Using AI To Uncover The Hidden Patterns That Hold You Back, with Tech Blogger Robert Scoble

My psychiatrist uses ChatGPT in our therapy sessions.

No way.

Oh yeah, I'm patient number one of her. So she has hundreds of patients. So it's fantastic. It listens to us as we're talking and writes seven highly detailed scientific notes about our conversation in real time, right from different points of view because she has to take very detailed notes on each patient. She doesn't do the notes anymore. Her system does the notes so she can actually listen to you and provide a better level of care. She says it's going to be malpractice soon not to use a system like this because it catches mistakes that she makes. It sees patterns over time. So If I do 20 meetings with her, it watches over time how a patient is doing and sees patterns in between meetings, which it's really hard for a human being to remember what we talked about 20 meetings ago.

Robert, welcome to the augmented life.

Thank you for having me on.

It's my pleasure. I think that we often talk about, I don't know, what might be labeled techno optimism on the augmented life since we're often in the pursuit of making ourselves and our lives better. You have, I think, an interesting perspective on this, having not just covered technology for a long time, but, but been in a place where, you know, some of the world's greatest innovators were like pitching you, like, like bringing their breakthroughs to you, you know, maybe most notably.

I wrote seven books about this stuff, you know.

Yeah, right.

Yeah, it's a lot of work. No, I, I've had a fortunate life growing up in Silicon Valley. I was Apple's first child laborer back in 1978. It's true story. My mom got a job building Apple II motherboards and we built them in our kitchen. But I've launched many companies on my blog. I had a famous blog for about a decade in Silicon Valley. So I got to go and see a lot of startups. And that continues to today. You know, I'm, I'm putting up a couple videos this afternoon with the right entrepreneurs who are building this cool stuff. Nice. That's changing our world.

Right, right, right. So the, so let me start with a longitudinal question. I think that the. You've looked at a lot of innovation that has sorted itself over time between novelty and things that have actually changed human behavior. Right. And at first, perhaps you can't tell the difference.

You just said novelty. I, I have this brain computer interface from a company called Next Mine, which Snapchat bot. It's, it's a novel thing. Right. Sort of can tell what you're looking at because your processing is done in the back of your mind for your visual. So when you see something, you process it back here. And this thing has electrical sensors that pick up the, the electricity off your brain and try to tell what you're looking at. It's not yet a mainstream product, you know, it's, it's, it's big, you know, and it doesn't work that well, you know.

Okay, let's try to dig into this, this exact example. Like I, I, I will put forward the theory that the difference between novel and useful in, in this, you know, let's call it just transcranial brain computer interface is resolution. Right. So if it's low resolution, it's just kind of like neat. And when it starts getting high resolution, then it can become a behavior changing tool. But that's weird because it's just a matter of degree. Right?

A lot of sensor work, a lot of AI work to understand what's coming off your brain. I mean this is very early days product. Right. You know, and it sometimes it takes a decade or two before these things cook, you know, and get good enough that, that they start working for every everyday people, you know, instead of just the nerds.

Yeah, I'm, I'm like a meditation nerd. And so in, in, in, in the, in the community of meditation nerds in the United States, the, the Shinzen Young has been a big influence and, and I, I, I caught up with him at this event thrown by Brain Mind just a couple weeks ago and he was really enthusiastic about ultrasound, about transcranial ultrasound stimulation and the through line. I don't know if Shinzen would put it this way, but the through line to me was transcranial magnetic stimulation was like really promising, but, but ultimately a novelty. And I think that's because like the effects that it had involved basically spamming the whole brain with, with energy. And what's exciting about using ultrasound is that it can be much more targeted though. Here's an application that would have never occurred to me, but is exactly what Shunzen was, was wild about. If, if you get good at targeting, you can give people the experience of equanimity that you otherwise have to spend, you know, maybe a couple of years meditating on the cushion to achieve. You can just make someone feel it and then say, this is it, this is the destination now, now go get there on your own.

Right. So it helps you get there faster.

Yeah, in expectation. Yeah, yeah. One of the hardest things about teaching meditation is that it's um. It's really hard to get feedback. It's really hard to tell if someone is doing it. Right. Because it's entirely an internal experience. Right. Like, I can hire whatever. A chef to teach me how to cook and like, he will watch me do it and tell me all the things that I am screwing up. It's like, really hard to do that with meditation.

Yeah. Or, yeah. Or what's not on the recipe. Yeah.

Oh, yeah, yeah. Right. Well, yeah, that, that's just it, right. You. You can. You can decide to go off on your own from, from any level of culinary knowledge. You can be like, whatever, I'm just going to experiment. And of course, very similarly, you can do this playing around with your sense phenomena. But. But it's. But at the end of a cooking experiment, you can, like, serve someone a meal, right. You can see the work. Whereas it's like, really hard to see what happened.

