Your Always-Available Second Opinion

Hello and welcome Mechanism. My name is Justin Robert Young, joined as always by my friend, your friend, Andrew Main. How you doing? I'm doing fantastic. Good to see you again, sir. Oh, always an enchantment. We got a bunch of stories to get to, including a viral uh story about a man who might have cured his dog's cancer by way of Alpha Fold and Chat GPT.

We've got uh a a new push forward with health for JetGPT Health, uh with the podcast that uh you did for open AI that just came out today. And then of course an NVIDIA. sailing to at uh yet larger and larger projections. Uh big, big, big advanced guidance that has a lot of people excited. Again, more uh

More contradictory evidence. Everybody's got evidence that they want to say to the health of the AI industry or the existence of a bubble. We can get into all that in a second. But let's start with the good news here.

The uh a man's dog is now uh on the road to recovery after he took. matters into his own hands, sequenced his dog's uh uh DNA or genome. I can't I don't know the exact uh uh specifics here and then used both Alpha Fold and Chat GPT to help diagnose a specific mRNA vaccine that he then had to clear through health boards.

And now it appears, at least according to all available data, that the dog is now recovering from cancer. Did did we just cure cancer? Did is this dog, is this going to be the the beginning of the end of disease? Well, I I am gonna be the butt of the joke of a tweet that I put out and do the actually. So I I I put out a tweet that's I got like fifty thousand views and it was I said Uh, we've gone from J J P T is just the next token predictor to like

Actually, uh ChatGPT didn't really custom design that RM and R M RNA vaccine. You know, and and and and the the idea is that I was trying to call out the fact of like I don't know what worked. I don't know how effective this was. I don't know if it was the other treatments the dog got. I don't know if it was completely wrong. Did nothing. Whatever.

All I know is, man, it's the conversation shifted now. That we're debating whether or not and and and it was to be fair, it's Chat GPT Alpha Fold. I think he did part of it in Krok. Um The now the conversation is, well, did it really you know like, okay, we're we were a year ago, we're like, can it spell strawberry? Now we're like, oh Did it cure cancer? Did it. Yeah. Um, I think and and it's funny'cause the the reaction to this story was extremely telling.

There are people who jumped in with technical notes and stuff like this, like, Hey, it's maybe not quite this, but

Totally fair, totally get it. I understand that. Um there was some other reactions though, like, well, yeah, sure, if anybody can spend a hundred thousand dollars on I'm like, Oh, so that's the price to cure cancer per individual. Like, like it's not, but you're you're you get this other sort of weird I would say how weirdly some of the people the medical establishment reacted was really what bothered me the most.

was like what's not ethical? Like, well, you know, if you got a terminal disease, you know, it's just Trivially easy to cure uh to to to sequence DNA and animals. And I regularly do it with mice and blah, blah, blah. The real challenge is in randomizing it for humans and having it clear ethics boards when it's like, wait a minute. That's holding us back. Yeah, you're telling me that we can one shot

specific like cancer. We we we can we can one shot diseases if we just focus on only a cure for us and we're not thinking about it being a mass distribution thing. And just to be sure, there are people listening to this who are only gonna hear you and I thinking, Ah, it gears like no we don't we don't

We don't we understand it. No idea. We have no idea. We're just saying, man, this seems like a cool tool and it's interesting conversation. And some people are kind of trying to do like, you know, I it's a weird, it's a very, very weird space to be. Um I let me let me just let me just make a note journalistically because

There's a few red flags here. As much as it's a heartwarming story, you know it's gonna go viral because man saved dog because of computer, right? Like that's a great story. Everybody loves a story like that. That being said, there were two vectors that this thing went wide on. Twitter, and then a what appeared to be a morning television interview with the man and his dog on Australian television. Uh

There's room for discovery here is all I'm saying. I hope that we have a lot more that goes into this. I hope that we look into it. I hope that that that that we we continue to track the progress of things like this. But to me, the remarkable thing about the story was. That this guy You know, in a bygone era, your dog gets a cancer diagnosis that's really kind of a rap. You're gonna follow whatever your vet's guidance is, and then that's that's gonna be what it is. You know, in the past.

