# AI Shifts Computing From Engineering To Capital Constraints

**Podcast:** a16z Podcast
**Published:** 2026-08-25

## Transcript

Right now, if I give 20 people a billion dollars, they can actually use it usefully.
We've kind of moved the industry from like this engineering-bound problem to a capital problem that's fundamentally very different.
Math is very much a leading-edge indicator of what the market might be interested in and why.
Some people will walk in and say, the foundations to AGI and to reasoning is going to be math.
But like, that doesn't tell you anything about reality.
For me, it's still in the domain of like, it's really good at playing a game.
The startups don't aim straight at the incumbents and the incumbents just don't pay attention.
Microsoft is worried way more about what Amazon and Google are doing than anyone in a startup space.
Everybody who's from a big company in Silicon Valley, you always think, oh my God, we're just going to crush all of these little companies.
And then you realize they never get crushed.
And I think this is why we're seeing...
such meteoric pros of the cursors, the anthropics, and the open AIs.
Although capital is scarce and it's hard to get and all of these other things, once you get it...
For most of modern computing, the bottleneck was engineering.
AI may be turning it back into a capital problem.
In this episode, I sit down with Martin Cassato and Steven Sinofsky to ask whether AI is changing some of the fundamental assumptions we've built up over decades of computing.
They start with AI's recent progress in mathematics.
What these breakthroughs actually tell us about reasoning, why mathematicians are paying attention, and whether math is a leading indicator for where AI creates economic value.
From there, they easy amount to a much bigger shift.
For decades, giving a small engineering team vastly more money couldn't make them build vastly faster.
Today, a team of 20 can productively deploy enormous amounts of capital into compute.
Martine and Stephen explore what that inversion means for startups, incumbents, venture capital, And what happens when previously intractable problems can increasingly be turned into capital problems?
Well, Martin, when you're not making major acquisitions or having big news, you're also very curious about what's going on on the frontier of AI.
And we're actually going to start getting into math.
So first off, thanks for both of you making time.
It's great.
Jared Sumner tweeted a few days ago something along the lines of how he told Claude to try to solve the Riemann hypothesis and to try harder.
And I don't know if there was actually any progress made, but it's part of the larger conversation around, hey, it seems like there's some accomplishments that are being made.
How do we make sense of this in terms of what is actually happening and what does it mean for math?
I'm no mathematician at all, but I mean, I think it's an important moment because it sort of divides the world into two groups.
The groups that are just very, very excited.
That, oh my God, these things are being solved.
It doesn't matter if you understand them.
Actually, nobody understands.
The universe of people who understand what these things are is very small.
And then there are the people who are just like, oh, it's fake.
It's going to put people out of jobs.
Then no one's going to know the future of where these fields go.
And the most interesting thing about it is the group that's most excited are mostly the mathematicians.
And so that actually confuses everybody.
Because if you're of the school of the people who are like, it's going to put people out of work and we're going to all get dumber and it's the dawn of idiocracy because computers are doing all of our work, you're confused that the people who are impacted most by what this level of AI did are the most excited.
I think that is itself shining a light on this moment that we're in right now.
You're talking to two systems, guys.
Two product guys.
I have the same caveat.
I feel there's some things like we're actually...
both very expert on.
This is not one of them.
So I'm going to kind of, from the peanut gallery, I've got two comments.
So one of them is, okay, so I view economic utility to be a very important measure when you're talking about AI, right?
So I was trying to think.
There's a lot of hours been trying to solve some math thing, right?
But if you sum up the entire postdoc salaries of all the people that have been working over the years on these problems, it's probably not very much.
And so part of me is saying it's great that there's these capabilities.
I'm not sure that the fact they've been longstanding is that much of an indication because there hasn't been a huge economic incentive in order to solve.
Now, that doesn't mean that it's not hard or whatever.
I just don't think we have that validation of this unlock some deluge of economic value.
And the second one is, it's kind of not surprising to me that AI is very good at solving.
almost purely axiomatic domain that, you know, requires knowing a whole bunch of different things and putting the solutions together from very disparate spaces.
Because often really when I read, so I've been reading all these, like everybody else has been obsessively and they're like, oh, like it came up with the solutions.
Yeah, the solution was pretty straightforward.
It's just like barred from a bit of math that I didn't know.
And so I think if there's like a meta learning here, the meta learning is there is a set of problems that probably require you to be too broad for most humans or most education, and it's going to solve those.
It's clearly very good at solving axiomatic systems, but I don't think it provides a strong indication of, is this solving things that the market hasn't been able to solve?
Because there really hasn't been a market around these.
And so I think that's the next question for us to answer.
So very exciting.
It seems kind of reasonable, understandable.
Not sure what the longer-term implications are.
I do think that there's something interesting, that math is very much a leading-edge indicator of what the market might be interested in why.
I remember when I was in school, like there was some big thing that someone at AT&T invented a new algorithm, like a new program for doing linear algebra, like a new way to solve linear algebra, which is super important right now in the AI world.
But his big thing was, well, now we can just calculate like the United Airlines flight map in three hours less time than we could.
last week.
Right, but let's take it to us.
So it's just not clear to me that the problems being solved are those that are roadblocks to like existingly economically useful tasks.
Right.
And if they were, it's not clear to me that they wouldn't have been solved.
Like post-doc that's been ruminating on a problem getting paid 30K a year for five years, like it's very different than...
Like the market has decided that this is like the one thing to unlock.
And maybe they're there.
Maybe these problems that are being solved are like the key problems to unlocking some big economically productive use case.
I just haven't seen that yet.
So that for me is like the next thing I'm kind of looking for.
I don't even know what 12 dimensional space is or what that means.
And so like I'm completely with you.
Like I don't even know what problems are in 12 dimensional space.
Are you very skinny?
Are you very tiny?
I'm really confused by that.
And maybe I'm wrong here, but for me, it's still in the domain of it's really good.
playing a game.
Yeah, yeah.
Like, this is the best StarCraft player ever, which is cool and it's very powerful, but, like, I have a hard time connecting that with, A, like, maybe the reason we didn't have them before is because there just was an economic need, and B, like, how does that actually map?
And so, listen, there's a huge range of these things.
We get pitches all the time.
Some people will walk in and say, the foundations to AGI.
and to reasoning is going to be math.
And once you do that, you'll be able to answer every question because the universe is based on some fundamental mathematical principles.
And once you understand that, you understand everything.
And then there's other people candidly that walk in the door and they're just like, listen, that's great, but that doesn't tell you anything about reality.
And so I think that there's more work to do.
And this isn't just about getting better at math.
I do think what's interesting is that part of the reason that mathematicians are very excited about it, though, is because they work a certain way.
If you work in history, There's basically no abstraction in history.