It's really hard to see inside your head and what your experience is. And your brain is different than mine a little bit. Right. I have an aut holistic sun and he's. His brain is very different than mine. Right. So it's. It's hard to figure out how to communicate with him even about basic stuff, much less, you know, deep meditation techniques.

So I think that's the success condition of, like, the device that you just held up, which is. Can. Can you show me enough bits of resolution that I can tell the difference between your brain and your son's brain? That's probably good enough to be like, a genuinely useful thing.

Yeah, man. It's a new world coming at us. I'm playing with these brain computer interfaces. I'm talking to Dylan, who's a researcher at Unity Labs, and he's been doing work with them for years with Mark Zuckerberg, bought a company that makes an armband so you can think about shooting a gun and you can have eight arms in a video game, all shooting, all from one arm, because it just picks up the electrical signals going through your nerves. And you can just have eight arms, you know, all controlled with this band, you know, and it's still not a product. Right. They bought that eight years ago. And it's still coming because it's. It's one of those things that has to cook with the. The AI really has to come along to make it useful. Right. Beyond a parlor trick. Right. But yeah, you know, it's. There's a lot coming. I mean, you talked to Hugo her at MIT Media Lab, and he's making legs for People who lost their legs or prosthetics that jack into the nerves and their little robot legs. That's, that's crazy, you know.

Yeah. One of the things I like about his lab is using neuroplasticity. So one thing you can do is you can make your BCI better, but another is you can create some sort of feedback loops with the patient so the patient can iterate in their brains and figure out how to make the thing go. So that, that's really, that's really, really.

Cool, you know that that's the singularity joining up with AI, joining your mind in with AI and learning to work with it at a high rate of speed. I mean, if, if we had a time machine, we could go 10, 15 years in the future and get a neural link put on our brain or something like it. That's what we'd have to learn to do, wouldn't it be?

Yeah, yeah. So I've been wondering if, yeah, like the casual pursuit of mine, or may even not so casual, an obsessive pursuit of mine, is what I should be doing now. Oh, here's the way to put it. Like, I'm sure we all have people in our lives who have decided that they're old and they're just going to stick with whatever technology learned at some point. And they're done, right? They're just done.

Yeah, my dad was like that. He's a PhD, very smart dude. At the end of his life, he refused to learn new things. He's like, I'm done.

Like, I'm not going to disagree with anyone making that choice. I just want to say that I don't want to the. But, but, but, but maybe you can accidentally fall into like the gravity well of, of, of slower and slower adaptation. So. Yeah, like, I think one of the best things you and I can talk about is like, what heuristics do we adopt? Like, like what behavioral patterns do we adopt to, to not be like that. Like, I just want to continue adapting in, in my 40s, 50s, 60s, 70s, 80s, you know, until I upload into silicon what. And I, I, I strongly suspect you're doing the same thing.

Yeah. One way I did it for myself is to build 70 different lists on x.com where I could watch like a hundred thousand of the world's smartest people, like neuroscientists, roboticists, you know, tech journalists, tech futurists, tech AI leaders. Right. And, and they bring me new things and force me to consider them. And even if I don't feel like learning this morning, like I already just shared two things this morning that came out, you know, and it's like you gotta, you gotta see, you know, by being force fed. New thing, new thing, new thing. It's sort of, you know, if you're gonna keep watching the screen, you might as well try them. Right?

I know, right. Okay. All right, well, hold on.

Generalizing learning system is what I've built.

You know, that is smart and. All right, so would you accept this generalization? Many of us feel at the mercy of the algorithm, let's say. And what you're doing is designing your own algorithm. So you're using the basic primitive of. Of of lists in order to.

Yeah, I'm using X. I, I wish I could show it to you, but I use X Pro, which lets me put each list as a column. The, the lists are in that context, are not filtered. There, there is no algorithm. It's just live. What if somebody posts? I see it and I see it in real time and that, that way I. The algorithm is not picking things for me. Because you know how algorithms work. They work on. If you start clicking on cat photos, you're going to cat photos, right? Because it's trying to addict you. I'm trying to get beyond that where I just get to see everything and then decide what I engage on. But I think this goes back. I spent a lot of time in an old folks home and the reason I'm still. Yeah, a lot of. A lot of time. My, My mother in law had a brain tumor and she couldn't walk. And so she spent almost a decade in. In one of these homes because she couldn't get out of bed and needed to be taken care of. And so I stud people, I studied how dismal these places are for us, right? And, and someday we both might end up in one of the. In fact, we'll probably end up in one of those. Well, I want a Vision Pro, you know, because I started studying the hos. The room that she's in, she. There's another patient there. There's no privacy. There's machines beeping at you all night long and there's a small shitty TV on the wall and it's probably playing not your TV channel, right? And so that's why, that's why I bullish on this stuff. Because I want, I want the end of my life to be at least more fun than that, right. You know, and give me a, give me a Vision Pro so I can go to Yosemite or watch a movie or, or you know, do a 3D scene, a game with my grandkids or something like that. Right. Maybe I'm not learning that much at that point, but at least I'll have a better experience.

