I remember about a year ago one of these conversations around are we going to have AGI, what would that look like? What one of the arguments for it was that well like at some point the AI will get good enough to just like design chips better than like humans can and then it

will just like eliminate one of its bottlenecks for like getting greater intelligence and so it feels we're on the pathway to that in a way that we just weren't before the last episode we were talking about you know what are you going to do with these two more orders

of magnitude since then Sam has uh told me that he actually wants to go to four orders of magnitude. It's the worst that these models are ever going to be right now, right this moment. Week to week you know there things that you couldn't do maybe a month ago that you could do

really really well right now so that sounds like a pretty crazy moment in [Music] history. Welcome back to another episode of the light cone. I'm Gary, this is Jared Harge and Diana, and at White combinator we've funded companies worth more than $600 billion and we fund hundreds of

companies every single year so we're right there on the edge of seeing what is going to work both in startups and in AI. Recently Sam Altman wrote this pretty wild essay that predicted that AGI and ASI are coming within thousands of days. Seeing him on Monday he actually

directly estimated you know between four and 15 years. Have you guys read this essay yet and what do you think? Yeah, I read it and one one of the interesting places where I think we have a unique perspective is that we were we had a front row seat to the very

beginnings of OpenAI because OpenAI basically spun out of YC and so what was cool to me reading this essay is that it's literally the same ideas that Sam was talking about in 2015 when he started OpenAI, like he's been talking

about this like basically since I've known the guy. Um, and in 2015 when he said these things he sounded kind of like a crazy person and not that many people took him seriously. And now 10 years later it turns out he was right and actually we were much closer to AGI

than anybody thought in 2015. And now it doesn't sound crazy at all, it sounds like totally plausible. I mean the essay itself is pretty much the most techno-optimist thing I've read in a really long time. Some of the things that he

says are coming are pretty wild: space colonies, fixing the climate problem, um, your intelligence onto being able to solve abundant energy. Uh, yeah, I think he's basically ushering in this sort of Star Trek future on the back of literally human intelligence being able

to figure out all of physics. Yeah, Sam was always a re- like, I, I remember back when he was starting OpenAI one of the things that really motivated him to do it was he believed that when we actually had AGI, basically it'd be better at doing

science than humans were and therefore would accelerate the rate of all scientific progress in every scientific field. That was, that was part of the motivation from the very beginning and I think is really connected to 01. Even when Sam came and spoke at our, at our, at

our batch a year ago this is long before GPT-3 was publicly released but it being worked on in you know in uh in secrecy by OpenAI that was the thing that he was most excited to talk about was giving GPT more advanced reasoning capabilities and I think this is the

reason it's because like the thing that's missing from its ability to actually do science and like accelerate technological progress it needs it needs to be able to like think through things one thing that really stood out about um in particular is if you read um one

of the papers talking about it so capabilities and potential for the future it talks about how it does really well in chip design um and I remember about a year ago one of like one of these conversations around are we going to have AGI what would that look like

like what one of the arguments for it was that well like at some point the AI will get good enough to just like design chips better than like humans can and then it will just like eliminate one of its bottlenecks for like getting greater intelligence and so it feels like that's

already kind of like we're on the pathway to that in a way that we just weren't before Diana's going to show a cool demo of it doing exactly that it's fun because we run this um hackathon with OpenAI and Sam came over and judged

the winners and one of the participants was actually chip design this company is called Diode Computer I think we mentioned them earlier what their building is basically AI designer for circuit design and their previous product it could handle in the you think about PCB design there's four major steps the big

expensive part that you need a lot of. All of these need a lot of expertise. It's like the system design. How do you really put together the architecture of it? How do you design all the components, like the resistors, the sensors, the specific processing units, then

you need to go do the layout of schematics, placing them, doing the routing. And routing is known to be an NP-complete problem because as you have different layers in circuit boards, there's interference. And this is why companies like Nvidia, Intel, Apple have gazillion electrical engineers because