There's just a bunch of facts.
And then people develop sort of these models that you can think of almost as force diagrams that explain war or famine or whatever.
Whereas mathematicians and mathematics has this super long historic arc of layering on abstractions after abstractions.
Don't worry, we'll get to OSI in a minute.
But this idea that why they're so excited is a bunch of math all of a sudden becomes a new level.
Of abstraction.
So is it true that they're, so I've found that it's a mix.
I've found that some are very excited and some are in an existential crisis.
The ones that are excited, they basically say, listen, it solves 20% of my job.
Is it 20% I didn't like anyways?
Yeah.
So this allows me to explore a new frontier that's very important or whatever.
And what I've always wondered is, is that a function of the type of problem being solved?
I just can't imagine if AI came and solved cancer.
like, whatever, someone that works on Castro would be like, oh, I'm so existentially depressed.
This is amazing.
But let me, we're on the other end of, oh, we saw this math for a while, I'm so depressed that saw the math problem.
Maybe, like, literally the entire utility of that problem was keeping somebody employed to solve the problem.
Or just writing articles in the back, like, another attempt at, and here's why I went wrong.
So let me offer it this way.
There's nothing on the other side of the solution.
And so, like, we're depressed because, like, now, like whatever, like this useless activity is gone.
Let me, let me stop.
No, I mean, I'm too cynical on this thing.
I love math.
It's great.
I think we caught you, but like being a little cynical, but not really, but more that it's just, let me take a side of it this way, looking at the history of computer science.
Because I had to take this class, which I looked at all the course catalogs for a bunch of schools.
You don't have to take it anymore.
That was like discrete math.
Yeah, yeah.
Or algorithmic complexity theory, which was a required class for a very long time.
And now— Do you remember Concrete Mathematics?
Oh, yeah, yeah.
From Donald Canoes?
From you.
I mean, you're a Stanford guy.
I'm not.
But, like, I stayed at school.
We didn't have that.
But that's a Cornell joke for us Cornellians.
But my class, I got taught by one of the luminaries in the field of algorithms, ironically, a Stanford PhD, John Hopcroft.
Oh, of course.
Who invented— For the people who are pragmatic, invented like two, three trees and a bunch of stuff.
As his thesis at Stanford.
That's a legend.
But John was our professor in all this crap.
And we had to learn like all this P equals NP stuff.
And I remember like this is the four- Is that the four-color?
This is the four-color.
Proven by computers.
No, but that's where I'm going.
You just buried the lead.
But like so those of you that don't know, we had to take a whole course in college that basically boiled down to this problem.
And the interesting thing is why?
And it was because.
The theoreticians have postulated that if you can solve this problem in an algorithmic, in exponential, in non-exponential time, in polynomial time, then you could solve all these other problems, like the traveling salesperson problem and all these other problems much, much faster, which mattered because all of our computers were just so compute bound.
So if you were the AT&T people that gave, you know, like, here's our node of like 6,000 switches.
Like, how do you route optimally?
You'd be like, well, we don't have enough.
That's like two years of running the simulation to solve this.
And so it turns out that one of the interesting things was they proved the four-color theorem, but they did it.
Which, by the way, I mean, just the four-color theorem says for any 2D planar map, you can color it.
You can use only four colors such that no two adjacent areas have the same color, right?
Exactly.
And you only need four colors.
You'll never need five colors.
And we learn it just so people, like, you kids know.
That's literally how we learned it, and we could all repeat it like that.
It's this very weird imprint over this problem.
And so what sort of happened was no one ever arrived at a basically what you could think of as, like, a proof that looked like calculus.
Instead, what they did is they actually proved that...
the number of potential solutions was finite.
Have you actually seen the proof?
Yeah, yeah, yeah.
200 pages of combination.
But they basically prove that there's a finite number of them, and then they just computed all of them and said, look, it's only four colors.
And so it was this sort of bank shot proof.
But it was only possible because of compute.
And to your point, that was actually very, very useful in the practical application.
Right, right.
And certainly as a topology person.
Like setting strong bounds and things like that.
And I think that that...
to me, was just a really good lesson in when you have like a new level of abstraction that says this is a whole class of problems that can be solved.
You can then build tools working at that level of abstraction.
And everybody doesn't have to start from like, okay, what's the two, three, three representation of what we're doing?
So it's hard not to get philosophical when you're talking about AI.
So I'm going to get philosophical, and you can tell me to shut up.
But I just can't.
Like, you kind of do.
So this math thing seems to me a little different because it kind of begs the following question, which is, will math ever be represented a physical phenomenon, right?
Like, has anybody ever, like, taken a bunch of equations and actually predicted something, like, physical?
And I don't know the answer to that.
So I worked in these large simulation codes, and these large simulation codes are actually...
trying to compute physical phenomenon like the explosion of a star or like, you know, what would happen to like, whatever, an airplane in like an air simulator or a wind simulator.
But all of those, and even though they're just calculating these like large, you know, differential equations.
They were all based on empirical results.
Yeah.
Like, literally, the equations of state for the— Well, they were a model.
They were like, we can measure temperature in these places.
That's exactly right.
So it was all based on empirical equations of state.
And so I've always wondered, like, is simulation computationally irreducible?
And so you actually have to actually run the simulation.
In that case, it's not clear to me to what extent AI helps.
Like, I know people are trying to solve this problem with AI, but, like— I don't know if these math answers have any impact on that type of stuff, right?
So maybe there's some separate algorithmic domain, to your point, where they do, or maybe like modeling or logistics.
But when it comes to like, you know, will this star explode?
Will this building stand up?
Like the actual simulation?
I think these things are pretty disjoined.
And then I read a lot of these discourses on the math solutions, and there's kind of these claims where if it can solve all math, you can predict anything.
And I just think that that's a huge, huge logical leap, which is not clear to me that it's obviously true.
Or there's any indication it's true at all.
So the way, one way that I think I might...
talk about that, you know, again, like, this is so out of my league on the actual math.
This is what I'm doing to systems people.
But I'm inherently a tools person.
And so I kind of get this part of it, which is that what's happened is that AI might not be the next tool to solve math problems at some scale that matters.
But it might lead to the development of a new level of model.
And so I brought, like, props to show this off.
So, of course, this is the original map tool.
And so before something like this, this is, you know, one of these real ones from, like, Beijing, Marco.
Oh, it's an actual.
You know, it's the ones they tell tourists.
But I'm very proud of that because I negotiated it down to, like, seven cents.
But, you know, that became a level of abstraction.
And all of a sudden, like, you just had this basic math thing.
And then you just fast forward a whole bunch.
I brought this because it's just so freaking cool.
So this, everybody knows what slide rules are.
You know, nobody knows how to use them.
This is called a Kürtta, which is an Austrian, basically it's a round slide rule.