Well, as a hacker, I'm often asked for, I guess, foundational security advice. And I came up with this idea that the true foundation of personal security is actually close interpersonal relationships. Because everything can be taken from you. You can lose all of your wealth. You can lose all of your things. A skilled adversary, if they really want to take you on, can ruin your reputation. Your reputation can be taken away from you. The. And if you live long enough, you will have the good fortune of even losing your mind. Right. And there are, There are two ways this can go, which is that there you can be incapacitated and someone will, you know, use a little napkin to, like, dab at the drool coming out of your mouth because they are remembering the times you were there for them. Or, or there can be no such person. So that, that, that's. That's the last thing the, the. Your last bastion of security is the close personal relationships that you've earned by earning people's trust. When you had stuff.

Yeah, it's a big one.

Yeah, yeah. But, but, but, but then, you know, technology is just a fancy word for tools. And, and tools literally increase your capabilities. So, so I'd say, you know, the step up from.

By the time we're both in that home drilling on ourselves, we might have a humanoid robot doing the dabbing. The kids won. We. We might have a robot.

Yeah. Oh, man. Hopefully remembering what we like.

Yeah. Well, this is a big one. You know, I, I talked with. I did interviews with Skip Rizzo, who does brain research down at usc, and he studies dementia patients and PTSD patients, and people have various kinds of problems with their brains. He says with dementia, if you show the patient a picture of their childhood home, it reconnects the brain enough to get lucid and have them remember something. And once they start remembering, once they get their brain back remembering things, then they're lucid for a bit and they're happier because they're remembering something. Right? Because dementia patients get real angry at themselves because they can't remember the. Their kid's own name or something like that. Right? That makes you really angry at yourself. So using these techniques, that's why I, I'm like, put me in a vision pro and take me to my photos that I shot all my life, right? So that I can keep my brain connected and, and re. Re. Establish who I am to it. And AI can tell you who your kids names are. Right. Or, or it can remind you of who you are or, or help you out when you're having trouble. Right. That's, that's, that's why we're doing this work is because by teaching AI who we are, we will have a better old age.

I wholeheartedly agree.

And maybe live a little bit more on our own without having people have it, having to take care of us and stuff like that. Right.

Since I've been doing obsessive lifelogging, I've noticed this about my own memory, which is recall is not deterministic, but it is extremely rich. And so what I mean is I've been trying to dump a whole bunch of stuff into my Flickr data store so that I have visualization so I can go back and pull on any thread. If I, if I'm trying to surface a vague memory, maybe I'll get to something related like where I was, what I was doing something like in the ballpark. And then if I can pull on that thread because, because I can pull on, on my own data, then something triggers this actually incredibly rich recall. That's like way better than what's in my computer system.

Right.

Like I'll, you know, I'll remember sense, you know, relationships, I'll remember emotions. So it was all there. I just didn't know how to do the lookup, you know, so I'm just trying to give myself like new lookup tools and then it brings back the fullness of the experience.

Yeah. You're not recording this into a, into a transcript, are you? Because that's one thing I do.

Kidding. Yeah. This is all going to be super transcripted. Absolutely.

Yeah. You know, and that's, and that helps too, right? Because now you have a body of work that you can talk to and I use a couple tools like Limitless and Read AI to record my meetings. And they're pretty useful. And they're useful. I like read AI better because it gives you a lot more data. Like on this call. It keeps track of where our eyes are and how attentive we are to each other and it uses that data to build its summary because were. Yeah, it knows what we're really highly animated about. So, so that's an important point you just made. Right. Put that in there.

Nice, nice.

But it, the world is changing, isn't it? It's like, wow, it's changing fast.

Right. I've been thinking about, I'll probably run this as a, as A small experiment. I've been thinking about personal audio recording to just record like everything. Not just meetings, but like record an entire day. And it's not for memory retrieval. It's actually that, like, I want to do some sort of dumb. Remember back when AI sucked but people were still selling AI based sentiment analysis? Like, oh, we're not really good with language, but we can kind of categorize things by emotion. Well, I want to do that on myself. How positive was I today? How negative was I today? Did I get angry today?

You're gonna have a German mother in your ear. You're a little depressed today. You should go for a walk.

The answer is always going for a walk.

Actually get some exercise.

Yeah, this helps.