15 years you've had people that wanted to clone their dogs and and you you have you have things like that that have kind of moved things forward in terms of the technological side of it. But this is something new. And and now it just shows you like. He wants to try it. He's willing to put six figures into the idea. He's worth he's w worth uh uh you know the the time and effort to try and and find a a cure. And maybe he did.

But maybe he didn't. I think it speaks a lot to the faith in the technology that he wanted to try and he wanted to invest in it. Yeah, I think it's an exciting time to see this happen. Uh and I'll tell you that there is There is sort of like a spectrum of clinician sort of reactions to things and we sort of saw that. And you know, we have just the the the Paul Ehrlich, the guy who wrote the population bomb and advised some of the most disastrous policy

in history. By accounts people know him so he's a nice guy, but a guy that literally was predicting, you know, billions and famines and whatnot. Wanted the you know, wanted the government to stop promoting large families on TV and said that we should step in and do all these sort of terrible, terrible, terrible things.

Um, there's a world where I talk to people in biology and life sciences who'll be like, Well, you know, he had a point. And like, well, no, he didn't. He was wrong on everything. Yeah. And also you get this thing in science, you'll hear like, Hey, maybe we should make like MRIs cheaper and whatever, like, it's gonna give a lot of false diagnosis. Like, I know, but your chances of discovering something is better than not doing anything at all.

And like, yeah, but and it's like like no, like we we can we can argue what people do with that information is one thing. But to say that people should not have that information because, you know, they're not gonna be smart enough to make those decisions about it.

You know what, people make people people are across the spectrum too. We make good decisions and make bad decisions. And I and I think that you sort of see this sort of this patronizing kind of expert class response to this, which is like, you know. Well, you know, and and you know, maybe maybe, you know, it needs to be done this other way. And I'm like, okay, cool.

when you're an actual person dealing with a loved one who has has a condition and you're told, No, trust this system and we know this system's broke and you tell us this system is broke doesn't inspire confidence or make me think that you're really in it for me. Yeah. Yeah, there's a larger conversation. I've been thinking a lot about the idea of Risk.

And and whether or not we are kind of under indexed on risk as a society and and we have moved more to a world of an expert driven consensus, which in general is good when you are coalescing around ideas that are known to work and then it can be stagnant and corrosive in situations where people are afraid to say a forbidden truth or to uh uh move something forward for other reasons, for personal reasons as opposed

to scientific or societal reasons. And so it's like for this, I don't know. Every once in a while there are just kind of paradigm shifting ideas that happen in the the the realm that they happen. For a reason. And and I would say with L LMs and AI in general, uh you know, our modern era of artificial intelligence.

A lot of that has been these odd, you know, ha how many what is it? A move thirty seven with the the the go, the the deep mind go game, right? Like That's that's known as this big moment of moving artificial intelligence forward because that was the place that you could do it, that you could show something that was like, okay, this is this matters in a different way.

Who knows whether or not this is one of those of guy cures dog for the price of whatever it costs to to to do it um plus waiting. But if it is, it's I I mean, remarkable. I think that's the other reason why it went as viral as fast as it did is that the as you say often, if this is true, what else is true on Guy Cures Dog of Cancer is like Pretty profound. That's pretty interesting.

Yeah, I so we just did an episode of the OpenAI podcast where we talked with Nate Gross, who is the the head of OpenAI's Initiative for Health.

And we talked about where this is all headed and what the state of things are right now. And we're in a cr pretty incredible place that's just It was three years ago that I was sitting there with GPD four talking to people about how we roll this out. We realized it was really good at diagnosing diseases. How do you do this in a way that you get it to benefit people as quickly as possible without too much pushback that then all of a sudden shuts it down? And

Now we're in this approach where doctors are getting better at using this. Doctors are now learning to not just use the free fast model, you know, to actually use, you know, maybe paid subscriptions and they're coming out with ways for doctors to do this to make sure it's HIPAA compliant.