And so it's like a coffee grinder or a pepper mill.
And you have all these ways you set the numbers on the side, and then you turn it one way to add, another way to subtract.
And it's this thing.
Wait, is that used for, like, multinumber arithmetic, or is it used for stuff like logarithms?
No, it's only arithmetic, I think.
But, of course, it depends on how you use it.
But it's from the mid-20th century, I think.
And my uncle brought this back from the war.
But what's incredible is this is, like, 600 pieces of machined metal inside this.
It would cost, like...
$50,000 to make one now.
Do you try to use it?
I actually did, but I'm not going to try to do it.
I actually, for prepping for this, I wouldn't be a complete moron.
Just go, ooh, look what I have.
I actually went through the trouble of learning how to use it, although it's been sitting on my shelf for years.
But it's, but the interesting thing is, you know, then all of a sudden a whole new level of problems gets solved.
So you're saying that the new model is the new calculator or the new graphing calculator or the new, I actually remember when, like, remember the TI-85?
Oh, of course, yeah, yeah.
I remember when that came out.
They're like, all of the math teachers had this crisis.
I'm like, you know, we used to give you a piece of paper where you plot the X, Y equation.
Now they can do it on the calculator and they can solve equations and our field is dead.
But what's interesting is, this is why it's so important to AI today.
Those people didn't complain about when calculus came out because calculus was a baseline to them.
And what it is is there's this notion, this people react to change more than they react to the baseline of where they all started.
And so, so much of, like, the concerns in a—I lived—I literally got, like, the TI-35 were the first calculators in schools.
The only advanced math they did.
They had a percent key and factorial, which we didn't even know what it was.
And you could do, like, 59 factorial, and that was the max that you could display.
And, you know, I went to college, and the classes were no calculators allowed.
The whole—I was on that—my whole life I've been on the cusp.
of allowed and not allowed for everybody.
I was there for the graphing calculator.
Oh, the graphing calculator.
You literally have a blue book just to show all of your work, just to show that you weren't plugging it into the graphing calculator.
See, I missed the graph.
Most of us were like actually writing video games in the back and could care less about its ability to write math.
Absolutely, absolutely.
But you just play that backwards and you realize that after, you know, after these guys, you went through this march of...
of algebra and then linear algebra and then calculus and, you know, all of, and then, you know, and then with calculus, then you ended up with Fourier transforms and thermodynamics and all of that was first to your earlier point were all based on need.
I mean, so much of this map.
Well, all of computers, computers are basically from difference engines which are just trying to calculate integrals.
And, of course, but to be really clear, to calculate integrals so that we could shoot missiles and cannons at each other.
Okay, so that's what, yeah.
Which I'm not judgmenting it.
I'm just saying.
Well, yes.
I don't mean to be pedantic about this, but it's one of my favorite parts of history.
It actually started with tides, which also has massive economic value, which you're trying to calculate the tides.
And this is where you kind of have like the old...
You know, and then that, those architectures got co-opted into, of course, the Ward effort for the logarithms for that.
That's what Enneac came from.
It's just great.
Actually, it was very interesting.
Enneac was about 5,000 times faster than a human being when it came to, like...
you know, doing this.
And then, of course, Mark...
And it didn't make mistakes, which was sort of the...
But it was very specifically math.
It was very specific for economic utility.
And the interesting question to me is, is like, these models are clearly going to do a type of math.
Is it one that has somehow blocked some sort of economic?
Yeah.
And I don't know of the answer to that.
Oh, yeah.
But I think it's super interesting to keep going with that because to me, what's so cool is that doing that basic calculus...
for the war and making those missile tables and things like that.
Then it unlocked the space race, basically, and jet engines and factory automation and all of these things.
And, you know, people were cheering that on.
Like, that, to me, culturally, is the most interesting thing.
Like, there was just this, not only were they cheering it on, every parent was looking at their kids saying, go learn that in school.
Go win the Westinghouse competition.
Go win the GE math competition.
Was that because of the Cold War?
Was it because?
Well, obviously, the Cold War was a big cultural part of it, for sure.
But it was just a general, the future.
I found this incredibly cool brochure from IBM.
It's from 1953.
Do you just have this stuff in your house?
I just stumbled across it.
This one I just got.
I can't even believe this exists.
But this is a brochure about the future of this computing.
I want to see it.
But like...
First, you got to look.
It's got like nuclear, like the whole thing.
The future of computing is like a guy with like atoms racing around his head.
Oh, wait, we're zooming in and doing the Carol Merrill thing.
So, but the fascinating thing is it's from 1953.
So your ENIAC point, like that's it.
That's the computer at the time.
This is pre-704, pre-370.
And so it's a brochure from IBM explaining.
what a computer might be, not even is.
And it's like, it took millions of years to invent and recognize the usefulness of the wheel.
That's the opening sentence of the, and, but like people were eating the stuff of it.
But here's the part that I want to get to.
It talks about computers and it's the two families of computers.
And so of course you get the slide rule and that's explaining the history.
And what this is really leading up to is we could do this for text too.
And so the idea, I mean, like, imagine who's reading this in 1983 that it has to explain hex and decimal and binary and compare it to Roman.
That's amazing.
Because, like, nobody knew.
What is the name of that thing?
It's just called IBM Light on the Future.
I love this.
With, like, a rocket test tube-ish spotlight-looking thing.
And it's incredible.
There's, like, oscilloscope waves in the back.
It is the most incredible thing.
It has this dictionary in the back.
Imagine the first time someone explains the computer and the dictionary is, you know.
Arithmetic unit.
Binary digit bit.
You know, like cathode ray tube.
Electrostatic storage tube.
And, you know, but the thing is, the reason I open this is because there's one cool page that really matters.
What is the organization of digital computers?
And so this is the thing that gets to this point about abstraction for us and AI.
These have been, for 75 years, how we thought computers were organized.
Input, storage, arithmetic, control, and output.
And that's all, that's what we learn in school.
You took courses basically in each one of those.
Last night we were going back and forth on the abstractions that will remain in computer science.
And you tossed in networking, which is sort of control.
Everybody forgets networking, by the way.
Of course I will.
Well, because most people stop worrying about networking.
Stop worrying about it as soon as the packet leaves the computer.
Well, and probably, I would say, you know, the late 90s was the end of basically a mandatory networking class.
Because, like, it was solved.
Like, there was no, you know, for me, it was the transistor.
I was, like, the last time that computer science majors had to know what a transistor was.
And trust me, I actually don't get what one is now.
It's like a triangle symbol.
But...
But the interesting thing is those abstractions led to, okay, so now we have those abstractions.
They were basically fields that did each one of them.
Like you spent 20 years of your career on storage and you watched them march from tubes to drums to spinning discs to tapes and so on.