this is an NP-complete problem. Up to GPT-4, which is what this company had built, it actually put some constraints and was able to automate a lot of the schematic design that you as a human had to design, what components it needed to go and the design and to some extent the

routing. It was simple, which is still pretty cool up to that point. So they were able to automate all that, but the thing that they demonstrated now with O1 was actually able to do the system design and component selection, which is crazy. So it would be able to read all

the data sheets and select the right components. So the way the product would work, it could say, "I want to build a wearable heart rate monitor with an accelerometer and a microcontroller." Very high level, and given this constraint and looking at the database, it would be able

to match the specific accelerometer and microcontroller and heart rate monitor sensor and connect it and just output the end result: what we are trying to build today is a wearable heart rate monitor, something like you would see in a Whoop, for example. Um, the O1 is amazing.

but one of the downsides is that it's a bit slow so we actually cached a pre-generated like system diagram the other one was able to generate it it's pretty good it has a USB-C connector an IMU like we requested a heart rate sensor um and uh like this is a

microcontroller so I'm going to show you how you can go from this uh and like build a PCB so we are going to like build the project um the output of this is code we actually use Aril which is a electronic HDL code uh language and you

can see that it took all the blocks in the block diagram stitch them together exactly how we want um and the second step is it actually is going to generate a layout for the board um and so now like we can directly open it and here you go here's the board it's uh quite

it's quite nice uh there's still like a couple of like fine-tuning steps required for example um we could like move like this USB type-C connector slightly um we can like change the shape of the board but but these are all the

components um and then like thanks to the system that we built uh we can like call the auto router on this specific board and actually get a fully working uh printed circuit board back so this is uh actually one of the examples on the 01 paper that it would do EDA but

actually they went a step further because the example on the paper they described the EDA a step process is this set of tools for circuit design it does sort of the design of the schematic also the simulation and bug verification it's easier to verify stuff than to select

and write it so this company actually went a step further beyond the paper because the paper did mostly the last stages of uh verification and simulation I guess it's an interesting example of using different models for different tasks and in different workflows so in order to

pick the correct components off the bat you know even before you you know place it on a circuit board you've got to actually have probably rag on structured you know taking unstructured data like PDF documentation and turning into a structured form that then 40 mini sounds

like is being used to actually extract the data and then put it into um format for 1 I think this is actually a very common pattern that we're seeing a lot of the interesting products built with they at you use different kinds of models so yes 40 minis for PDF

extraction and then 01 for the reasoning because it's actually very hard to select the components for parts I know jar also work with a lot of hard tech companies and the whole part of selecting whatever the servos the motors the sensors it's like so is it takes a

lot of thinking for human yeah the the other thing I think is interesting about this example is like during the batch before 01 came out diode had tried to do this with GPT 40 and it just flat out didn't work and then they basically

tried the same thing the same prompts but fed it to 01 and boom all of a sudden it worked and so there really is a sort of like step function capability unlock they were so excited when I talk to them and they show me they were they had this

big smile It's like wow, they themselves were super impressed This hackathon that Diana ran, incidentally, I think it was a really interesting concept for a hackathon Like most, most hackathons are like people who are just sort of like building something that they plan to

throw away Um and the cool thing about this hackathon is it was all actual YC-funded startups that have real businesses that have funding that have like a real thing with real users and they were all building actual features for their product that they plan to

release to real users It was, it was really cool I think for us to see how 01 unlock capabilities for real companies, not just like toy projects Y There's another one that was similar in here in terms of reasoning for a one I think, Har, you work with uh, Camper Yep So want to

tell us what Camper does Uh, it's I mean the tagline is Dev for Cat But basically they um let you create CAD designs with just natural language Like you just type in like you know something that you want to design and it just like spits out

like a CAD design for you So, can you design me five airfoils optimized for 50 mph with a minimum drag to lift of 15 at a 5° angle of attack That's very specific Y Normally this would require a actually a mechanical engineer to be

running all the simulations and solving through the equations and what you're seeing why it's like flashing is like running all the multiple simulations, four of them at the same time So it's actually kind of like a co-pilot to uh, Solidworks Yeah, they actually built