And I you know, two things is we're gonna see more of this accelerate. I think we're gonna start seeing that orphan diseases, things like this. I had a friend that her her son was born with this condition, uh, Peyton, who was An orphan disease was maybe one or one specialist in the world trying to do this and solve for it as kind of a side project. And it's because it's very data intense. There's a lot of things when you want to try something, you don't know what's going to work.

I think we're gonna see a lot of opportunities come up now. There are gonna be a lot of different stories, mostly good, some bad, whatever about things people try and stuff. But I'm very, very optimistic. And the only thing I'm optimistic too about is If you wanna be a healthcare worker, I think actually now's a great time. I think that we're gonna see Yeah, I think that we're gonna start to realize that when we take away a lot of the drudgery of the office work.

when we figure out how to lower the cost of certain things and when you get to spend more time being a clinician or spending more time being the diagnostician working with these tools and seeing that impact. I think a lot of Even conventional medical practices often feel more like triage and they're just trying to deal with what they can. They don't get to spend the time to figure out how to get the best outcome to the patient.

I I think we could live in a world with more doctors or we come up with a specialization, you know, like you know, nurse practitioner, an another category like that where it's somebody who is Very good at understanding what the AI is able to do, able to do prescriptions within constraints of that. And I could see a world where we we spend uh more time talking to people in medicine, not because we're sicker, but because we're healthier and we have even bigger goals we can try to achieve.

solutions are like here's where I plug in AI, or if the real work is let me lay out my process, let me identify where human my human ingenuity is the most valuable. Let me find the places where AI is the most valuable. Let me lay it out in a chain. and figure out the combination of it. And I I I think of a place like medicine.

Like health, which is an extraordinarily ossified profession that people have kind of done things the same way just because that's how they've been done for a very, very long time. And you've seen a little bit of modernization in the world of the smartphone. But other than that, it's all kind of about probably 15 years behind where the, the, the, the bleeding edge technology is, at least from a consumer perspective, that now that can catch up.

really, really, really, really fast. And I think that that forward thinkers are now, I mean, I partly it's already happening. Like I recently got a new doctor. I have ChatGPT Health. I uh loaded all my records into there and

Uh, got labs done with my new doctor. And now I already have all my labs. I already have what I know I'm flagging for myself in my labs, what I want to talk to my doctor about in my labs, what my results were immediately that a a year and a half ago that I could just check in one chat thing. So now my conversation with my doctor is going to be

entirely different. It's going to be a good conversation, a second opinion based on what I already know. I have a level of knowledge coming in. And I don't find that to be redundant. I find it to be important that that I'm going to be able to talk to somebody

that that gives me a a a a way forward. And I don't think that that's gonna be unique. I think we're just gonna see more and more of cases like that. We're gonna see more and more of a reimagining of how we interact with our own data and with our own health decisions.

Yeah. I my wife and I switched healthcare providers and uh I haven't seen a primary physician in a long time'cause just trying to get somebody booked and get into their plan and getting into their list is so hard. It's frustrating. Yeah. Frustrating. Um, but uh you know, people like why are people talk to

Yeah, and that's that's the thing. It's like, okay, so in in the meantime, like why why are you adrift with your own health if you if if somebody is if you're not able to go see somebody, right? Like why can't you make good informed health decisions right now? That's that's the that's the promise and and It's always baffling to me when people talk about like, Oh yeah, well, I mean, is it really impressive technology? And it's like Yeah.

It it gave me it read my health charts. It it read read my test results and told me what was above and below what it needs to be. You know, gave me an estimation of uh uh my 10 year risk of all these various things. It's like I I don't even know if in that moment, if I were sitting with my doctor for a physician for an annual physical, if I would ever get to a level where I would ask those questions in in that in that one postage stamp, you know, period of time that I get to see that.

