UL NO. 432: Can You Summarize Your Work in a Sentence?

relies on. Join the 5000 fast growing companies that leverage Fanta to manage risks and prove security in real time. Vanta scales with your business, helping you successfully enter new markets, land bigger deals, and earn customer trust. Unsupervised learning listeners get $1,000 off Vanta at Vanta comm slash unsupervised. That's vanta.com/unsupervised. Welcome to Unsupervised Learning, a security, AI, and meaning focused podcast that looks at how best to thrive as humans.

In a post AI world. It combines original ideas, analysis, and mental models to bring not just the news, but why it matters and how to respond. All right. Welcome to unsupervised Learning. This is Daniel, episode 432. Can you summarize your work in a sentence? All right. Lots of stuff here. Got a new fabric pattern called git wow per minute. This is basically like an estimation of the

value density of any piece of content. So what you do is you essentially take the content and you send it to this fabric pattern, and it'll give you a score from 0 to 10. And if it's anything above like a five, it's got really it's got decent value. Okay. So 5 to 7 is like decent value and then like eight, 9 or 10, 8 or 9 even is like top. I've not seen a ten yet. So just consider nine to be like the top. So it's like how much value am I going to get for this thing?

And basically what it's doing is it's looking at novelty, it's looking at insights, ideas, novelty, surprise and wisdom and basically collecting all those together and saying those are all instances of value. And then it's kind of putting all those together into a score between 0 and 10, and

it is really working well. I put it through some stuff that I knew was going to be weak sauce, and it scored like a two or a three, and I put it through a couple things that I knew were super high value, and they scored eights and nines. So highly recommend this if you want to like, test your gut or use it programmatically for something to rate value. And importantly, it's doing it value per time, right? So that was key to me. So OpenAI did some cool

stuff on Monday. So new model the O stands for Omni GPT four O. It's not a zero. It's not GPT 40. And it's about as smart as four, but four times faster and twice as cheap. So half the cost. And the big thing is that it has vision capabilities and better audio. And this is not fully released yet, but what it's going to allow you to do is basically just talk with it like a normal person. And even better, they released a desktop app which will soon

have the capability of monitoring your screen. So I could be like scrolling through this right now and just like looking at stuff and it'll be like, oh yeah, so here's this. Don't forget to mention this. I could say, give me a summary of what's on the screen right now, and it'll just talk to me. And even cooler is like, if I'm looking at a data visualization or some math or something, it could, like, walk me through that math. It's just unbelievable. I've seen a couple of demos of this.

I'm not sure if they were internally built or if they were third parties. Actually, I saw one by Khan from the Khan Academy and he was walking his son through like teaching him geometry. It's literally you're just talking to this thing. And what's really crazy about this release is that this is going to be for free users as well. So they opened GPT four to free users, which is absolutely incredible. And I'm just super excited about

this whole thing. It's it's really, really exciting. Again, one of the things I've been talking about here with AI stuff is that it's extraordinary when AI is used to do things that everybody needs, but very few people have access to, so that's creating movies, that's creating ideas, writing books, but it's also things like getting a mole looked at, most importantly, education. I mean, the education use case for

AI currently is insane. And it's really, really incredible when you start doing this for Zero Vibe, where one they made the whole model free. I'm not sure what the usage is going to be for that. Hopefully it'll be decent, but not only do they make it free, but it's interactive. So you could literally just say teach me calculus, teach me this infinite patience. It doesn't get mad at you. In fact, it now detects emotions, so if it hears you getting frustrated, it's going to be like, hey, let's

slow down. Let me explain it in a different way. Let me use an analogy that is unbelievable. We're talking about massively multiplying the percentage of people on this planet who are educated because they have access to an extremely high quality superhuman intelligence, superhuman patience, tutor everyone in the world having access to this. And I'm not sure everyone in the world has access to GPT yet. And of course,

you need a computer and you need internet. So that's not exactly true, but the barrier is massively lower as a result of this. It's just unbelievably positive and awesome. And obviously that doesn't take away any of the negatives or the dangers or the alignment problem or superintelligence. Like we've got issues coming up, no question. But the positives are just undeniable. And this model becoming. Free. I mean,

and having these interactive conversation capabilities and emotion detection. Just unbelievable. I predicted there would be more agent stuff, which I think that will be coming soon. I still believe in these predictions, not so much about like what OpenAI specifically is going to do. I'm making broader predictions about what OpenAI and anthropic and probably Google and all other people will do, which is essentially moving prompts into or no

