Transcription
Atro, I'm an engineer on a major 15 billion USD data center construction for OpenAI and Oracle and I can boots on the ground. Confirm that these people are running out of money. For a good half year, I thought it was an infinite money glitch, but now your comments make much more sense as they started cutting out so much scope and trying to cut cost wherever they can. Few months after getting a basketball hoop for our construction trailer, they cheaped out after the basketball hoop. At least you got the hoop.
Uh, brother, not only am I very interested in what you're saying, I would love for you to DM me on Discord, ideally with some proof of your job and and I would love to talk to you and get more information. This would be very something I'm interested in talking more about.
However, I I think I can I can riff off what you're saying here because I don't know if you guys have noticed, maybe some of you are, you know, we've advertised anthropic on the pod. I'm a Claude user, but I'll just say it. If you use Claude, and I think this happens with OpenAI, too, they have been rate limiting, bro. In recent weeks, even they've been giving you less quality for the same dollar. The thing thinks less. And it's because the bill is coming due. It's because every person was getting the Movie Pass experience, the Uber experience where it's like they were getting more than they were paying. So, of course, they liked it. It, you know, it's like the Movie Pass where you get 15 tickets for the price of one. It's like, well, this is a pretty good service, but it doesn't make any money. It doesn't make any sense.
So Claude and Open AI and and not yet Gemini. That's why I think, God help me, I think Google's going to be the monopoly on search and the monopoly on AI later on. I literally think consumer AI is going to end up in [ __ ] Google's pocket cuz they have so much money to blow that they're going to be able to keep like an adfree and better quality longer than everyone else. I legitimately think that's like their biggest advantage is they could just burn more money. But I think Claw was the best one, but right now they're sort of in shitifying. They're all in shitifying a bit to save money. They're like trying to hide it. Basically, you'll like ask it a question and then it'll try to guess whether or not it can get away with not thinking at all. And 90% of the time it does. So, it'll route it to a weaker model and it'll just like [ __ ] [ __ ] out an answer with no thought and it just hopes you don't notice or care. But that's what humans do. So, how is it better? It's supposed to be BETTER THAN A HUMAN. THAT'S WHAT HUMANS DO, BRO.
So yeah, it's a lot of people are complaining though. And what's weird is you could even consider it. I mean, obviously the terms of service probably cover their ass, but it's a bit of like a breach of contract. It's a bit of like a bait and switch because they are operating off of a subscription model, not a like pay-per-use model. So technically, you bought maybe a year supply or a you know, you bought the yearly subscription when it was a certain quality. You were you were like implicitly promised a certain level of thing and then they cut it back after you've already paid. So people are like angry about that.
I mean, it's it's it's a very interesting discussion but especially I got to be honest with you. I mean like clearly especially among workplaces the demand is there. They can't fulfill the demand but the workplaces can pay the higher prices and they probably will if AI were to be charging at cost what it actually costs. I think regular people would not pay for it because it would be very expensive at at current levels. They just not it's not worth it at that price for most people. but businesses would. And so I think that's why Open AAI has been pivoting all the business business clients. My company has 25K employees and they pay for C-pilot and shove it down our throats daily. The thing about Copilot is it just [ __ ] sucks on all levels. Like at least with Claude, it's like the old one was relatively useful. I mean, it had some value, but Copilot sucks, bro. The only reason your company is using Copilot is cuz they're cheap. is because Microsoft is giving insane discounts to allow it and it's included with the larger Office deal that your company already has. It's like why companies use Teams over Slack or whatever. It's because they're cheap. It's because they already had a deal with Microsoft for Office and they're like, "Well, we'll throw in this." And every company executive is being tasked by like the CEO to be like, "You need to do more with AI. Get everybody do more with AI. Get the employees to get all all future forward on AI." And so if they can do that while saving money using co-pilot, they will. And that's what but copilot sucks. So it's not it's not effective.