Yeah, it's it's the hard part. And you if you get them after lunch or they're distracted and you're like, oh, this is my window to get every health question I have answered. And if I forget to ask about this question. I die. Yeah. See you in a year, right? Yeah. When when everything is gonna be way worse. Uh And I'll I'll go there. I think it's more important for men. It's more important for men than women. Every every woman I know has more of a connection to

their healthcare professional than every man I know. And there's a reason why married men tend to live longer than unmarried men. It's because women get on them to make sure that they go to the doctor. And part of it is just It's frustrating. It's annoying. And also it's like, uh uh, I know for me, I'll only speak for myself, like I get resentful and angry when a doctor comes in and I see them for 10 minutes a year and they're kind of like frustrated or checked out or uh annoyed that.

we're having XYZ or a conversation or like if I I I'm struggling to think of exactly what I wanna ask them and they're like, you know, in my mind like rolling their eyes. They're like in a telly counter waiting for you to finish your order in, you know, exactly. Um yeah. And the lack of judgment on a chatbot is huge. I mean, like it's it's I I I really can't speak highly enough for it. Yeah. Uh Let's talk about

Well, anything else in in the in the health uh podcast that you found? I just recommend people check it out. I think it it was very interesting to see the efforts by which OpenA is going at this. Uh ChatGPT in the metrics is uh way ahead. They've been consistently ahead of everybody else when it comes to health. But being that this is an area that's getting of more interest, you know, Anthropic has uh has their Claude Health, uh Microsoft has like health co pilot.

or something along those lines. I think Google's gonna be working the space. Expect to see everybody get more into this because one of the things that having consistent conversations, long threads. Then it was memory to have threads, you know, communicate from back and forth. And then you start getting into security, uploading documents. Now we're getting to the health space. So we're going to see everybody try to improve these things.

and make them move a lot faster. And I but I really recommend the opening iPodcast because it was just a great example of just seeing how thoughtful they have. Dr. Gross, you know, and his team have been working on this, thinking about this really methodically. And I think it's going to be in a very exciting time. Amazing anecdote and you should listen to the podcast for the fuller context, but uh they ran a pilot program in Africa with ChatGPT Health and then

The program was like, Oh, I think we want to do a further study. The only problem is we don't feel that it's safe to have a non-AI control. Like that's how far it had come in just the time that they had used the tool was that they didn't feel comfortable giving patients non-AI assisted care if even just to understand the exact worth of it. It was just like all of a sudden having like, you know, penicillin available.

All right. Let's talk about NVIDIA. Big forward guidance for them. Uh they continue to say that they're uh, you know, the the demands for compute uh are

A a hockey stick going upward and that they are a huge part of that. Uh, what are your thoughts? Yeah. Disclosure, uh you and I are both NVIDIA investors. So Hell you are. Thank you, Deep C take with Yeah, take he take with that as you will. Um

There is this, you know, so I run a venture fund with some of my former open AI alumni and you know, people ask your thesis and I'm like, well state my thesis. And I go like, oh, cost of intelligence get lower, we're gonna use AI more, whatever. And people are like, Well, yeah, everybody feels that. I'm like You know, there's a difference between feeling it and kind of betting everything on it.

And and you know, that that's been my kind of thing with people is it's like how how committed are you to it? You might think that this is a possibility, but the only thing is you think like, no, this is really a possibility. Yeah. And

You can look at people like, ah, are we, are we overbuilding on data centers? I'm like, we're not building enough. We're not building enough. And in one of the analogies I give, it's like, Imagine it's, you know, we travel back in time to nineteen ninety nine and we go to this little fishing village off the coast of China and we hear that they want to build factories there and it's called Shenzhen. Ah, should we build this many factories? Like what

What is the total demand for manufactured goods on the planet? And plus where will it go? And it turned out that was actually a very smart idea. I think that here's what we know. Two things are very important. The number of API calls that get sent

increased dramatically. That when Deep Seek the Ne did the whole thing that Deep Seek said, Oh, we found this radical new way and all this and Wall Street's like, ah, they've lowered the cost of it. It's like, okay, every eight months somebody comes up with a new thing that they they figure out a more efficient thing. But also what did we see post that deep seek announcement a year ago was the rise of super capable coding models, the cost of tokens actually went up.

people were willing to pay more, particularly to anthropic, to have these things. They said, okay, it's one thing to have a cheap thing that can tell me how to make a grilled cheese sandwich. It's another thing that can go through this entire code base and factory. And we just saw this this area of all of a sudden not all tokens were equal. More demand, increased demand, people willing to spend more on this.