moving agents into prompts. So basically, prompting is the most natural way to interact with an AI, and I believe that it's the most natural place to build agents as well. In fact, you'll just describe what you want. Like I want a team of 100 people creating ideas. I want a team of ten people filtering through those ideas to figure out the best ones. In fact, you won't even

have to say 100 or 10. It'll just figure out based on how much money you have to spend, based on how much time you have and the time constraints and your deadline or whatever. It'll spin up more, spin up more or fewer based on all those constraints, right? So when you say, I need a team of people to do this, a team of people to do that, team of people to do this, then I need the final result to be rated and tested and validated. It'll just go and build all those pieces, which are actually

teams of agents all collaborating together. And all you had to do was put that in the prompt. And that's why I think prompting is everything. It's an article I wrote earlier called prompting. Most of AI is prompting or something like that a couple of weeks ago or a week ago. So I would check that one out. And yeah, they got a desktop app. And yeah, they basically created digital assistants, which I've got a book out from 2016

that was basically talking about this. And it's also in the Predictable Path video, which you might have seen, which is kind of like an updated visual version of that book, but with lots of new stuff, because I didn't know about Gen AI back in 2016, obviously, but they basically built her. Not only did they built her from the movie, but they're also using for the demos the voice that I use for my personal AI with OpenAI, which is Skylar,

which is the voice of Scarlett Johansson, essentially. I doubt that it actually is, but maybe it is. Maybe she got paid for this. RSA was fantastic. Really good. Caught up with a lot of people, did a couple of talks and panels. Probably should have brought something to sell. I didn't really like go in there trying to sell anything, which I guess I should have. But I did talk a little bit about threshold, so that was cool. The

energy and optimism around RSA was extraordinary. I think this is largely in part to AI, and there were tons of products talking about how much they're using AI and everything like that. So we've all made those jokes already. But speaking of threshold, holy crap, I am absolutely in love with this product. Okay, I am in love with this product. And yes, it is a paid product and it is my product. But look at this. This is my current feed right now. I didn't plan on showing this,

but whatever. So how do I blow this up? Yeah. So how to cope with the fear of aging? I haven't looked at that one yet. 2025 models will be more like coworkers than than search engines. Again, this I did not explicitly go follow this person. What is this? This is. Uh dwarkesh.

In order to show this to me, it came into the top of threshold and all my different AI stuff ran against it to assess it and determine what the content was and how high quality the content was, and it had to be above a certain threshold in order for it to even show up. Now I have it set to 50, which is the minimum because I want to see a lot of content, but I can move this up and I could turn off a lot of these things here. This is the first time I've actually

showing threshold kind of interesting. So I can move those and I can say save and then look at this completely change the feed. But now it has to be above an 80 and it has to be in these categories or these categories have to apply. So just I mean every single thing listen to me, every single thing I click on here, I love I absolutely love it. And I don't have to stress about oh did I watch all the. Tests, from leks and from Huberman and

all these different people and all these different I people. No, I just put them into the top of the funnel. I put them into the top of the funnel for threshold, and the whole I pipeline runs against it and I get the analysis. Oh, and I forgot to mention watch this. Okay, this is the sickest video. I want to talk about this anyway. So this is the sickest business video that I've ever seen and I forget how long it is. Soon we're going to have duration things in here. But

watch this. If I go click on the video, obviously it's going to open up.

and more recommendations. And if I go to deep level auto updates, summary, core ideas, core recommendations. Unbelievable. I mean, I've always wanted this and that's why we actually created it. I've always wanted this as a tool and the stuff that we're about to be adding to this, the I'm not even going to talk about that yet, but insane stuff along the lines of everything I've been talking about for this last year and a half with with AI content rating, all that stuff like the new features that

are coming are just going to be insane. And again, the guiding light that I'm using here, like I haven't raised any VC money. I'm not worried about investors. I'm worried about making the best possible tool for me. And this is the tool that I've always needed, and that's why it exists. And of course, when people are like, hey, I wish I had this. Well, if it's something that really resonates with me and I think everyone's going to love, then yeah, we put that into the pipeline and we

actually build that. But ultimately it's it's a thing that solves a problem that I've had of collecting as much information as possible, as high quality as possible, but at a threshold level that I could control the flow. So I could basically go in here and say, you know what, I need it to be a 90 or a higher, and I want it to be only about AI. So watch this. And that pulls it down even further. It's just it's just unbelievably useful. So that's my pitch for that.