And, you know, if you did output, you watched the invention of going from a teletype to a line-oriented teletype to a terminal, black and white, to color, to...
vector and the whole deal.
And all of those were the fields and they all rose in parallel.
NECS department, which came out of the MAP department because of the missiles, ended up being like departments made up of those things.
And then it all collapsed and produced us to the systems group.
Yeah, yeah, yeah.
So let me just push on one angle of this.
Sure, sure.
Because I, listen, I clearly love the framing and we move up in abstraction and every abstraction there's still a set of problems which is a higher level of abstraction.
But I still think this kind of notion of economic need is very important.
Oh, yeah, yeah, right.
So, for example, we, you know, Bletsley Park was about cracking a code for a war.
And so, like, there's this effort that created innovation that the outcome was, you know, winning World War II.
ENIAC, we were trying to do nuclear...
not just research, but like innovation, you know, in terms of a war effort.
And so we needed to like calculate integrals and we were doing it by hand.
And so at that point, these things were lauded as like saving humanity.
Everybody was super excited.
All the physicists loved computers and used computers.
For me, the thing about the current solving math is I don't know what that thing on the other side is.
And I do think we've had that in the past.
Well, we did.
We had the AlphaGo moment.
It was the same thing.
Like, we did a podcast, not in this room, but...
But even before AlphaGo, we had, like, remember when Chats were all bitten in the chest?
We had the IBM chess thing.
And Frank, Chad, and I, we did this podcast about AlphaGo, and we had to try to make...
people understand like why it was a good idea.
And I think it's actually pretty reasonable for us to ask the question, which is there's things that these things solve.
And, you know, there's a lot of utility and value in that.
And like that's going to move things forward.
And when that tends to happen, people tend to be excited and get behind it.
And it's these things you solve, which I think people don't have like as positive of you.
And I would submit that's because it's almost like solving the problem had become the...
the end as opposed to the actual end.
But we should maybe all step back and be like, if you're really, like, sad about something being solved, maybe it wasn't worth working on to begin with.
Right.
And so, you know, and you're like doing the Sand Mandala and like you think in her piece or something like that's not moving the economy forward.
Right.
Well, we're both like we're systems people, but I'm actually an apps person.
I know.
Yeah, we're both systems people.
But like I, I, of course.
absolutely think that the wave that matters are apps.
And of course, the internet also, this same problem happened in 1995 and 96 with the internet, which was, it was very exciting, but most people just sat around saying, I don't know what that does for me.
Look, there's a great book out now called Steve Jobs in Exile, which I...
absolutely think is required reading if you're listening to this podcast.
So, Cain wrote the book, but it's with Catmull at Pixar and with Dana Lewin, who was at Steve's, super good friend, and was also at Microsoft.
They all, this book is just fantastic because it explains all of, it encapsulates all of this notion of like building things that people actually need and solve problems.
But it pointed out Very clearly, you know, the Next was actually the machine that Tim Berners-Lee used to write the HTTP protocol.
Right, so he actually...
A Next machine?
And he used the Next machine.
That's interesting.
And it's super interesting because nobody knew what this machine was for or what it did.
But then he built that, and still nobody knew what the machine was for or what it did.
Because he's like, well, it's to find the phone numbers and the other researchers and to share papers.
And I'm like, eh.
And I think there was a great example of a company, a Seattle-based company.
that was called Cyber Pizza.
And this was like a .com thing that didn't even make it to the 2000, I think.
But the idea was, it was basically on, or DoorDash for pizza, only pizza.
And they would basically, you would order, and then they would figure out a pizza place near you and send the pizza.
That was the launch demo for the Next on stage.
They did that.
And they had actually pizzas in the back in case it didn't work.
And I should say for Next Step or Open Step.
But...
But the idea was that that was showing what you could do with it.
And literally, the reaction was like, wow, that's really cool, but have you heard of the telephone?
Yeah, yeah, yeah.
Your point is not everything we've known how to use.
So I'm a little focused on, like, there was a solution on the other side that people were going for.
You're making a point that there's a lot of platforms that get built where that's not clear, but clearly they did.
Well, the strategy was like, I will show, here's probably one of my last visual aids for today.
But, like, the word processor came out.
And this is in 1982.
And people are using it on Apple II computers and this new kind of computer called CPM, which is the origin of DOS.
And people were like, I don't understand why you can just type.
And the people, once you used a computer, the idea of typing really, really just didn't work anymore.
And so some people at law school.
Well, you have to show it now.
Yeah, I will.
I'm just like building up.
But these people at Harvard Law School, they brought in the first laptop.
So that's the first laptop.
Weren't those called lugables?
Well, no, this was just called, this was literally just called an Osborne and it was the only one.
So, so as a guess, how, Eric, you're, you're a kid.
How, how much, how long was the battery life in this?
Um, not long.
There was no battery.
This giant case that weighed 25 pounds, there's no battery in it.
It just plugged in.
But that was a trick question because every time I've ever lugged mine out, I have mine from college.
Like, people are like, well, how long does the battery last?
And so it's literally the size of a sewing machine.
It's bigger than a legal carry-on ever was.
And that was my college computer.
But in my senior year of high school, it got banned from Harvard Law School.
So someone showed up to do their exams.
So at Harvard, they used to bring your typewriter to exams because that way the professor could read it.
And two kids brought computers in.
One brought an Apple II and one brought the Osborne.
And the school banned them.
They just said this is, and for all, every reason you could read, and I have the Time Magazine articles and the New York Times, every article you could read reads like, don't use the graphing calculator, don't listen to rap music, or don't read, don't know jazz, or the arguments that are going on now.
Three years ago, I tried to get Cornell to use AI in freshman writing.
they just stopped talking to me.
Wow.
But here's the irony of that.
My freshman year, when I had this computer, I was, of course, the only person in my 90-person dorm with a computer.
And I had to get permission from the dean to use it to write my papers for freshman English.
This is the fall of 1983.
And so that's exactly where we are now on all of this stuff.
But the thing is, you can also think of it as a level of abstraction.
Because, like, no one's going to college now without a computer.
Like— Can I just put— Yeah, no, put— So I agree with you, but let me— Right, right.
Every once in a while, I'm like, well, maybe it's a little different.
So here would be the argument.
I don't think of the history of computer science that I can recall.
Have we ever abdicated actual reasoning or logic?
It's always been a resource, right?
It's been like compute never can storage.
And like, that's what you're providing.
And then the human is like putting in the high level thing.
And then it's using the compute network and storage to like calculate the answer.
But like all of the kind of initial setup we're providing.
And I guess maybe it's not true for the internet.
But now I feel like you're actually abdicating thinking in a way where you're like.
tell me the answer.
Like, I'm not even really sure what the question is.