There, like initially, they were going to build this as a plugin to SolidWorks, but they went for like the even harder technical approach, which was like, 'Now this is just like an executable that runs on desktop and it essentially opens up SolidWorks for you, and then just'

starts like clicking around in the UI pretending to be a person, you know? Nice, and you saw there what was really cool earlier, they flashed the math Trace, so 01 was actually able to write all of these equations, all these partial differential equations and solve basically Navier-

Stokes questions to actually solve airfoil. That is really cool. The last episode we were talking about, you know, what are you going to do with these two more orders of magnitude? Since then, Sam has, uh, told me that he actually wants to go to four orders of magnitude

to get to a trillion dollars in, uh, you know, sort of spend, I mean, pretty wild. But on the other hand, like you could see where that might go. You know, you could imagine an airfoil is still, it's, you know, very impressive and complex, but

sort of what we're capable of doing today in 2024, you could imagine abstracting that to like understanding the nature of physics. I suppose like it, it would be sort of hard to see that maybe in the current version of 01, but if the scaling laws hold, it seems

entirely plausible that you know far more difficult engineering challenges such as, um, you know, room temperature fusion, like these are all sort of ultimately engineering fluid mechanics. There's weather prediction, there's all these complex physical phenomena that are very hard to solve and you need basically PhDs and

to Sam's essay, this is a glimpse into what AI and where 01 is heading with this Chain of Thought and reasoning, especially like, like Sam says, the vibes are sort of training intelligence and this new age of intelligence, and then the 01 paper

just that I think this whole idea of now you can actually give like feedback, not just on like the outputs and whether you got the correct answer, but like on all of the steps to get there and like you're basically teaching a model how to think, like the camper guys are

mentioning too, right, the reasoning traces and they can probably go back and like fine-tune the various steps for like every output to make sure that the model's thinking how they want it to think, that one that just is again very like the AGI conversations I feel like a

year ago were all sort of in this direction of like what happens once you can actually start teaching the models to think better versus just like, um, spitting out the correct answers. Uh, and then the scaling laws, this is like even more surface area for like throwing

compute at the problem, right? Like now you can just basically put compute at the inference step and and iteratively have something come out that you know you can actually spend more money and more time and have a result that iteratively gets better, similar to what you might expect from a human scientific

organization. Yep. Maybe more consistently, even. Do you want to talk about the architecture and how they actually created A1? I think a lot of it is inspired from what they've been working for many years since the beginning of, uh, OpenAI. I think one of the inspirations is a lot

of the work they did with DOTA, yeah, does everyone remember when like before OpenAI was famous for GPT, the one thing that they were like kind of famous for, that at least people in the tech industry knew, was DOTA, was like winning video game

competitions, that was their first big breakthrough, and the funny thing, I think back then, DOTA wasn't something that took the world for storm. I mean maybe only the research community kind of knew about it, but it wasn't anything practical, but what was impressing was

beating a lot of the best DOTA players. So DOTA is this complex game of resources and planning, right? And they implemented a lot of, uh, kind of reinforcement learning type of techniques in there, which I think were also inspired early days from AlphaGo

and AlphaZero as well, on how it solved Go. It wasn't just brute forcing through it, but actually having a reward function and and trying to solve towards it, and even this is why there's just so much talk about Q-learning, because that's sort of the fundamental algorithm behind

the family of algorithms behind RL. So yeah, so like because of DOTA, they got really good at doing reinforcement learning. That's how they got it to work, they just had it play against itself like a million games, um, and then, how does that connect to to to GPT-1? So I think

this is where there's a bit of a big stepping function because how do you then incorporate that into the family of GPT type of model? GPT is all generative based on predicting the next token and patterns and then getting those results to check that they're correct. So I think

a lot of it is you had to have a lot of data that was factually correct and fed into probably the model and the training and having a reward function that gets it to reason a bit more about the output and make sure that it's

correct so they probably done a lot of um interesting techniques with that and there's probably a lot of secret sauce on the type of uh data sources they done maybe one of the speculations we could do is a lot of very factually correct information like math problems and

science problems and things like that yeah and that's why I outperform so much on those yes yeah one of the things I think is interesting Gary to your point about the scaling laws is a lot of people are really focused on the next like scale up of the model like the GPT