Don't you think business need even more compute to process requests? Yeah, they do. Everyone needs more compute. That's why I mean I don't think they're going to slow down on building data centers, but the companies that are building them now can't afford them. So it'll be someone else probably. I mean, I don't there's clearly demand like like large de like every time the models get better, the demand spikes even higher, especially from companies. They're using infinite tokens, bro. So, it's interesting. I mean, I don't know. It's it's a weird dynamic, but it's hard to tell because everything is cost. Everything is priced under its cost right now. So, you can't get a true idea of like is there a viable business at the current market dynamic right now? Clearly, it's useful to some people. Clearly, it has some value, but people are judging it based on the price, which is not real. Clawed Code is an insane discount. You're using a renting a GPU at 10 cost. Yeah, 100%. I mean, again, it's like early Uber. It's like early Uber when you can get a ride anywhere in the city for five bucks. That's so far long ago. Maybe some guys weren't even around. It's like 2014. But in 2014, you could get an Uber and it would go anywhere for five bucks. It was incredible. Early days of Netflix had every single movie and TV show you could want for like six bucks, seven bucks. The early days of food delivery, there was a company called Sprig that I talk about. Incredible burritos and cookies and everything freshly baked delivered for like four B. It was insane. Like all this [ __ ] was VC money subsidized and and AI is the gi most gigantic VC subsidy ever. There's never been more money pumped into it and there's these companies all competing to keep the price low. Le Paul is pivoting towards AI data centers now, bro. So is all birds. I'll believe it when I see it, dude. Everyone everywhere is pivoting to AI data centers because it's an easy way to get money. If you just say the word AI data center, someone will throw you a check, bro. Whether or not they're actually going to build it and it' be useful and at cost. Like to to really build them, you need a lot of energy, which is actually where America is getting limited. We threw up we're throwing up bunch of AI data centers that we can't even turn on. America does not have enough energy to support them. We can't turn them on, bro, without [ __ ] breaking the grid. We off the grid. Grid grid. Well said. Well said.
Hey, I have a question. Do you think that given the current demand for AI and the power and energy it requires? Does developing new specialized chips like TPUs could bring down the overall cost that make companies switch from traditional GPUs or the company's not switched because they have crazy commitments to Nvidia? Yeah, of course. You know what's funny is um there was an interview recently with Dwaresh Patel and Jensen Hong, the CEO of Nvidia and Dwaresh asked a similar question and uh the contempt the [ __ ] disgust with which Jensen said TPUs tensor processing units I guess can do some things. Okay, like from his POV they're dog [ __ ] They're dog [ __ ] Listen, having worked at Nvidia, I'll just tell you the real advantage of the Nvidia GPU is not that someone can't come along and make something that is uh more performant on a price per watt thing. It's that it has CUDA. It has this uh essentially like programming language for AI on the GPUs. It's the software and that stuff is tied to Nvidia GPUs and that's what everybody learns on. And so that lock in which was designed like 10 years ago. I mean it's an insane for level of foresight from Jensen that I was able to benefit from financially. It's like an incredible moat. Big A.
Is it bad that my peers are extending and pretending graduating by either getting a masters delaying graduation to try and get jobs or internships cuz there are situations make me feel guilty for finding a job under graduation even though I did from competit. Is this just a long flex? Is this whole post just to flex on your friend that you have a job and they don't? Jesus Christ, bro. Don't feel guilty for getting a job. That's insane. Wait, you should just have you should have bombed the interview so that you're What are you talking about? You just got a job after. And I don't blame them for trying to do the the pretty standard thing in like a recessionary job environment, which is they go back to school or they find some other way to opt out of the bad labor market to stay busy and keep their skills above. It's it's a normal thing to do. I'm not saying it's got a guaranteed success rate, but I'm just saying it's it happens relatively frequently. Hey, truck. All my friends are broke, but I have millions of dollars. I feel bad for them that they're poor. It's basically the same question, bro. Basically the same question.
What should I do? Did you see AI creeping into physical 3D art service through machines? Yes, dude. In fact, again, I found this out through my wife, who is a she's an avid cosplay prop designer, both personally and professionally. She has a 3D printer that is always working in her studio back there, and she's making a millennowy staff. All right. But she was like talking with other props designers on Discord or whatever, and one of them said, "Yeah, AI is getting [ __ ] crazy. It can 3D model things based on a photograph." And Ari was like, "I don't believe you. I don't think that it can do that." Like, probably not. Well, right. And then they're like, "Test it." So Ari sent a photo of this. I can't find the exact photo, but essentially it is the flat [ __ ] Friday gator. She sent a photo of the flat [ __ ] Friday gator. Okay, just this photo, flat still image. And then her friend fed it into AI and it sent her back a perfect 3D model. generated of the flat [ __ ] Friday gator and then she fed it into her printer. Bro, look at this. Look at this, dude. It's crazy. He's a flat [ __ ] This is off a photo. I don't know why to be huge. No, it's I mean it's crazy. It really is crazy. So, she was like, "Damn, that's scary." That's what she said. She's damn scary. Cuz that's crazy. I feel like to cuz I I saw the 3D model on her computer in Rhino or whatever and it looked good. And it's like to make that would be hard and take time. Ah, it's just crazy, man. And it did it it did it in like a second. It's It really is. I mean, it's impressive, but it's scary. It's weird. I mean, it all this stuff is advancing so fast, man.