Right, right.
And again, like, I think maybe you could say, well, Google was kind of like that too, but it was still very much a social thing and not very much.
And so it does feel like that's a little different than just going up in abstractions because going up in abstractions, you still tend to have like a deterministic system that's a higher level of abstraction that like you have a computer and like it's a human being that's kind of defining everything about the problem statement.
It feels a little different.
Well, it definitely feels different.
Here's, I also think for me, graphing calculators felt different.
To me, graphing calculators felt like cheating.
And because, you know, the test question was make a graph.
And so that's what's going on right now is that the capabilities match the test question.
Now, getting us full circle to what we were talking about, about computers and mathematicians, my freshman year, also a new product, a new thing came out, and it was Maxima, which was the MIT symbolic math package.
And so this was a way you could literally type in like an integral.
I remember the first time I saw Mathematica, I'm like, this stuff is black.
So Maxima is, you know, machine-aided, what was it?
Machine-aided computation symbolic math, I think was the, and that was the lab at MIT started in the late 60s, early 70s.
And that had started to sweep through.
So my freshman engineering class, we had a version of it that ran on IBM PC.
It was called MuMath.
And, like, you could, like, we got our calculus homework.
You marched over to the engineering library, checked out a PC disk, and then just typed in the answers to.
So you don't think that.
And that was cheating.
Let me just push on just a little bit.
Because I tend to agree.
Every once in a while, I have, like, moments of doubt.
So I don't remember writing programs where you actually abdicate logic.
Like, if I'm writing a program, I'll, like, whatever.
I'll use a cloud database.
I'll use storage, I'll use networking, you know, whatever it is.
But like correctness and logic for the program is under the programmer's control.
Maybe I'll use a third party library.
But again, like I'm choosing the library.
I know the inputs, I know the outputs.
And I feel like we're entering this realm where you're actually abdicating logic to a third party.
You're like, tell me the answer.
So maybe that's just a higher number of protection.
It feels a little different to me.
No, look, that's the debate.
I'm like, I'm all in on the debate.
Like, here's an example of that, a Stanford example.
So in the, during the AI winter, that was the 80s, Stanford, the biggest.
One of the AI winters.
Stanford was, and we have a podcast on that from 15 years ago.
One of the AI, one of the biggest things at Stanford was to combine new AI with the medical school.
And so there were all of these projects.
to do like medical diagnosis, chemotherapy kind of stuff.
I worked on one that was doing organic synthesis with a team at Harvard.
And all of those were sort of the earliest, like, let me turn over the decision-making.
In fact, that's the whole era of...
the 80s in computers were the dawn of what they used to call expert systems.
I remember then.
And so expert systems were the first time we got a taste of this debate.
I remember very well.
Your classes were all this.
It just didn't work.
Your classes were mostly about, like you had a bunch of classes on this stuff.
A hundred months on expert systems.
I've had to build expert systems.
Right, right.
I've written a lot of prologue.
Exactly.
I very much understand it.
I just thought like that never really worked.
Right.
So the big difference is that stuff was working.
But now it's work.
And we're abdicating logic.
Right.
So it's interesting because to compare and contrast.
And even in the case of Prologue, you're kind of like for these declares.
You're coding it.
It's algorithmic.
You're still providing the end state.
And it's just like finding a way to get to the end state.
Where here, like you're almost asking what the end state should be.
So it just feels a little different.
And it's especially, so I agree.
Like I love having this debate because I think.
So much of it boils down to the concern and the willies that you get thinking about it.
It's actually because of the context we're in.
And like, because, you know, think about like we have all this stuff going on where people don't want to build data centers.
But like two years ago, people were like beating each other to, please, governors were racing to have data centers built.
Or, you know, 10 years ago, like build a car factory in our state, the one that billows smoke and is really hard labor.
And so.
The context really matters to these discussions.
You can't separate them from...
Right, but I said, I just want to go back to this later.
Yeah.
I don't mean to...
I just think, so, your and my entire career has been moving up layers of stack.
But, like, there's always a computer layer of stack.
Yeah, yeah.
You can always map it down to, like, the next layer in basically a deterministic way.
Yeah, yeah.
There's higher levels of, like, compute abstractions.
this is the first time it feels like a different layer of the stack.
Like maybe this is like really is the next abstraction, which is more of a human level abstraction, which doesn't map directly.
And so it's actually different.
So it may like, I think, you know, whatever it is, starting with like, you know, transistor logic and then going to compute and then going to like hardware and then going to OSs and then going to like applications and then going to platforms.
Like you've been moving up the stack that way.
It could be the case that like.
We're at a layer where, like, we have to rethink fundamentals.
Oh, yeah.
Because it feels a lot different to me than just, like, this is the next layer.
The big difference is, and you can argue this or debate it or label it, either side, which is we actually are making the leap from calculating to imperative programming, which is where we've been and where everybody is, to now.
And then we were in this, for a brief time, we were in this mode where basically the data really determined the program.
And that was the first recognition, all of the inference and everything.
And now we're at this where it's arbitrary, it's random, and it's statistical.
Right, so the way that I think about it is the following.
So...
In imperative programming, you know all of the steps, so you write the recipe, and it follows the steps.
Okay.
Then there's declarative programming.
Declarative programming is you know the end state.
Which is this prologue kind of thing for people to follow along.
Or data log.
Right, right.
Or SQL.
You know the end state, but then the computer does all the stuff to get to that end state, and you can't really bound the compute time.
So you're like, this is like a makefiles.
Yeah, yeah.
Like, here's what the end state looks like, and it does it.
And this is like this new thing where it's almost like, You don't really know what the end state is specifically.
And you just kind of like, you know, you kind of like, you know, pray to the model God in like the right words.
And then it produces the answer that just ends up being useful.
Yeah, yeah.
In a way.
But I look at it.
That's a factual statement like that.
But it's also interesting to think about it going forward in terms of is that itself the next layer of abstraction?
in how we think of computing.
Yeah, and it may be like compute in a baby, like this is where I compute and like natural like phenomenon actually intersects pretty heavily because the answer is produced from like human output, which is language, which is kind of different than like...
If we do need to rethink some fundamental assumptions, what may that look like?
Well, I just think that like people like, you know, Steve and myself have built these...
deep intuitions on how systems function and how they hit the industry based on 40, 50 years of like watching this stuff.
And I just don't know, like things like will value go to the model or to the app?
How much capital can you apply to this stuff?
What classes of problems can you solve versus not solve?
What guarantees can you provide?
How does this impact productivity?
There's a lot of things that we've got intuitions on.
And for me, the big question is, do we have to reshape those assumptions or not?
And to what extent do we have to?
Because a lot of physics feel a little bit different.
I'll just give you one example.
I mean, I've said this many times.