5 series of models which are being trained now and people are working on them and they are going to come out um but I think people may be underappreciate how big an unlock this other direction is because there's there there two research directions being

explored in parallel right like one is the straightforward scale up of the underlying LM and then this 01 direction is like a totally orthogonal research direction in which you unchain the model by having it do reinforcement learning while actually trying to do things in

the real world and getting better at them the version that's come out so far it's still only 01 mini and if you look at the actual sorry 011 preview and like if you look at the performance asset they released like the full 01 model which is

coming out any day now is a huge step function above even 01 preview which is what enabled all these incredible results of the hackathon Sam was just telling us that like 002 and 03 are not far behind and so like I think people

maybe under appreciating just how big an unlock we're going to get yeah and 01 also is really opaque still I mean if from a you know sort of business perspective this is a new method I think at Great cost to themselves they actually did create a new data set to

train the chain of thoughts you know it's essentially a giant data set of you know given task X can you break it down and uh into you know break it down into parts and you know what's funny is this sort of rhymes with what Jake uh Heller

figured out for case text that if uh a given task that you give an LLM is hallucinating or you know not consistently giving the output you want you're trying to make that particular prompt do too many things you need to break it down into steps and so what's

funny is Jake's prescription is really two parts you know one is break it down into steps and then the other part is evals and it sounds like basically with 01 the chain of thoughts will replace the workflow so you might not need to break it down into steps yourself but the

evals are still really important um even like in the aftermath of that episode with Jake Heller it sounds like some YC alums are reaching out and saying that episode helped us figure out and unlock something really big like a lot of people really were just raw dogging

their proms yeah they got to uh you have an example of the company you work with J that got to 100% yeah just by doing exactly what Jake recommended which is like having a really big eval set and being very careful about testing every

step of your reasoning pipeline so one of the theories that I have now is ultimately like if you superimpose that on what is a moE I mean that's one of the key questions that everyone's sort of asking themselves right now you know

okay like GPT-5 is coming two more orders maybe four more orders of magnitude are coming in terms of a trillion dollars spent on more training that's pretty wild you know if I'm a rapper company or I'm trying to do vertical SaaS or I'm trying

to you know build my own business what do I do my theory would be it's the events it's you know write the 10,000 test cases and the only way you get access to the test cases that are proprietary data that are not a like

commonly available is that you literally you know that's what a bunch of our companies in this current YC are doing they're doing the hard work of doing Enterprise sales they're getting embedded and sort of going quote unquote undercover into these uh you know

sometimes really boring jobs sometimes really complex or Arcane jobs you know it's everything from you know I think accounts receivable all the way over to how do you do like Financial Accounting or forensic accounting like it's just all kinds of things that are really uh

not readily available. Um, you can almost argue that anything that is consumer and publicly available on the internet, that's going to be in the base model. So then your MO ultimately is for all of the other things that are not already online, whether it's, you know, for

case, like, being a lawyer or maybe over here on science, or you know, in terms of building airfoils, like what you're trying to find is the data that is proprietary in some use case, some vertical that allows you to build that 10,000 test case. Um, even Val, and then

that's actually the value. I mean, this is just a crazy theory, but it's something that might, might be what happens. Totally an interesting implication for startups based on everything you just said. Um, is it may be worth thinking about who, like, of your customers, picking the ones that

will pay you a lot for that final like 10% accuracy, like, imperfection. I think like Camper are actually a good example of it, where there's lots of interest in this sort of text to CAD design amongst like hobbyists or people who want to

prototype things and get something up and running very, very quickly. Um, but there's also like a segment of the market where it's people who are literally designing like, you know, airplane parts, where there is no room or like margin for error and 01 makes it

quite easy or easier now, right, to get to like the prototype, like, you know, 80% of the way there, but I think the strongest technical teams have the option to go all the way and go after the segment of customers who want like a 100% accuracy.