Yeah, I'm using. I mean there's certainly some, some of these things coming on. Like there's the humane pen and there's the Ray Bans, right, that have a microphone and a large language model running them. You can start using it that way. I've been picking the moments more like when my friends and I are really having an interesting conversation. I'll be like, can I turn on Chat GPT and I'll start recording it, right? Because it, it's really interesting when you're having a good, like, like this conversation. If I had Chat GPT going, it could get involved with us and augment our conversation, make our conversation better. And then you can use it to say, hey, can you take the last 10 minutes of our conversation, write a blog post or, you know, or, or give me a report or summarize it or, you know, or tell us what we're missing or tell us, go out and do more research on the topic we just talked about. Like, you know, we're talking about brain computer interface. Just go, go and run a deep seek research project and find us more information on that topic.

Okay. You know, there's something here, there's a thread I want to pull on, which is one of the things that I've enjoyed about your writing for a very, very long time, like over the years is the, is, is that you pick out, you pick out the important stuff. I'm going to call it salience. You've got the eye or the nose for the salient details.

And.

I'm just not getting the problem with distillation from any OpenAI model is not that it. I think that if you know how to prompt the models well, you can kind of minimize hallucination and accuracy problems, but I just cannot teach the damn thing like an eye for salience, you know, so, so, and, and distillation is a compression, right? So preferably what you want to do is compress while keeping. Let's call it the sexy bits, right? And I don't know, like OpenAI, especially like Claude feels a little bit more clever or has a little bit better emotional capture. OpenAI, like makes everything boring. In my experience.

There's something there. I mean, I just shared a item about Luma eyes. New model, new picture to video model. Right. And why did I pick that one out of the thousand that's come across my screen so far? Because I had dinner with the guy who started Luma. I've been following it since the beginning and I've been following all the AI artists since the, the beginning. And, and so I have a lot of experience in my head of what is interesting to that community because I've been watching it over and over and over again and I have personal interviews and, and context, you know, for my career, right. So I. It's really hard to teach anybody else. People ask me that all the time. Can you do a class on how you read X? I'm like, it wouldn't really, it would be entertaining, but it wouldn't be help you because, like, I know who Tim O'Reilly is because I sat next to him at the first Google Advertising conference, right? And he wrote the Web2 memo. And I, I've been to his parties and I've, you know, I've interviewed him a couple times. So when I see him come on the screen, I have some context to who he is and what he's, why he matters to the world. Right. I can't teach that to somebody else because, you know, how do I teach, you know, all that to somebody else? I can teach the facts, you know, but, but yeah, unless you have personal experience, it's, it's hard to do that. Salience. That's an interesting way to put it.

Right?

Yeah. You know, and everybody has a different salience, right? Everybody has different set of things they've done in their life, right. If you talk to the guy who invented the backward high jump, you know, Dick Fosberry, he has a lot about sports and politics because he's a senator too, right? But he has his name on the sport of high jumping. So he has a lot of understanding of high jumping and the technology there and the people, all that. I have no effing clue about any of that. So you're not going to call me up for interesting conversation about high jumping? You're going to call him up?

Yeah, yeah. The I had a mentor, like one of my mentors had the, a profoundly positive effect in my life by being hard to impress. He, he had just seen so much in, in, in my domain that you, you know, he was comparing anything that I did to, you know, approximately the best in the world. And so what, what do you think effect that had on me? Like, of course it meant impressing him became the most important thing in my life because he was the hardest to impress. Right. So like, like very, very similarly. Yeah. Fosberry is going like, it's going to really be hard to impress in his domains. Right. It's not just expertise, but a long memory of, of what everyone else in the field has been, has tried or failed or succeeded at. So let me ask, I'm going to put you on the spot. What does this mean for AI agents? I think what we're putting our finger on here is that tacit knowledge that actually leads to badass execution is just not in any training sets. We see a lot of people making the jump from the AI models of today to AI models being good, like task oriented, completers of things, agents. But where are they going to learn all this tacit expertise?

Well, they have a lot of expertise to pull on. You know, in fact, if you don't have any creativity, right. You can ask it. Hey, hey, I don't have any creativity and I want to use one of these new AI art programs to make a movie. I don't know how to make a movie. Can you write the prompts for me? And it does, and it does a pretty good job. Now will it be as good as Steven Spielberg can do, you know, but by sitting down and talking and talking through a movie idea? No, but it gets you started, you know, and that gets you 80% of the way to Spielberg.

Right?

You know?

Yeah. Okay, there's. Okay, you raise a good point. I've noticed that the people who are kicking ass as users of AI tools have adopted little meta strategies like this, where it's precisely a meta strategy. Like you ask the tool for help using the tool.