I think it's so important.
20 years ago, if you're a startup of 10 people and I gave you a billion dollars, what would you do with it?
You would end up spending a...
ton of money on building, buying your own computer or something, if that's where you're going.
Well, you hire people, you buy computers, you blow up.
Like, you wouldn't know what to do with a billion dollars.
Oh, I see what you're saying.
Yeah, yeah, yeah.
Yeah, 10 years ago, I give you a billion.
You hire engineers and you'd be fucked.
Right, right, right.
Like, what do you do?
Right, right.
Write code, you've got to, you know, you've got, you know.
The billion, the important part of that is it's a billion.
It's not that you got money, it's that it's a huge, it's a ton of money.
If I give you a billion dollars.
Two years ago.
Because $10 million, you'd buy a bunch of stuff from Hewlett-Packard, and the money would be gone.
Yeah, for sure.
Right, right.
This one is a billion dollar.
Yeah, I mean, like, in software, you hire people, and then it's a very job.
There's nothing you can do.
But the FK scale, the mythical man month is very real.
Yep.
And right now, if I give...
20 people a billion dollars, they can actually use it usefully.
It's very, so it's like, it's like we've kind of moved the industry from like this engineering bound problem to a capital problem that's fundamentally very different.
We've never been like that before.
And so like, this is like a law of physics where like our early intuition, which is like all problems, your engineering problem starts to change.
So I think there's this very open question we should, especially people like us asking, which is like.
To what extent do we have to reevaluate our priors on this stuff?
And it's not just one level of abstraction.
It actually changes the nature of capital versus innovation versus competition versus defensibility, et cetera.
That's a great way to think about it because it forces you to think about a new model.
It's also interesting that computing was capital bound for the first 30 or 40 years.
Like if you wanted to do something.
This is such an important point.
If you wanted to do something with a computer.
Like your first step was we have to get one and then you couldn't.
You were capital bound and then you were engineering bound and now we're capital bound.
Capital bound, yeah.
Which is crazy.
So it's almost like you have to like hop back 40 years.
Yeah, Mad Men goes through the scenario where the computer shows up at the advertising agency and they run around trying to figure out, explain what it does for people, which they also got a copy machine and the same thing they did.
But it was interesting because they couldn't figure out what to do, but they were excited that they had the capital to acquire one and it made them look like they knew what they were doing.
Five years ago, Patrick Olson interviewed Sam Altman in a podcast and Patrick was saying, hey, you know, we've been in this era of lean startup, but for your projects, you know, OpenAI, this sort of energy, you know, project you was involved with a few other aging thing, you've raised colossal amounts of money right out the gate.
Is that underrated?
And it's sort of just speaking to what you're saying.
Yeah, yeah.
You know, it's interesting.
So prior to AI, there was always this battle between Eric Ries and Ben Horowitz, right?
Yeah, yeah.
So you lean startup, and then, you know, Mark and Ben wrote, like, the art of the fat startup.
Which basically, you raise the money and go for it.
But there's always been this natural limiter, actually, which is the engineering.
Yeah, yeah.
Complexity is, oh, that's actually been the reality.
And so Patrick Collison is right.
It's like, we now...
have a discipline for taking a lot of money with small teams and using it productively.
That's a very, very big change.
I don't think we've internalized this.
Which also, it's incredibly, that is why there can be so much optimism now.
Because although capital is scarce and it's hard to get and all of these other things, once you get it, you, the...
As we know, the building based on people was also hard, like just scaling that and doing more.
And then nine people can't do anything faster.
I'm telling you, yeah, my 10-year job, you know.
Was recruiting.
And giving, yeah, it was like literally giving.
these early teams' money and then helping them accrue and then watching and waiting for two years.
And I think engineering just doesn't scale.
It also has implications for venture capital because for the last decade, people have been saying, hey, there's way too much capital.
I just think this is such a crazy view.
So there's been this view in venture, this zero-sum thinking, which is funny from the people that shouldn't be zero-sum thinking.
Too much capital is chasing too few deals.
You're a venture capitalist.
Don't you believe in positive-sum stuff, right?
But if you look at the numbers, The more capital that flows into private markets, the larger the market gets.
And there's a couple of reasons.
One of them is the one we've talked about, like technical waves that actually are able to consume capital like AI.
But there's another one is if there's more capital available on the private markets, companies will stay private longer.
So more value accrues on the private side.
And so I think capital going to private markets grows the TAM.
It's not a limited TAM.
And I think the people that should be, there's so funny, early stage venture investors who should think of like, you know, positive sum outcomes need to stop thinking about zero sum.
Well, one way to think about that is, I'll bring it back to what I think, your foundation enables what I think is the most exciting thing, which is we're really on the cusp of a wave of apps.
And like the fact that now you can apply capital without also being a recruiter for 10 years.
and have output.
Now, all of the world that's unserved by software, which is literally all of it, like everybody who complains about whether it's medical records or scheduling to go to a doctor or my favorite are lawyers.
Like nobody has cheered more that, oh my God, we're finally going to be able to automate lawyers with AI, which is the weirdest thing in a world where everybody is against everything except having more lawyers.
But like all of this This means that the person who has the domain experience, like we used to love like venture capital thing, like, oh, you know, it turns out it's like really, really hard to like build commercial real estate.
Wouldn't it be great if somebody who understands commercial real estate built a software company, but then they don't know how to build software.
Well, they should get a co-founder knows how to build software and teach them about commercial, 20 years of commercial real estate.
It's really hard.
But now the path.
from that kind of idea is a capital problem.
And that's a new level of abstraction.
And I mean, I remember my very, very first customer visit as a professional product developer was to visit a doctor who happened to have gone to medical school after majoring in the earliest computer science.
And he wrote...
like a DOS program to schedule a doctor's office.
That's amazing.
Which you'd think it's just scheduling.
It's a calendar with hours.
But it turns out this was me, a 20-year-old me, hearing this guy explain, no, you don't understand.
You call the doctor and, you know, you're talking to a scheduler.
So they're listening for keywords to decide, is this five minutes, 20 minutes?
Do they need the x-ray machines?
Do they need the EKG?
And so they're actually scheduling like a blood draw and all of this stuff in parallel, not just the 10 minutes you need with the doctor.
And so that's what his software did.
It took him years to bang that out himself.
And that's the kind of thing that's going to be able to happen.
Like now that problem can get solved by the person who knows.
No code is finally here.
Well, it could really be.
And it might actually be that you're not just building this throwaway code that's hard to do, but also everybody else's abstraction layer is rising.
You know, you don't need to design that piece of code.
Like, you know, if you're doing it for a phone, well, the phone's abstraction level has risen.
So you're not building a text control.
You're not building UI controls.
Whereas 20 years ago, step one of building a company was building all of those things.