and will pay a lot for it always go all the way yeah to go all the way but I think it's interesting because one of the things that gets pushed is does 01 or does AI in general actually make it commoditize a lot of the tech and

make it less important to be a strong technical team um and it just seems unlikely to me it seems like actually the last it seems it's the opposite yeah yes like all of the value is probably going to be captured by like the strongest technical teams who can build

on top of whatever the base level of tech is and get the final 10% yeah Gary I think it's the prompts it's it's the evals and it's also like the UI layer and the integrations that go around it because like just the prompts themselves

are not a product for a company to actually adopt camper like it needs to actually integrate into their existing tools it needs to have a well thought through UI and workflow and all the tooling to sort of make the prompts useful yeah well and then it's

distribution right like how do you actually get in front of people do you establish your brand and then uh a perfectly good moat is difficulty switching actually and once you have all your data and it's working and you're paying $10,000 or $100,000 ACV sometimes

a million to 10 million ACV uh you know man it's going to be hard to switch so all the classic modes still apply you know this is still software but you know you can unlock this capability and you know this is a moment you know another

point to double down on the importance of evals is that that still applies in the world of AI as Founders are wondering how am I going to still build the best product on top of it or does it change and everything we discussed in

the episode with Jake Kerr applies because Gigl is this company that uh, you work with, yep, and Gary too, Gary right? I adopted them, maybe tell us a bit about what they what they do. The full full back story is we funded them for a completely different

um idea. Something like they're an Indian founding team and the original idea was something helping Indian high school students apply to US colleges. A very like niche idea but they're super cracked AI engineers, researchers, yeah. And it just it just happened like we

were like this is not a great idea. AI is you know changing the world and you like you're research like that you've been doing at University or college is all aligned with like um, in like fine-tuning LLM models. Originally it wasn't even the AI version of helping

Indian high school students apply to, nothing to do with AI actually. So it's like a classic YC story. It was like these are two clearly brilliant engineers, um, we don't like the idea at all, but we should just fund them anyway and hope something works out. And the idea they

actually pivoted to initially, which they raised a seed round for, um, was helping companies fine-tune um open source models so they could get to like equivalent performance um uh as you know at the time it was really only OpenAI but um what I think in general what we

found is that those have not proven to be great businesses because just like the cost of the models has gone down and AI, like you know, the performance of the open source models has gone up, you just haven't had to fine-tune as much as people thought you would need to because

the models are just keeping getting better. It's kind of betting on the different, on the opposite direction on AGI that let's just trust that these models are going to keep getting better and better, which AI doesn't require as much fine tuning. Yeah, and so they pivoted again into, well

like let's just find a vertical, like let's just find like a we're really good at AI now, we're like we experts in like finding, squeezing performance out of these models, so let's just like find a vertical application for that, and they

went into um AI customer support, which is like competitive, but again I just think if you're an intensely technical team you can still find ways to squeeze out like a comparative edge against other teams in the space, and I think that's what they've done. The problem with

Why The Next AI Breakthroughs Will Be In Reasoning, Not Scaling

Videos to help you build a successful startup. Subscribe for startup advice, founder stories, and a look inside Y Combinator.

What is Y Combinator?
We invest $500,000 in every startup and work intensively with the founders for three months. For the life of their company, founders have access to the most powerful community in the world, essential advice, later-stage funding and programs, recruiting resources, and exclusive deals. 

Visit ycombinator.com to learn more.

Video Team:
Zach Both - Video and Content Lead
Ryan Loughlin - Senior Producer



Y Combinator

Discussion on the possibility of achieving Artificial General Intelligence (AGI) and its potential impact.

