The Exchanges

Q What's your day to day? Like you have a lot of meetings. You have just a few.

A Yeah. I have a lot of, you're talking about my, my, my lived life. Um, a normal weekday for me is, um, I wake up about seven, 15, my wife and I get the kids out the door and she drives them to school at about eight. I usually do a half hour, a 45 minute walk with my dogs, get exercise, strenuous up and downhill. Heart rate goes up, which is good. Listen podcasts, for example, yours. So that's, you know, why, why I have time to actually follow Exciting, cool things that other people are doing. Get to work at nine. And so work nine to six or seven or something like that. And most of that is meetings. And so that's me trying to solve whatever the problems are of the day, get home, dinner with the kids. I insist on eating with my family, hang out with them until bedtime and then crush through for like two, three more hours until I pass out. And do that regularly.

AI assessment note: “work nine to six or seven or something like that. And most of that is meetings.”

Q Last but not least, the question that everybody wants the answer to. Have you finished building the Lego robotics table for your kids that you mentioned last time? And what's the, do you have a new project that you're working on?

A Massively forgot about this. So we still use Lego robotics table. So this is a big four bait sheet of plywood with a bunch of, uh, two by threes around the edge, but a four by eight sheet applied was pretty hard to work with. And so it breaks apart into three decomposable modular sections. And so it's super great. And the kids are getting way better at programming and Legos and all this kind of stuff. Um, gosh, what's my most recent project? I was just building swords in the shop with kids on a bandsaw. And so you take a piece of wood, you have a bandsaw, give it to a kid, and you say, don't cut your fingers off. Turns out that the bands, I mean, I'm, I'm obviously joking a little bit. They get a lot of, they get a lot of oversight, but a bandsaw is actually a very safe tool. And so the reason for that is that a bandsaw, which if you, probably a lot of people have never seen a bandsaw, you can do a Google search for, you have two wheels and then you have a blade that goes around the wheels. And you've got a table, and the cool thing about a bandsaw is that you can crank down the opening towards the blade, so it's just, just as big as a piece of wood, and also, it pulls the wood into the table, and so, because it pulls the wood into the table, the risk of, like, something called kickback, and there's a lot of other things like this, is very low, and so you can basically tell a k…

AI assessment note: “So we still use Lego robotics table... what's my most recent project? I was just building swords”

Q Actually optimization. How long did you work on CPU and then how much of a jump was it to go from CPU to GPU?

A So, so the way I would, the way I explain modular is if you take that first three year R and D journey, and if you round a little bit, first year was, um, prove compilation philosophy. And so this was writing very abstract compiler stuff. And then prove that we can make a matrix multiplication go faster than Intel MKL's matrix multiplication on Intel Silicon and make it configurable and multiple D types and prove like a very narrow problem. Um, and do that by writing this MLIR compiler representation directly by hand, which was really horrible, but we proved the technology and some technology milestone. Year two was then say, okay, cool. I believe the, the fundamental approach can work. But guess what? Usability is terrible. Writing internal compiler stuff by hand sucks. And also matmul is a long ways from an AI framework. And so year two embarked on two paths. One is mojo. So programming language, syntax, member of the Python family, make it much more accessible and easy to write kernels and performance and all that kind of stuff. And then second, build an AI framework for CPUs, as you say, where you could go beat OpenVINO and things like this on Intel CPUs. End of year two, we, we got, and we're like, foof, we've achieved this amazing thing, but you know what's cool? GPUs.

AI assessment note: “first year was... Year two... build an AI framework for CPUs”

Q Can you just maybe run people through the pieces of Max? Yeah. Because I think last time none of this existed.

A Yeah, exactly. Uh, last time has been a long time ago, both in AI and also in modular time. Um, yeah, so the bottom of our stack is, we, we think about it as concentric circles. So the inside is a programming programming language called Mojo. Chris, why did you have to build another programming language, right? Well, the answer is that none of the existing ones actually solved the problem. So what's the problem? Well, the problem is that all of compute is now accelerated. You have GPUs, you have TPUs, you have all these chips, you have CPUs also. And so we need a programming language that can scale across this. And so if you go shopping and you look at that, like the best thing-ish is C++. Like there are things like OpenCL and Sickle, and there's like a million things coming out of the HPC community that have tried to scale across different kinds of hardware. Let me be the first to tell you, and I can say this now, I feel, I feel comfortable saying this, that C++ sucks. I've earned that right having written so much. And so, and let me also claim AI people generally don't love C++. What do they love? They love Python, right? And so what we decided to do is say, okay, well, even, and even within the space of C++, there isn't An actually good unifying thing that can talk to tensor cores and things like this. And so we said, okay, I want, again, Chris being unreasonable, I want som…

AI assessment note: “the bottom of our stack is, we, we think about it as concentric circles”

Q I realized like it impresses me that you named the company modular and you've designed all these things in a very modular way. Yeah. I'm wondering if there are any sacrifices to that or is modular everything the right approach? Like there's no trade-offs.

A If you allow me to show my old age, um, like my, my obsession with modular design came from back when creating LLVM. Okay. So at the time there was GCC. GCC is, again, I love it as a C and C++ compiler and stuff, right? And I have tons of respect for it, but it's a monolith. It was designed from the old school Unix, like, like compilers are Unix pipe. Everything is global variables, C code. It was written in KNRC, if you even know what that is anymore. And so it was from a different epoch of software and LLVM came and said, you know what? Everybody teaches you in school is you have a front end and optimizer and a code generator. Let's make that clean, right? And so at the time people told me again, first of all, it's impossible to replace GCC because it's so established, et cetera, et cetera. But then also that you can't have clean design because look at what GCC did and it's successful. And therefore you can't have performance or good support for hardware or whatever it is without that. They might be right if in an infinitely perfect universe, but we don't live in an infinitely perfect universe. Instead, what we live in is a universe where you have teams. Like you have people that are really good at writing parsers, people that are really good at writing optimizers, people that are really good at writing a code generator for XA six or something like that.

AI assessment note: “you can't have performance or good support for hardware or whatever it is”

Q Was there a company mandate that they had to write Swift from now on?

A No, but still it's like, wait a second, this isn't the company I thought I joined, right? And stuff like this, right? And so, and so there's this huge amount of turmoil and drama and nonsense that came out of that. And so, okay, fast forward to Mojo, lessons learned. Hey, uh, one, don't have a hot start. And so we launched Mojo A long time ago, before it even made sense, we called it a zero dot one. And so that, how's that honesty in advertisement, right? Like this is zero dot one, please don't use it, but if you're interested, we'd love your feedback, right? And so soft start go. Second thing that's very different is that in Swift, we had one demo app. And so you have, you know, a very, I think high power team building a language that had done lots of credible stuff. That was building a language for iOS developers and the compiler is written in C++.

AI assessment note: “No, but still it's like, wait a second”