Resurgence of AI Researcher-Founders

    1. DHDiana Hu26:58

       And certainly, that's actually what a lot of, uh, also not just founders, but smartest researchers who are really at the cutting edge, is I went to NeurIPS this past December, which was incredible to see the energy in there. The conference has grown so much. I think it's like over 10,000 attendees. There were 3,000 papers, more than 3,000 papers accepted. And I think, um, back in 2017, there was only around 600 papers.

    2. HTHarj Taggar27:24

       Wow.

    3. DHDiana Hu27:24

       When I went back in 2010, it was just a, in a ski lodge and maybe like a hundred papers. It's crazy the kind of exponential growth.

    4. HTHarj Taggar27:34

       Yeah.

    5. DHDiana Hu27:34

       And one of the big topics of interest was a lot around AI ethics and regulation and how do we measure that. So, that, that was interesting. Um, but the thing that's different about typically... that was interesting in this conference is the amount of interest from researchers wanting to start companies too. One interesting data point is, um, a lot of this era with GPT came about from one foundation paper, is All Attention You Can Need. It was this paper that got released, got launched in a NeurIPS back in 2017. It was a team at Google who was trying to figure out how to make a machine translation between languages more cheap. Because English translation to any language was actually pretty good. But if you wanted to do, I don't know, German to Japanese, there's not enough data. So they figured out this way to compress data, which became the transformer models for GPT, and it was like groundbreaking, and this is the foundation for LLMs. That paper came out in 2017. And the fun fact, I was just looking this up, out of all those authors, eight authors, seven of them started co- different companies. And all of the companies in total, their ra- their worth valuation more than six billion.

    6. HTHarj Taggar28:50

       Oh.

    7. DHDiana Hu28:51

       And now people are seeing, "Oh, these like industry pioneers did this," and it's creating this new crop of, I think, founders that I don't think would have started, because I talked to a lot of AI researchers, and I don't think they wanted to be founders. And I got this question, "How can I turn my paper into a company?" Which I think is cool because this is like going back to the root of, um, YC fund- funding hardcore technical founders. And I think it's cool to see that energy there. So when we went and host our event, we, uh... I didn't plan and it was like 3X oversubscribed.

    8. HTHarj Taggar29:24

       Nice.

    9. GTGarry Tan29:25

       Standing room only, huh?

    10. DHDiana Hu29:26

        Yeah.

    11. GTGarry Tan29:27

        Yeah. It's, that sounds like really the new Homebrew Computer Club, so NeurIPS in December.

    12. DHDiana Hu29:32

        Yeah.

    13. GTGarry Tan29:32

        We got to mark it on the calendar.

    14. HTHarj Taggar29:34

        (laughs)

    15. DHDiana Hu29:34

        We'll come back.

    16. GTGarry Tan29:35

        Yep.

    17. JFJared Friedman29:35

        Diana, I love your point about how this is sort of like returning YC to its roots. It definitely felt that way last summer, because when YC got started, the internet was really new and the people who were building stuff on the internet were mostly technologists, because it's actually pretty hard to build websites back then and pretty hard to build like good software. And like as building software and building websites got commoditized, a lot more people came into the space.And this is a cool reversion back to the, like, origins where, like, the people who are building the most interesting stuff are, like, mostly really hardcore, like (clears throat) researchers and technologists because there's actually real new technology being invented.

21. 30:15 – 32:06

    Periodic Dismissal of Emerging Tech

    1. JFJared Friedman30:15

       It's not just, like, innovating on business models with the commoditized technology.

    2. HTHarj Taggar30:20

       I mean, and just like every great technology is being dismissed, right? It's going back to, like, the ChatGPT rapper meme. Uh, again, I actually think that was great for YC because it meant we only got the people who are, like, tune- who could tune that out and would just say, "Hey, like, either I'm just so interested in this technology, I don't care, like, what the memes are, or I'm just too busy building it to pay attention to the meme on Twitter," (laughs) which is also great. But, like, th- th- I feel like this has always been the case, right? Like, Homebrew Computer Club, like, PCs are, like, dismissed as, like, toys. Like, the internet is dismissed as a toy, like, all- all of these things, so feels like that moment again.

    3. GTGarry Tan30:55

       Yeah, there is a- a classic, uh, essay that I love that I saw off Hacker News. Do you guys remember this? It's geeks, mops, and sociopaths-

    4. HTHarj Taggar31:04

       (laughs)

    5. GTGarry Tan31:04

       ... in, uh, Subculture Evolution. And, you know, I- I think that that actually is the one thing that's quite durable and, like, keeps returning, right? It's always the geeks who are gonna be into the tech no matter what. They're on the cutting edge. You know, uh, I always think of Steve Wozniak talking about, like, you know, "We started Apple Computer with no idea that it would ever be a company. Like, we just wanted computers for ourselves and our friends." And so, you know, at some point, the, you know, sociopaths come along (laughs) and they start sort of, uh, monetizing the people who, you know, come to the scene, and then the cycle returns and repeats. So, that's why I like being at the beginning of a new cycle. And clearly, AI is exactly that. So don't- don't count it out. (laughs) Don't write it off. (laughs) It's one of the most interesting things that are- is happening out there. Um, but, you know, there are clearly things to be careful of, like don't be, uh, attracted to the new shiny thing. Uh, instead look for the muck because where there's muck, there's brass.

22. 32:06 – 32:26

    Outro

    1. GTGarry Tan32:06

       So, that might be a great place to call it for the very first episode of The Light Cone. We'll see you next time.

Episode duration: 32:26