Analysis of the rapid advancements in AI capabilities, exceeding initial expectations.

Sam Altman's prediction of AGI/ASI within thousands of days, later revised to 4-15 years.

Discussion of Sam Altman's essay predicting AGI and ASI, and the unique perspective from Y Combinator's involvement with OpenAI.

Comparison of Sam Altman's current predictions with his earlier statements from 2015, highlighting the increased plausibility of his vision.

Discussion of the optimistic predictions in Sam Altman's essay, including advancements in space colonization, climate change solutions, and energy.

Sam Altman's belief that AGI will accelerate scientific progress by surpassing human capabilities in scientific research.

Focus on OpenAI's work on enhancing GPT's reasoning capabilities to facilitate scientific advancements.

Highlighting GPT's proficiency in chip design as a step towards AGI.

Demonstration of AI's capabilities in chip design using a company called Diode Computer and their product built with GPT-4 and O1.

Explanation of the four major steps in PCB design and how AI automates these processes.

O1's ability to automate schematic design, component selection, and routing, solving the NP-complete problem of routing.

O1's capacity to handle high-level design requests and select appropriate components based on provided constraints.

Demonstration of O1 generating code and a PCB layout for a wearable heart rate monitor.

Applications of AI Models in Various Tasks

Discussion on using different models for different tasks within a workflow, such as using 40-mini for PDF extraction and O1 for reasoning.

Comparison of using GPT-4 and O1 for component selection, highlighting O1's superior performance.

Description of a hackathon where YC-funded startups built real features for their products using O1.

Discussion on Camper, a company that uses AI to create CAD designs from natural language prompts.

Camper's use of O1 to solve complex equations and simulations for airfoil design.

Discussion on the potential for future AI advancements based on scaling laws and increased computational power.

Exploration of the possibility of AI solving complex engineering challenges, such as room temperature fusion.

Connection between Sam Altman's essay, O1's capabilities, and the concept of training intelligence.

Importance of providing feedback on the reasoning process, not just the final output, to improve AI's thinking capabilities.

Discussion on the potential of scaling laws to improve AI performance through increased computational power at the inference step.

Discussion on the architecture of O1 and its inspiration from OpenAI's previous work on DOTA and reinforcement learning techniques.

Explanation of how reinforcement learning techniques were incorporated into GPT-type models.

Importance of factually correct data and reward functions in training models to reason correctly.

Speculation on the types of data used in training O1, including math and science problems.

Discussion on the two research directions explored in parallel: scaling up LMs and developing O1-type models.

Implications for Startups and Future Trends

Discussion on the implications of AI advancements for startups, focusing on the importance of proprietary data and creating comprehensive test cases.

Importance of focusing on customers willing to pay for high accuracy and the potential for strong technical teams to capture value.

Discussion on whether AI commoditizes tech or enhances the value of strong technical teams.

Importance of prompts, evaluations, UI, integrations, and distribution in building successful AI-powered products.

Discussion on Gigl, a company that helps fine-tune LLM models for specific applications, and their pivot to AI customer support.

Gigl's success in automating customer support tickets for Zepto, highlighting the impact of O1 and the importance of evaluations.

Gigl's significant improvement in accuracy from 70% error rate to 5% error rate using O1.

Discussion on the challenges and opportunities in AI customer support, focusing on handling complex edge cases.

Discussion on the need for interpretability and editability in AI models to enhance usability and fine-tuning.

Discussion on the current state of AI models and the rapid advancements being made.

Analysis of which industries might be negatively impacted by advancements in LLMs.

Discussion on the potential impact on AI coding agents and program engineers.

Discussion on the opaque nature of Chain of Thought and the challenges in making it directable.

Discussion on new startup ideas unlocked by recent advancements in AI models, particularly in phone calling applications.

Discussion on the potential of AI to improve various engineering fields and create real-world abundance.

Why The Next AI Breakthroughs Will Be In Reasoning, Not Scaling

스크립트