And API calls continued to go through the roof. The number of times that we hit the server to say, can you give me an answer? keeps going over geeps and people go, Yeah, but what's the limit? And I'm like, you're you're still in grilled cheese sandwich world. You're still thinking that that's what these models are for. And I see people who should be otherwise intelligent, who have very limited exposure to how these tools are used.

And don't understand what agents are or sub agents or the fact that like OpenA just announced you can now do sub agents inside of Codex. Well what does that mean? Well that means My agent that's running code can actually spawn a bunch of other smaller agents that then go run code. And what is open, I say, hey, your daily usually is gonna go up. So we've seen this pattern of ATI calls going up considerably. The other pattern we have is demand. If you're trying to get GPUs

If you're trying to order, if you say, Ah, I don't really know the demand is great. Find me a thousand H one hundred. Find me a thousand Vere Rubens. Find me these. Can you get these for me? Can you give me some blackwells? No, you cannot. The waiting list for that is extremely long.

And and that's people are kind of like I'm like, there are things you could check to go look at, like the waiting list for GPUs should tell you a thing, the number of API calls, and that waiting list stretches f way in the future. There was recently a little kerfuffle because OpenAI and Oracle were trying to build out another data center on their campus in Abilene, Texas. And that they changed the deal. Oh, but I decided to do something differently.

And people are like, oh, is this about demand? Look like well yes, the fact that this data center wasn't going to be ready soon enough and OpenAI had got hold of a bunch of next generation GPUs that like we need these things running now. We cannot have them sitting on a shelf. We're gonna go deploy them elsewhere because the longer these things are sitting unused, it costs them. And it was easier to just say, let's just go read let's just let figure out what happens to that data center.

throw that one as a loss to go build it. And I I I see people who just don't understand that. They don't see the pattern because they're still in this world of like, well, we're probably gonna reach capacity soon, right? We're we're probably gonna reach this. And it's like You know, we're it's like we're seal we're selling stainless steel, you know, and you know, our steel and you know, eighteen eighty with Andrew Carnegie, you know, and we're like

Yeah, we'll probably figure out the peak demand for steel in like five years. And meanwhile, we're like, you know what, you can build with it? Bridges. Wait, bridge out of steel? That's insane. You're a crazy person. No, we did it. Build a bridge on steel. You know what else we can do?

Golden Gate Bridge. Well, if you're gonna put a bridge across this, it's impossible. You're a crazy person. Oh, we do it. You know what else we can do? We can make buildings higher than ten stories using steel. No, that's absurd. That's impossible. And then The uses for steel kept increasing, increasing, increasing. And you know what's really hard to get these days? Steel Steel. Yeah. So

Yeah, I guess that is really the fundamental question whenever when everybody talks about like a bubble, right? Is just like, are we going to get to a point where there is

the saturation. Like there there's the limit where either through popular uh interaction with the tools, they're like, well, I don't need this all that much, or this is not worth it. And While I can understand that perspective from a certain point of view, because there was a lot of crap that like said it had AI on it or uh dumb features that were slapped on top of already existing products that were like, oh, here's the AI version that felt somewhere between foreign and incoherent.

The reality of the usefulness of the tools is pretty indisputable at this point. And if we are to understand that, okay, downstream of that are better commercial tools. commercial I mean, sorry, uh c uh uh uh consumer tools, better tools that people are going to use and fall in love with than like

Man, I I don't know. I don't even know where to where where to begin with that. I mean, when you continue to see the growth of all of these products, you know, the products that are leading the market, like ChatGPT. I mean, hell, there was a thing that came out that Sora, which obviously had a phenomenal debut and then uh uh cooled off the Sora app. That still not only retained its audience, it's it's grown.

you know, uh in terms of its its uh daily active users, which I don't I don't think people view it like that. I think people view it as like, oh, it was this you know, uh a a flash in the dark and like maybe they'll come back with more computer, a better model, and maybe it'll be better going forward with like the Disney thing. But no, it's it's still a destination for millions and millions of people on a daily basis.