Yeah, my first favorite first prompt is please show me all points of view on neuroscience. Right. And it'll give you all points of view. Right. And, and that's really cool because one that gives you a menu of things to talk about. Right. To dig in further. Right at. By the way, with AI, it's very. AI is very different than using Google because you can go multi stage in the prompts. Right. So second prompt you can say, hey, can you give me a name for each of those points of view, like a Persona. And it does. And it does. Right, because you like if you're talking about electricity, Ben Franklin would be one of the names that might pull up. Right. And then third prompt, can you join all those Personas into a debate club or a South by Southwest panel discussion or whatever, a board of directors. And then you can talk to the board of directors and it's on point to the topic. Like, like if you really want to learn neuroscientists, this is sort of what I do. And now you have a group of people who are famous neuroscientists to talk to on in the engine, you know, about neuroscience and it gives you a start, you know. Yeah. And people who use AI all day long or who go to a lot of AI meetups or, or watch X, you know, the AI community on AX.

I'm.

I'm watching 26000 people in the AI community. Right. So I see a lot of these prompt trip tricks that people discover like oh, this, this prompt really did amazing for this topic. Right. And, and that gets you started. But you're right. I salience. I, you know, it, if, if you watch the AI artists, the people who are really good at it, were good at it at art, right. They, they know something about the people who, you know, like who is Picasso and what kind of technique did he use and what kind of ink or paint did he use. Right. What kind of brush did he use? If you know those kinds of things now you can talk to the engine and get better results out than I can because I don't know a lot of that, you know, that art history or, or the, the techniques. I know photography a little bit. Like I know what Kodachrome film is by the way. If you prompt mid journey with Kodachrome film, that'll give you back an image that looks a little reddish like a Kodachrome film did. Right. And so you know, if you have that menu in your head of, of.

Things to talk about the dimensions of specification.

Yeah, yeah. Like if you're going to do brain search. I don't know how to do brain surgery. So I don't even know how to start or talk to it about brain surgery. Even if I had a, a robot to do surgery, I'm not qualified. Well, not even qualified. I don't have the menu in my head to be able to talk to that machine and, and explain in a technical way what to do a surgeon does, Right? Yeah. Because they've spent nine years or 10 years in school learning it, you know, whatever it is.

A variation on the debate club or board of directors is probably my, my best AI model advice for my friends. My friends and family, like many I think you know, have turned to ChatGPT for, for medical advisor, you know, to, to like ask questions about, you know, something having to do with their body. And it, this works out so much better when you have the model make a differential diagnosis like medical drama scene. Like what you want is doctors debating what's wrong with you and working it out instead of just trying to get a single answer.

That's why the prompting technique I talked to you about works. Right. If you have a very specific kind of cancer, please show me all points of view on this specific kind of cancer. Make some Personas for those points of view. Those are going to be really high end, you know, people in the field and then join them together into a little club. And now you can talk to the little club about your cancer and it'll give you really pretty interesting answers. Right. Teach you a lot.

Yeah, yeah. I think this is the top way in which AI models in the last year have actually enhanced my thinking. Like, like I've been incorporating them as little reasoning assistants in this way to think through topics that are outside my area of expertise. That's just a general purpose leveling up of my life.

The multimodal models are really amazing. My friend had a back surgery a month ago and a week into he was having me change his dressings because he couldn't reach them on his back and he wanted me to take a picture so that he could send it to the doctor and find out if, if everything was healing properly. Right. Well, I sent the picture to Grok AI over on X.com and it went through the. It's like, oh, I see a little bruising. No, don't worry about that. Yeah, I did prompt it and told her a little story about what? You know, it's six days after the surgery and my friend had back surgery and here's a suture, here's a picture of his suture. Can you check it over for me? And it did. And it gave me the exact same information that the surgeon did. Right. Yeah. So it is. The medical side of this thing is pretty interesting. The psycho, my psychiatrist uses chat GPT in our therapy sessions.

No way.

Oh yeah, I'm patient number one of her. So she has hundreds of patients so it's fantastic. It, she, her thing listens to us as we're talking and writes seven highly detailed scientific notes about our Conversation in real time. Right. From different points of view because she has to take very detailed notes on each, each patient. She doesn't do the notes anymore. Her system does the notes so she can actually listen to you and provide a better level of care. But in 30 minutes it nailed everything about me. Because, you know, in your first 30 minute conversation with a therapist, you know, that's, you're sort of spewing like what's going on in yourself and why, why you're there to see her and.

Right.

You know, oh, I have this mental illness and it' with me, my life isn't going good. So I'm trying to work on this. Right. But yeah, it's, it's amazing and it, she says it's going to be malpractice soon not to use a system like this because it catches mistakes that she makes. It catches, it sees patterns over time. So if I do 20 meetings with her, she's had several patients that have 20 meetings with her. It watches over time how a patient is doing and sees patterns in between meetings, which it's really hard for a human being to remember what we talked about.

Of course. Exactly.

Right, right.

Yeah.