And so there's a lot to how important this is in terms of what you're able to do.
I want to talk about any other fundamental assumptions that might be interesting to revisit.
Is it sort of, how about incumbents versus startups?
You know, we've talked a lot about innovators dilemma.
Does that, you know, now that these startups are, or these incumbents, you know, have the capital advantage, are they able to do more?
By the same time, we're seeing startups that you would think incumbents would just destroy.
This is the crazy thing.
If you would have told me, six months ago, you would have asked this question, say, like, what advantages do incumbents have?
They have the same advantage incumbents always have.
They have the capital.
And they have the cash flow and they have like whatever.
Distribution.
Distribution.
And like what's crazy is AI, A, solves the distribution problem.
It just solves the demand problem.
And B, these companies are able to raise so much money that they're actually on competitive footing with like the Microsofts and the medicine and the Microsofts.
And so I think we're in a very new territory when it comes to these new challenges versus the income, specifically for these two reasons.
You know.
I think that the distribution point is often misunderstood how impactful it is.
In the past, if you had a company and you wanted to get people to use your stuff, it was hard.
You'd hire marketing.
You have no idea how much to invest and where.
You didn't know what you were getting on return on investment.
But the demand is so unlimited for tokens and for GPUs.
Literally, you can just decide how much money you're putting into it in order to drive top of funnel and growth.
And so the things have typically been very, very hard for startups.
you know, are much easier now.
And I think this is why we're seeing such meteoric growth of the cursors, the anthropics, and the open AIs.
That results in capital access, and that has put them on uneven footing.
So very, yeah, I, and I think it's to your point about how hard, look, my whole life was managing thousands of people, of engineers to build things that couldn't be built anywhere else.
Like it was, the moat to build an operating system, it was infinite.
And I think— You have to have one Cutler?
Is that the other?
Well, but it really— That was mostly it.
No, I get it.
He's brilliant.
But read the Steve Jobs in Exile book because you really get a sense for building up.
In fact, of course, Next was famously just—it took the code from Mock at Carnegie Mellon and started from there.
We couldn't have done it from scratch completely.
But this whole idea— of just how important it is to think through the domain-specific and how you disrupt people.
Because there was an old joke at Harvard Business School when Clay was still with us, which was, they really, it's weird that they teach disruption as a theory in the business school when really it should just be a fact in the physics department.
And I love that.
I was there in 98 when he was writing the book and the paper and everything.
That's when I was teaching.
That's actually great.
And I really, I used to be, of course, there's a lore with everybody who's from a big company in Silicon Valley or when you arrive like I did, the theory is always like, you always think, oh my God, we're just going to crush all of these little companies.
You always think that when you're at the big company.
And then you realize that they never get crushed.
And that, Ben always makes this point.
Like they just, and Mark does in his, his movie did.
It was AWS actually put out of business.
Right, right, exactly.
And because, and the.
And because, you know, the startups don't aim straight at the incumbents.
And the incumbents just don't pay attention.
The incumbents are only interested in what the other incumbents are doing.
Microsoft is worried way more about what Amazon and Google are doing than anyone in a startup space.
But the elements of disruption that matter are the cultural ones of being a big company.
And those are constant.
Those are the laws of physics.
And so you can't, you just can't change those.
You can't change scorecards.
You can't change field sales and go-to-market and compensation and org structures and legacy and customers.
Because, you know, the way you behave, if you have, you know, 500,000 customers you're serving, there's a bunch of stuff you just can't do.
Like, you're just stuck.
And that is really the essence of disruption.
And that's why we're at a magic moment where It's not just that the culture is there like it always is, but the startup ecosystem, it's a reflection of what happened during cloud, which was a whole bunch of stuff that you needed.
Again, it's this abstraction layer.
You don't, you know, if you're a startup like you were, you don't have to go build a data center and build your own egress and call AT&T and do all of that stuff.
Now you're up and running in the first hours of...
your first dinner.
But the thing with cloud, I actually think it's, you actually articulated it very well.
I'm going to actually use this because this is great.
Like the thing with cloud is like, nobody thought they could put AWS out of business.
Right, right.
Like you just kind of accepted the oligopoly and you built on top of it.
And the question is, is like, will they kill us in our little kind of pipsqueak corner?
Will they just add you for free or for some price you couldn't add or whatever?
I actually think, you know, and that's always been the question.
Will Microsoft, right?
But now these companies are actually taking on the incumbents.
And you make this great point, which I actually had thought about it this way, but like, coffee's been defined by these very complex, large engineering efforts building a chip.
Building a System.
What's that?
The Age of the New Machine?
Yeah, yeah, yeah.
The Soul of the New Machine.
The Soul of the Machine, yeah.
What a beautiful book, right?
It talks about how hard it is to build, like, you know, these systems, building an operating system.
Even in the cloud, like, I mean, like, Jeff Dean in that era of people were building these distributed clusters, and they're the first time that people could figure out how to do that.
And once you had that, this was a massive advantage.
So these were these massive engineering efforts that no startup could do.
And now, for these models, it really is just capital access.
It's some very different laws of physics where, like, if you can amass the capital, you can do something.
Like, I mean, you know, I mean, Google is Google.
They have all the data.
They have all the intelligence.
And, like, their models are getting trounced by open AI and by anthropic.
And it just goes— Because the cultural element.
I think that people on the outside underestimate it until you've lived the cultural element of trying to— To do something.
I'll bet it's cultural.
It's not like a typical, you know, engineering problem.
Like, they're very good at it because they've out-executed.
Like, GCP is fantastic.
That's a massive engineering effort.
And then also, I bet it's probably hard to free up that much capital for one of these companies, honestly.
Well, all the big companies, you can tell from their earnings calls how consternated they've been over the capital.
You know, you have Google doing their bond deal to move it off balance sheet, basically, in some weird way.
You know, you had the rumors of, I don't remember which company, you know, the rumors of like, well, they're rationing the tokens so that they go to the enterprise customers and not to the internal products.
And so the internal products are AI starved.
And of course, none of their competitors to those products are starved.
And I just, I've learned to really appreciate the call.
I mean, look, I fought and fought and fought to not be disrupted.
by the mobile platforms, like by ARM, basically.
And Intel just didn't care.
You know, I came down here, I sat across the table from all the Intel leadership, and I pulled out the first Surface, and I said, here's our new computer.
And they got very excited, and then they were like, but what's in here?
I said, well, it's an ARM chip.
Oh, wow.
And, you know, like the fact that I even brought one into the building, you know, and it was very, very tough.
And...
And they just never felt that that was going to, that was like a chip used in a printer.
And also, and they looked at me and like, we're ARM licensees.
We knew all, and I'm like, but it's the power, it's the graphics, it's, you know, always connected, all of this stuff.