Yeah, that was surprising. The the numbers about how Sora has in maintained usage and grown, and it does show you how much of a bubble You can be in a bubble inside a bubble. And that's the thing I think people forget about is that you can you can be I'm in tech, I get it. And then there's bubbles in the net. And I see that with uh you know I'm I am I am over the moon for codecs. I'm using codecs all the time. I love it.

And I see other people and I I followed Cloud Code, played Cloud Code, I think it was great. I thought it was super cool. But then I got to codex and like all this. And now I watch people like, Hey, have you heard about the Cloud Code? I'm like, oh no, we're in the Codex world now. And you see that where, you know, C Dance Pro 2.0, okay.

There's that's like a crazy thing where C dance showscased this model where they're like they just they just YOLO'd the entirety of every single movie and TV show ever uh and trained on it. And then we're like, oh, there's some copyright concerns. Um

looks like to be a fun I have a I there are people like I wonder it's not like I know people who they the the weight's leaked. I know people who are losing this, you know, and you just sort of go, Oh, like there's just you know, an example I brought up, I was gonna bring up Sue's tweet about this. So like uh the identity of Banksy got released, right? The who Banksy is, right? Yeah. And and I'm like, oh.

Oh, I guess if you're not in the art world, you didn't know who he was. Like if you're not that world, you didn't know, and that was news to you. And not to be like, oh, I'm an insider. It was kinda like you know I remember like Millie Vanilli were like, oh the band that was like, you know, lip syncing. Yeah. Everybody in like in the music world, like, Yeah, we all knew that. You're like, Oh, well, you you could told us, you know, and with Banksy was one of these things, like you just go, Oh.

I guess that's just sort of a thing that you don't think about and like with Sora. We're all focused on this thing, but like meanwhile they're probably, you know Tons of college kids sitting around in dorm rooms using this thing all the time in the Midwest and elsewhere and all these other places. We're not even paying attention to because we're looking at the new thing, you know, or we're just so insider that we're like, you know, I don't know. It's fascinating.

The bubble, we are all in a bubble within a bubble can probably be said about all culture right now. And certainly I don't think that there is a that there is a an element of the seventy thousand podcasts that I do where that is not something that if you want to be smart about a field that you don't have to deliberately try to think, okay, where am I? Where is the narrative that I believe is the narrative?

what is the counter narrative and how are both of them how many people are listening to them? How much do I need to elevate myself out of that to have a better point of view on it? Here's here's the thing that we are biologically unprepared for, and that is when Windows, the first version of Windows came out and you used it and it kinda sucked. Yeah. Um you could not pay attention to it for a couple of years'cause it took a while to get better.

Then it got better, you know. Uh and when the first iPhone came out. You could look at it and go like, eh, uh, I don't need one right now. You were good for a year or two before all of a sudden that mattered. Then you're like, Oh, I need to get something. You know, we saw this with, you know, email. You didn't have to have an email address in the nineteen eighties, although you could have had one, but

By the mid nineties, you know, I remember there's a time when I had an email address on my business card and people are like whatever nerd. And then yeah now everybody uses email address. There's periods where it's okay, but the thing is accelerating. The point is It's getting harder to sort of say, I'm going to wait it out.

Because it won't make a difference. And we're seeing that now with AI capabilities where it literally it's a weekly bi-weekly sort of change. These things change, but also it's not just what changed, it's what's sort of affecting things everywhere. Um, my wife bought a shirt at a little, you know, st store that sells a little curios, whatever. And it was just that said something, I'm only talking to Chat GPT today.