And, oh, I like that a lot. She, she mentioned, one patient, kept, mentioned his psoriasis. Psoriasis is a marker for depression. It's. For some reason it's connected and the AI picked up on that and she did, she missed it. She said that that was key to helping this patient out. Right. So.

Wow, that's cool. I, the, I hadn't thought about the longitudinal pattern detection that can be big in multiple domains of life.

Oh yeah. That's why I asked you if you're recording these, you know, because you could talk to it after a year of doing these videos, like, oh, what do you think about my body of work? All these people I've interviewed, you know, who else should I interview? They'll come up with things. Right. Because it, it has ability to see it as a whole. But which humans. I can't remember who I interviewed a week ago. I do these things. Right. It's hard for me. Right.

You especially couldn't tell how common themes shifted or something along those lines.

Some of the big picture ones I can remember, but, you know, little ones. But, but it's, it's, it's interesting when you have a body like this in AI. It's real interesting. What, what you can talk to it about, you know, and learn about, and pull back out, you know, it's, it's pretty, it saves me time too. I I use descript for my video editing, and I put that in there, and it makes. As soon as you import a video, it makes a transcript. Right. The ideas, you can copy that transcript and take it to chat GPT with no extra prompting. It writes you a blog post. It's like, what. How did it do it? Didn't even know that I needed a blog. But. Well, it sort of wrote a, you know, a summary of it and said, oh, this is a great interview. You know, and then you can start talking to the interview you did. And. And it's like, wow, that's really crazy. Yeah.

Are you taking steps to save this corpus in one particular place?

I'm not real hardcore. I'm not like you, where I'm really hardcore about it. I'm using Limitless, so I have that file saved on my. On my desktop, and I'm using Read AI now, and that saves it. But no, I'm not real religious about it. I should be, because I'm getting value out of it when I use it, you know, so.

Right, right. I should talk to the Limitless guys about doing the. The vibe check I was talking about. Right. Like, how positive, how negative am I? I think. I think they could make it work on their platform, which would be pretty cool.

Yeah, There's. There's a couple people. Avi. What's his last name?

Oh, yeah, the friend founder.

Yeah.

Yeah.

He. He wears his little device going to business meetings, and it talks to him in his ear about what he's talking about with, you know, if he's pitching somebody or giving a presentation, it'll. It'll talk to him like, oh, that was a great point. Or, oh, you should. You should make this point. Or, oh, you were a little rude there. You know, you should. You shouldn't be so rude to people. You know, you should be nicer. You know, and so it constantly is listening and getting involved. I. I do do that with my friends, but only at specific instances where I, you know, sort of pitch it up, like, oh, we're having a great conversation about nuclear power right now. Can I pull up Chat DBT and have it record us talking about this? Because I think it would have some interesting things to say to us. And it always does. It's. It's always interesting. You let it listen to 10 minutes of conversation or, or even two minutes, and it comes back with like, wow, that's really cool. You know, it has a lot of it. It knows every topic at a PhD level, you know.

Yeah, right, right.

Even music. Right. We don't even think about using it for music. You ask chat GPT who Buddy Guy is and it goes in a depth, you know, it knows a lot about Buddy Guy and that certainly knows more about Buddy Guy than I do. You know, he's a famous guitarist.

Oh, yeah. No, yeah. Of the category of music that I happen to really like too. I, as a matter of fact have used the advanced voice mode while driving. I've like stopped music that I was listening to to ask for backstory about the particular song. Anything related to like the artist coming up with it, performing it. Yeah, it's great. Yeah. So I, I, I, I probably, yeah.

That'S what makes us different because on in my car I'm having it teach me how to play poker. Actually, I was going to an AI poker game and I was chatting up with it and it's really good, by the way, at playing any game really.

Oh, oh, I gotta try that. Oh my gosh.

Yeah. Giving you some hints on how to play poker better with your, with your friends or going to Vegas or whatever.

Right. What I guess what I want is for something to look over my shoulder and watch me doing something and tell me like what I'm missing. You know what I mean? Like, like I'm sure my poker game could be improved, but I don't actually know what questions to ask. You know, just, just like show me the ways in which I'm sucking right now. And I don't, and I don't even know.

Well, it might, it might have you working on, you know, telling whether somebody's bluffing.

Right, right, right.

That was some of the questions I was having. Like, how do I know? How do I know if somebody's left? Oh, oh, it knows the game. Oh, it knows the game at a fairly advanced level and it gives you things to look at. Like is, are the people playing with their chips in a certain way? Do they change how they play with their chips? That might be a tell on, on how good the hand is that they have. Right. So look for that or you know, look at their eyes. Are they doing something different with their eyes? Right. It's really knowledgeable. I mean, you go to Las Vegas and take a picture of a slot machine, it tells you how that thing works. And then you, then you ask it, how does that addict the human playing the machine? And it tells you about the dopamine squirting in your head as you're getting these kind of non regular rewards, you know, that the machine is giving you. And that's why it's Addicting, right? That's how it's working on your brain. It knows neuroscience at a fairly deep level. Maybe not as deep a level as you are, but, you know, pretty good. Certainly, you know, answers most of your questions.