And the culture was, they do Moore's Law at Intel.
And just like with Google, they do hyperscale.
So like if AI moves on device.
Yeah, yeah, yeah, of course.
Like, that's not what they do.
Yeah, sure.
And, you know, with Microsoft, they were squeezed.
They're squeezed now, you know.
And I think you raise super interesting points about the opportunity, though, with this capital inversion.
Yeah, go raise capital and go after the people.
Well, and it's not just, it's like you're also saying, like, we're actually not going to question you if you're trying to raise that capital.
Like, we're not going to look at you like you're crazy.
And you just look at the raises that are happening, right?
You know, these companies have been quite successful as a result.
The last thing, when Vishal came in and we had him on the podcast, he was sort of, he thought all of them were a great achievement, but he was bearish on their ability to invent new discoveries, particularly like scientific breakthroughs or things like that.
And I'm curious if you think the sort of math progress is consistent with that or what is your latest thinking on sort of the limitations of the current sort of model architecture versus like, well, we need more?
So here's kind of my new view, which is I think we know exactly how these things work.
You put a bunch of data in them, they're stuck to that data.
They can only do in distribution stuff and they can move along that manifold in a perfectly Bayesian way.
So we can say these words.
And then the question is, okay, but what are the implications of that?
Like what problems can it solve, right?
I think it's just so hard for a human being to reason about a digital artifact, in this case the model, that was built with $5 billion.
So like in the history of humanity, we've never created a single digital artifact that had that many flops and that much data in it.
So on one hand, we know exactly how it works from a mechanic standpoint.
On the other hand, That is so much data and it is so much compute.
Maybe all of that stuff's already in there and it can solve anything that you want.
And so, you know, the conversation has moved from the how do these things work?
We know.
Can it do out of distribution stuff?
No.
Does, you know, is there transfer learning?
Probably not.
Like if IRL, one thing, it doesn't teach something else.
Like is the singularity here?
Probably not.
I think everybody kind of, most, many people kind of agree on, like we're not in fast takeoff.
You know, we're stuck to it being in distribution.
We haven't closed it.
We all agree about that.
But what I don't think anybody knows is, okay, but you're still putting 10 billions of dollars in that thing.
What's it capable of now?
And if you consider this meta-economic machinery, which means, the ability from Anthropic to raise lots of money and then pour all of that money into this thing to create this super powerful thing.
I don't think any of us can predict what that means and where that goes.
And so it's a different conversation, but the question is the same.
It's like, will that be able to cure cancer?
Maybe.
But if you put $20 billion into something, maybe it can cure cancer effectively.
And that's where I think the discourse has evolved and where it is now.
And I honestly have decided that I cannot predict what an artifact that you use $20 billion to create is capable of.
Look, I think it's just so important.
It's important for people who are deep in watching everything that's new to admit that they can't predict.
And I think that that's great because it turns out, like, I wrote 58 memos on what the internet was going to be.
And I was wrong a lot of them, by far.
But I do think on the, and I think...
But even this one is a little different.
This is like, I take $20 billion and I put it into a model.
Right.
And then you and I look at that model and we can do whatever we want.
I don't think we can comprehend what that even means.
There's so many flops and so much data.
Like, I don't know what that's capable of.
And I think we are.
I think that that's really true.
And I think, but I will say on biomedicine in particular, like the other half of my household is a research doctor who uses AI.
We have a spark at home and she's loaded.
A spark?
Like a sun spark?
No, no.
NVIDIA Spark.
Oh.
Yeah.
No, no.
No, not with a C or the K.
Oh, yeah.
Wow.
We were in old times there.
I was like, not a Scott McNeely Spark.
No.
No.
Not a Scott McNeely.
No, and like, it's all AI.
Like, she does brain stuff and surgical brain stuff.
All AI.
And it's so interesting to see.
Because what it really can do is it just, it.
it sees the patterns that only experience could tell somebody.
But if there's 10,000 papers on a topic that's part of her model, then, like, it's just finding the patterns that no one has.
And that's a pretty basic AI capability at this point.
But it's actually opening up solutions or problems or research directions and things like that.
I will say, just for the, like, this is not a magic to discover drugs.
Because the hard part of drugs has always been candidates.
Not candidates.
It's always been efficacy and safety.
The candidates have, since the 80s, have been able to develop more than we could test.
It's human patients, and it's very, very, very hard.
Can I please tell you something that I got wrong on this?
So I love the question that you asked, which is how has our thinking evolved on, like, you know, whether these things, you know, like their capabilities in generality, which is I was responding to this boast.
Bostrom notion of recursive self-improvement, fast takeoff.
You create one of these things, you step back, and it takes over the world, right?
And so I kind of poo-pooed that because that's clearly not what's happening.
And I think a lot of people agree that that's the case, right?
But here's what I got wrong.
What I got wrong is I did not know that we could effectively just continue to pour money in this.
Like, the scaling laws are holding, and I don't...
I don't know what it means to just, let's say we do a $100 billion training run, to have this thing that you're putting $100 billion in.
And then that money comes from this meta-economic machinery that may be able to want to solve whatever.
They may want to solve cancer, but they may also want to create a weapon.
Who knows?
And so this concentration of this many resources in a useful way, I think, is very new.
I don't think we understand the implications.
I think you could reasonably argue that that's very dangerous if you kind of apply that $100 billion in the wrong way.
So I think that's kind of where this conversation needs to evolve to.
So less the foom, you know, and more the what does it mean to be able to concentrate resources.
Which also was, I mean, you're basically talking about exponential growth.
And this is just exponential in dollars.
Yeah.
We all know none of us can model exponential very well.
Yeah, but we've never been able to do that.
Like, complex engineering project, you were not, like, tackling one problem with a lot of money.
You're just kind of building this machine.
Well, we, you know, you're right.
You're 100% right.
And I completely agree.
But just, I remember just sitting and meeting after meeting, Intel saying we have 5 gigahertz, we have this many gigahertz, this many transistors, and literally nobody knows what we're going to do with them all.
No, you're building the machinery.
I'm saying in this case, if you're like, I want to exhaustively explore every protein combination.
Right.
We can just turn that into a money problem.
Yeah.
It's kind of very strange.
Which is a great way to say it, that we can take previously infinite problems and apply capital and it becomes finite.
It just makes it a capital problem and not an engineering problem.
Yeah, yeah.
Which is just a very different loss of physics.
Yeah, yeah.
Let's wrap on that because it's a great episode.
Thank you guys for taking care of us.
Thank you.
Thank you.
Thanks again for listening and I'll see you in the next episode.
As a reminder, the content here is for informational purposes only.
should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund.
Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast.
For more details, including a link to our investments, please see a16z.com forward slash disclosures.