Yeah. And and it was to me, it was this sort of moment where I'm like, this is like some quirksy Etsy style thing that you know, probably presumably a woman made about this, about this sort of parasocial relationship. with a system that I'm still can see being in the room with the meeting talking about the launch of this. And it's probably, you know, well, we we don't want the distraction for GPD four. Okay, go ahead. It won't be that big of a deal.

Yeah. It being a big deal and now approaching a billion users and it's a that it would happen this fast. This fast. It's crazy. That it's gonna hit a bill. I mean, like that's well insane if it hasn't already. Yeah. I mean I guess there's there's a a a question of timing there.

Yeah. Yeah. So about invi so so about we with it back to the NVIDIA thing. So we start off talking about NVIDIA. One of the things they talked about is they're building a version of their of their GPU to go into space. They've talked about space, you know, and we we've talked about before about All the challenges of trying to do training in space. Inference in space is one thing, but when you want to build data centers is like

You don't need a data center for inference. Inference is literally inference. You just need a cluster of computers. They could be any, they could be into the block plugged into your fiber optic line. They could be up on a satellite, whatever. Inference is one thing. When you start talking about data centers and hundreds of thousands of computers kind of connected there. Um

You can say data center in space, but what does that mean? But NVIDIA said, Hey, we're building you know we're gonna be building space hard and hardware for this, which I think is great. I I'm I love space. I am Yeah. I am willing to be sold. I try to say this, I'm willing to be sold on the data center in space argument. I just have these questions that I can't like I can't I can't get a good answer about this.

And then they're like, oh, you're a skeptic. Like, like, no, literally, it's like, you know, do you, would you like to breathe under the ocean? Like, like, yes, how? Yeah. Why do you doubt this? Like, wait, no, no, I'm into the idea. Yeah, I mean like your your line has always been it doesn't make a lot of sense to train a model in space. It would make uh it would make sense to serve inference in space. Yeah. Right? That broadly broadly speaking, that is your

Like like to to create the infrastructure up there. It's hard to make yes, the trade off number one, theoretically, abundant free power, although that's more complicated than you might think. But let's say you figure that out. Abundant free power, great trade-off. Now, the downside, or sorry, the other uh, you know, side of that coin is.

Much harder to upgrade, much harder to fix if something goes wrong. Uh uh, you know, there's there are there are trade-offs that you have for not having it on the ground. Yeah, I and and to be fair, and and I think it's I I did a talk a few months ago about the future of compute, whatever. My closing slide is literally a giant orbital ring data center around the earth that I'm like like I'm like, hey, this is this is where we could go. But

There are all these things between here and there I have to go like, you know, it's like I remember I remember talking like I love space, I remember talking to a doctor who had this plan in a PowerPoint deck to build old folks' homes on the moon. Yeah. And and I'm like, wait, what? He's like, well, yeah, lower gravity, whatever. I'm like

I'm like, do you know the cost to put anything on the moon? The fact that even when you're using like reusable spacecraft and stuff, that the energy c like, what's your plan for producing the hydrogen that you're gonna need? It was just like and I'm like, this is This was like borderline I drew a nuclear bomb plan using a crayon on a paper bag, you know, sort of thing. And and

grandparents to be on the moon because unless the grandkids are gonna be on the moon, they're not gonna wanna be up there. I I don't know. I just was like and and I I watched other spaces like, oh you should go talk to this guy. I'm like, why am I talking I'm this

Anyhow, um my point is is that some things you go like, like I'm not against this. There's just a couple things we have to solve for. You know, like we had our space elevator debates, the great space elevator debates of the weird things podcast circa the two thousand aughts, you know, and it was like Let's lower the cost to orbit, you know, because we could do that. And but anyhow.

Point is, I'm excited to see this. I'm excited it. I think we're gonna see inference in space. I think we'll see some sample I think we'll see some toy example to training in space. That'll be a different deal. But uh, you know, you start digging into like I saw somebody dig into the

During the pandemic, liquid oxygen was such short supply, SpaceX was limited by their launch being able to launch because they heavily timid on liquid oxygen. And when you start talking about what it would take to actually put all of your assume assume the cost is beneficial to put the thing in space. Just the logistics of getting enough fuel to be able to send things into space and also not to mention the launch cadence, these things are really, really complex. But anyhow, excited to see this.