I wonder if, so it's, it's clearly the case that some people are embracing using these tools while, while many who have equal access to them are not. I wonder if somehow we're training our dopamine reward system differently.

It's, it's hard, see, because of what I, I'm a, I'm a beta tester. That's What I did 20, 30 years ago, was test out software and, and I, I was the number one beta tester on Adobe Acrobat 1.0. Right? And so I, I forced myself to use it in a different way than most people would use it. I, I, I call this AI First Living. You force yourself to use it even to go to the bathroom. You start asking, hey, how can I go to the bathroom better? And it has an answer, by the way, and it's pretty interesting. It's like, but by forcing yourself to use it for everything, you really got to start to see some, that's why I have some of these insights because I forced myself to use it over and over again. And when I don't use it and I'm not perfect at it, like, like I went to see, I went to the Sphere in Las Vegas a couple weeks ago and the tickets I bought were on the first level and I could only see a third of the screen. The, the top was occluded. I didn't think of that when I bought the ticket. So I, so I asked Rock, you know what, what tickets should I not get at the Sphere? And it told me and I'm like, yeah, this is what happens when I don't use AI first for everything. You know, like, ask it for advice on what tickets to get at the Sphere. It knows which, which ones are good. That's amazing. And that would have made my life a little bit better that night, you know, if I had used AI first, I even use it, then I'd start using it for thinking about things that might happen someday. Like my wife might, might be on the kitchen floor not breathing one day. I asked it that and it's amazing. It's going to save lives if you force yourself to use AI first. Then when you're in crisis, you're going to rely on it and you're going to make a better decision and maybe save your wife's life. Right, Right.

Listening to this makes me realize that my earlier testing of ChatGPT has been hacker like, but I'll call it depth first testing. Like, I was like, let me screw this model. Let me find out how it errors out, right? Like I was trying to test its capabilities in that way. And what you're describing is breadth first testing. It's just like, let's just try all of the different situations in which we could possibly bring to bear this tool. Yeah, I thought I was doing great testing until listening to you.

No, believe me, I'm not doing it. I just re shared somebody who built his own evaluation suite for the large language models because he didn't like the tasks that were out there that other people did. So he built his own. I am not that guy, but I have figured out a few things like, like, I use it when traveling a lot. I'm always asking for audio tours. Like, what's that building? You know, if I'm in an art museum, can you tell me about this painting? Can you tell me about the artist? Sort of like your buddy guy experience? And it always does have an answer. It's really amazing. It knows everything in the Smithsonian, for instance, even the tractor and the back corner, right? It knows everything from a picture, which is amazing. You take a picture and. And then it'll say, oh, it looks like you're standing at the Smithsonian and you're at this exhibit. You know, I'm like, how do you know that from a freaking photo, man? That's, that's really amazing. Where I was going, it knows Disneyland in much more detail and, and has much more nuanced answers than if you go to like a university in South Carolina and get a tour. It does know that university at some level. It tells you what kind of telescope is in the planetarium there. It was pretty cool, but with Disneyland, because at Disneyland has 40, 000 people visiting it a day, every day, right? And everybody's taking pictures and writing notes or blogs or doing videos or you know, that kind of thing. And so the models have a ton more data about Disneyland to coalesce into the model than they do about University of South Carolina. Right? Because there's not much data about that. And so the, and that gives you insights of how these, where these things really do well and where they don't. Where they are going to hallucinate more. Right? Because if they don't have a lot of data and you ask a question, it's going to make it up, right? Sort of like a human tries to make up an Answer, you know?

Yeah, right? Yeah, yeah, you just kind of bullshit your way through it. Yeah, that's definitely what I.

But has a lot of data. So it's. You could take a picture of a light bulb on Main street and it tells you about it. It's like, whoa, that's really a lot of data. You know, it's really nuanced and could have conversations.

Well, that's what's driving my data hoarding, by the way. I, like, I want to convert you. Like, the fact is I, I guess I am anticipating, I'm anticipating a future where I do get to bring, you know, Tesla Optimus or something similar home. And I want to just feed it a whole bunch of data that I've collected about myself on day one instead of laboriously training it on what I like after I bought it. You know what I mean? Like, like we are all like that university. Like the AI doesn't know enough about us, so we need to consciously create the training set.

