Coffee With Scott Adams — Knowledge Archive May 24, 2026
Scott Adams Philosophy Archive
Search ideas
Episodes Episode #3003 Segments
MainContent AI & Technology

Back to episode — Episode 3003 CWSA 10/29/25

Context —

pes are in his catalog, but check it out. Just look for Akira the Dawn, spelled A K I R A the Dawn. You'll find it on X. I'm sure it's on YouTube too, but look for it on X. We're expecting an interest rate cut today, maybe a quarter of a point. Stock market is already responding to that and the fact that Trump seems to be having success in his Asian trip. Maybe there'll be something with China co…

← Previous segment →

n. Now, does that sound like a bubble to you? I don't know what else that could be. If that's not a bubble, I've never seen a bubble in my life. I've seen a lot of bubbles. There's no way in the world that's worth five trillion dollars. Because it's not like they have no competition or that they'll never have competition or that we'll never find out that maybe there was some other way to do this cheaper.

What would happen if somebody came up with a way to do this cheaper? Well, let's go to Elon Musk who says this. He came up with an idea on one of his earnings calls. One commentator is talking about this. Apparently, since every Tesla car is also a little computer and they're all networked, it wouldn't take a ton of work, says Elon Musk, to turn the collective cars that are on the road into an AI inference engine such that if you wanted to use AI and you were in your car, you could talk to your car and the car would use all of the computing in the entire network just like a data center would. So you wouldn't need a data center. You would just need the cars that are already on the road and suddenly you have AI. And then of course you hear all the people who are making their own local AI models. You know they use DeepSeek or something else and they're building home office AIs that don't even have any connection to the rest of the world.

So are none of these things a threat to Nvidia? I mean, I'm no expert in this domain, but you think they'd have some competitive threats, even if it's not those. Anyway, five trillion. Good luck with that.

Here's my experience. Yesterday I thought, you know, I'm going to look into this again. And I looked into it about two years ago and AI couldn't do it. But I thought by now it can do it. And what I'm talking about is not hallucinating. And I thought, okay, I have to create one of these special databases called a RAG or a vector database the AI can use without errors. Allegedly. I didn't believe it necessarily could, but I wanted to build one. And so I went to Grok and I said, "How do I do this?" and it recommended a few apps. One of them is called Pinecone. So I said to Grok, "If I use this Pinecone app, is this going to allow me to build a database that will be reliable and not hallucinate with AI?" And it said yes, that the Pinecone app would allow me to easily create one of these files because I was teasing Grok and saying, Grok, if you would know how to use one of these files, couldn't you tell me how to build one and couldn't you build it yourself and just say fill this file or fill this database and I'll be able to read this every time? Why do I have to build it? Like why am I even involved? We've got a five trillion dollar AI company, but a human is the only person who can figure out how to format the database. AI can't do that for five trillion dollars.

So then I said to myself, aha, I'm going to beat the system. So I'm going to have Grok walk me through what I need to do technically. So that basically Grok will do it, but I'll just be the one typing on the keyboard. So then I open Pinecone and it has its own set of instructions how to do it, but they didn't work.

What if I told you instructions on how to do anything technical in 2025, no matter where the instructions came from, whether they came from the company that does the product or AI or your smart technical friend or the people on X who gave you advice, which one of them accurately will tell you how to solve any technical problem? The answer is none of them. Every one of them will have a confident answer of what menu choice you should use that doesn't exist.

So that's the first thing. So the Pinecone instructions, I couldn't get them to work. So then I take Grok and I point it at the screen and I say, "Why isn't this working?" And Grok says, "Oh, those instructions are wrong." So instead of pip, just to give you one example, one of the commands you're supposed to do in this terminal window is pip. And then Grok says that doesn't work on a Mac. Like what? I'm looking at the company's own page of what command to use. PIP. And then Grok says, "No, it has to be pip3 if it's a Macintosh." Who's right? Well, pip3 didn't work either, right? And if I were to ask somebody to help me with it, they would say, "Do this command instead of those two commands." And it wouldn't work. In 2025, no one can tell you what to do that works. It just doesn't work.

So what I found so far is that anytime I want to do anything, now obviously I'd be in the smallest of small business category, but anytime I've thought I want to do something with AI, any kind of project, any kind of business initiative, do you know how every time it ends? It ends the same way every time. Somebody says you're going to have to hire somebody to do that for you. That's right. Every single use of AI that I've concocted, and there are a lot of them. You know, if you can imagine all the ways that the Dilbert creator and a podcaster can use AI, it's a lot. The things I imagine I could do with it would be amazing. Like I would have an AI cohort here that I would just talk to. I would make my comics with AI. I'd have a clone that would answer your questions about me and about my books. I mean, all kinds of AI amazing things I would do. And every single one requires me to hire more humans. And you know what would happen if I hired more humans to do that work? I wouldn't need AI. The AI is to replace the humans. But you can't do anything without a human. And I'm pretty sure that even with a human, you can't make a database that works. So that's my complaint about AI.

Anyway, Elon says that Tesla autonomous driving might spread faster than any technology ever. And I think he's right. And the argument for that is that they've been working for years to have the cars ready to just flip a switch. So when he flips the switch to autonomous driving and I believe that they've already satisfied every safety test that you could do, so it's already safer than human drivers. When they flip the switch, it'll be just this enormous footprint of autonomous cars that went from non-existing to existing with just one software flip. He's right. That will be the fastest spread of any technology ever. So that'll be fun.

Apparently UPS trying to adjust to this new world is using gig drivers for deliveries. Gig meaning that they're not the regular UPS drivers. But if UPS has one small package that has to go to one place in your neighborhood, it might not be worth sending the UPS truck there. But they might have so

Context —

mebody who had signed up to be an occasional delivery person and they get a message that says, "Hey, take this package over here." And apparently there's a lot of that happening. Esther Fung is writing about this in Wall Street Journal. So if I were a package delivery company, I'd be really worried about the Tesla autonomous cars and the Waymos and everything that works without a human. Well, mos…

Next segment → →