Yes. We're excited. We're excited. We're excited about the future and excited that NVIDIA continues to uh to to fund it. I see it's so strange, man, having these conversations with like average people. Like it's it's It really is one of those things and and obviously we're we're in a very, very complicated world in terms of you know, the geopolitical realities, but it's like I I feel like there's just so many things in my life where I'm like, Man I don't know

I don't know whether I'm like I I feel like I'm Charlie Day pinning things together on the on on the bulletin board and stringing red string around it where it's like like this seems clear to me. But I don't know whether or not I'm nuts screaming about Pepe Sylvia or everybody else is just like uh uh not seeing the forest for the trees sometimes. I Uh bringing up Paul Ehrlich again. Okay.

Yeah. I think the obituary said that he had like he had had yet uh his were predictions, you know, were like un you know, had yet hadn't happened yet or s was like something that was like not they didn't want to say wrong'cause he made specifics about times and dates. He was flat out wrong.

And and it you know, and as James Raney had said, you know, our prediction happened. For those for those who d who don't know, he he he believed that the earth was going to get incredibly overpopulated and that was going to be the greatest cataclysm of our time was overpopulation. Well, not just overpopulate. We'd be unable to feed ourselves. Mass famine and starvation by the nineteen eighties, you know, UK would collapse, all these sorts of things. He predicted the situ

He advocated against feeding starving countries that were going through temporary food shortages because it's only to create more miles to feed. You know, like he did not want India to be fed. And India India last time I did there didn't have a problem feeding itself. um extremely negative, nihilistic sort of view about this and is just totally, totally not understanding human imagination or ability to, you know, solve for these things.

and advocated policies that were other than that, nice guy, from what I'm saying. Yeah. But his policy wise was terrible. And they didn't want to really kind of they didn't Nick your time sent when I call out like, yeah, I was wrong.'Cause like, ah, but maybe no, wrong, wrong. Point is. When it comes to where things are headed, I don't know, but I do look at

Who is good at predicting things in the past? Who has made the best predictions about things that were kind of counterintuitive and not obvious? And those are the people I tend to think are probably pretty good about predicting the future. We can find people that agree with us about something with, well, this person says this, like, okay.

Who predicted where we would who will last year knew we would be where we are right now? Oh two years ago, five years ago. I can name a handful of people that had a better idea where we are now.

And I can name a ton of people that are jumping in to tell us, No, no, no, you're wrong. I'm like, Well, where were you? Where were you in twenty twenty? Where were you twenty twenty two? Can I go look at your record and you find out if they were making raptor predictions? They were wrong or they're just super vague. So

That's my thing. Like what is I I think that if if you we we get trapped in metaphors because metaphors lead us too much, but metaphors can be helpful in the sense of If you said, you know, if we're talking about data centers, all this is really about the value of intelligence. And the same thing with oil. Oil is about the value of energy. And if you think that energy loses value over time, you're not going to be correct.

If you think that we're going to reach a point where we have as well, you can't do that. left as much energy as we need, you would never be tr corrected any point in history. If you think that we're going to plateau with our demand for intelligence, I think you're mistaken. If you think that we're going to in the near future have as much electronic intelligence available as we could possibly make use of

You're really, really not paying attention. I think that's a great way to cap off this episode because I agree entirely. Mr. Andrew Main, where can people find you?

I am at AndrewMain on X and andrewMain.com and also check out AndrewMain prompts if you want to go take a look at the last six years of prompting from Somebody who was uh fortunate enough to be in the room early on and be the open AI prompt whisperer trying to figure this all out. There we go. And also the brand new open AI podcast available right now at dot com slash openai.

that you find the OpenAI podcast on the audio podcast player of your choice. OpenAI.com slash podcast. Yeah. There we go. That's another great place to watch it. Uh Justin R. Young everywhere on social media. Until next time, we'll see you later. Diamond Club hopes you have enjoyed this broker. Dog and Pony Show Audio