I mean, the thing you want your optimist to do, okay, you're going to have it do your laundry, undo your dishes, clean your toilets, weed the garden. You know, all the basics after that, cooking. And if you, if you videoed your mother in law making your favorite dish or something like that, then you can ingest that and it'll make that dish over and over again because it'll be able to watch and it probably can make the dish. I took a picture of my wife's ash soup, which is an Iranian soup, and it told me the Iranian name of the soup. It told me the cultural relevance of the soup. It told me what was in the soup. It gave me a recipe of the soup from one picture. Right? So you didn't even need a video. But if I had a video of, of my mother in law back in the day when she made dinners for us and recorded that, I would have had a lot better stuff to feed the optimist robot. I'm really sad I don't have those. So I tell people, interview your relatives about your family, right. Your ancestors. Because I haven't done that. A lot of that knowledge is just gone. Right. I don't remember who my great grandpa was really in depth. If I had a video of my grandpa telling me about that, I, I would have had a lot more data to remember and, and record people making your favorite dishes. Because then you could train your robot. Hey, can you make me, you know, this dish that I want? You know, this kind of soup or whatever you want? Right.

Okay, we're not gonna top that as like advice from this episode. If you're listening to this, capture the tacit knowledge of you and your family making stuff, the family history that could be lost. That's the most important and most fragile knowledge in your life. Capture it now.

Probably the most meaningful to you later.

Right.

Other, other than videos and pictures, your kids walking the first time or graduating from high school or graduating, you know, that, that kind of thing, it's really important. Most people don't take very many photos. I have a hundred thousand photos on my iPhone, so my robot's going to know me pretty well just from my photos.

Just inference on the photos. Yeah, good, good call.

Because it knows everything in the photo and it knows, it knows where those photos were shot because I kept my location on and stuff like that. So if you send me the three photos of the last hotels you were in, I probably know your credit score within 50 points just from that sparse amount of data. Right? Because, you know, if you're staying at nice hotels, I know you probably have 700 or more. And I, AI can figure that out pretty quick. Right. The more data you got, the more it's going to know who you are, what you. And also, I only take photos of things I really love. Well, add a car wreck or something here and there, Right?

Right. Things that are interesting, things that are.

Highly meaningful to you. Right. And so it's going to pick up what's highly meaningful to you real quick. You know, if you have a lot of photos on your phone of you and your kids playing Legos, it'll know you're really into Legos, right. And, and be able to, you know, play Legos with you. I mean, AI can do LEGO plans really well. Hey, did you ever think of making a, you know, this thing, you know, an Eiffel Tower? You have all the Legos. I saw one guy who. You put the Legos out on a table, take a video of them, and it catalogs all the Legos on the table and so it knows how many green pieces, how many long pieces, and you know, how many weird pieces you have. And it builds a little inventory of the legos that you have. And now it can tell you how to. What you can make with that inventory. Right. And that's pretty cool. Back to you were saying you wanted an AI to watch you. I mean, I visited a company in Germany called Reflect and they make augmented reality software for enterprises for. They actually made the. No, that, that was Mato. But Reflect made a whole bunch of stuff for like Mercedes Benz to help first responders augmented reality. They had me build an electronic motor using augmented reality glasses. And it showed me how to do that. Like take, take this screw, take this screw out. Take this screw out. What, how to do it. And then it, with the cameras, it was watching me work and make sure I did it right. So if I did it wrong, it could talk to me and say, hey, you didn't. You need to go back to the earlier screw. You didn't take that out right. And so it's possible to get there. It's just, we're, it's not quite, you know, that's not there for everything in your world. You know, it's close. It's. The pieces are all laying there. Like if you need to take apart a washing machine and get a, you know, fix a pump on it or something, there is that data on YouTube. Almost, almost everything in society now that you need help with is on YouTube as a video. But getting to the data is hard. And then visualizing it on top of the job and understanding which, which screws to take out, which clips to unclip and you know, how to do it all, it's a little hard. Right.

I do think that that guidance is what is going to make the difference between unaugmented humans and true cyborgs, is the real time guidance to do it right.

So we have, now we have models that can ingest YouTube videos and start laying them out for you. Visually, they're close. So in five years, I think this is a solved problem. Maybe in two years it's a solved problem. Right. So we're, we're about to get augmented in a whole new way. And then you'll have your robot do it anyways. Why, you know, if you had a robot, why, why are you, why are you taking apart the washing machine, trying to fix the pump in the, in the washing machine? You know, hey, Optimus, go fix my washing machine.

Well, if people want to argue with you on or or me on X about whether it's the robots or the augmented humans that'll be doing the best work in the future, how do they find you?

I'm on X. It's X.comscovilizer but I'm on, you know, join me on X. If you hate Elon Musk, then you gotta join me on LinkedIn.

Sounds good, Scobalizer. Thank you for joining me on the augmented life. This has been an absolutely fantastic hour.

I love it. I love how you're refreshing the world too. So thank you. Thank you.