It is a paid thing. So yeah. Whatever. Is that even called a sponsor post. Is it a sponsor if it's your own product? I have no idea. But anyway, highly worth it. I would pay thousands of dollars per year to buy this product. No joke, I absolutely would. Oh yeah, and this is the email. So I have it set to a daily email. So you get this email that basically shows you the stuff. Because sometimes like I'll forget to check, I'll just be super busy in

meetings or whatever, and I'll be super busy. And then at the end of the day I get this, look at these. How strong is AI as a tool, as a successor? Deferred happiness syndrome? This thing is unbelievable. And then I click on view right here and it goes yeah this is the business truth one. That's how I found that video. Also from what's it called. This is from Peter Attia on VO2 max and muscle mass. Anyway

it's it's fantastic. I just love it. I got a sponsored conversation had with BlackBerry and Cary Ransom about maritime security. My dad just went on a podcast in, uh, in a studio in San Francisco. I was actually giving a talk a couple of blocks away. My dad's in the studio talking about his music, showing his music, just really, really cool. And I got a link here to the full show, which is this one or no, this one.

And then this one is him playing one of my favorite songs that he's ever written called children of the nights, and you definitely want to check that one out. It's fantastic. And security Dell got hacked real bad through an API. I think if I could invest in any sort of legacy non-ai tech, which I guess it's quite AI related at this point. And that's kind of the point, is that API security is like everything, because everything is becoming APIs that get wielded by AI, and your company is

essentially about to be mostly your API. So you've got to lock that down. So I feel like of all the traditional security spaces that I would want to be in right now, which I don't want to be in any of them, but if I did, it would probably be API security. And speaking of that, my buddy Joseph Thacker has put out a couple of different pieces about API security assumptions about AI, and also one about agents and authentication. So check that out. Both of those and

CSA has a new alert system. They're basically allowing companies to sign up. And then they send them Kev alerts if they see something associated with their company. Attackers are using Microsoft Graphs API for malware comms. I love these side channel things, I just. Absolutely love them where you could, like, use something in a way that's not expected. The best example of this that I remember is Gmail drafts. You actually,

you you have a shared email account. You create a draft on one side, you don't send the email, and then someone else has access to that email account and they just log in and read the draft so the email never gets sent. So that's another example of just like using a a platform in a way that's sneaky. All right. So one password. Thank you for sponsoring. That's Collidin one password and Russian influence campaign looking at magnifying

and taking advantage of the campus protests. Google's making MFA setup smoother by letting you skip phone number for options like authenticator or keys. Marines are testing robot dogs with AI aimed rifles. Awesome awesome awesome awesome 95% of international data travels through undersea cables. Okay, so this is one that's really interesting. I've never understood why undersea cables weren't

more targeted and just like totally abused by terrorists. I guess there is the small matter of like you're in a boat, you're way, way above this cable. It takes time and money and skill to get down there. So I guess that's a barrier. But once you're down there, Holy crap. I mean, it seems pretty easy to do damage to these things, and the amount of disruption you could do if you're disrupting international internet traffic is is

pretty serious. Plus, you think this has got to get easier. Like, if you can send drones down there like underwater water, drones with like explosives, this wouldn't be easy for low level attackers to do. But for a state actor to want to just continuously like send drones after it or teams or whatever, it seems so easy for any state actor to basically sever internet communications through undersea cables. And the US just closed another door essentially for Huawei getting

AI stuff with Intel chips. So another attack against the Chinese trying to get AI chips. And someone from Andreessen Horowitz basically says half a Google staff or just pretending to work, they're not actually doing anything. Yeah, and a lot of people are talking about this trend of fake work, and I think it's a giant mess. It's fake in multiple ways. Right. So I talked about David Graeber's bullshit Jobs book. Really, really good. Basically, we have all these

jobs that shouldn't exist. You have all these people with massive salaries, hardly doing anything and somehow being justified, especially in Google where like they're not making much new stuff at all. Because they don't have a product management focused product teams. They're just like engineering focused teams. It's a total mess. And you add AI to this and it's just it's going to be a total mess. The safest thing to do, in my opinion, the safest place to

be is building new things, creating new things. Be a programmer, definitely be a programmer. But more so be a thinker who focuses on problems. Solve those problems. Have the tech skills. Have the humanities skills. You want to be broad. You want to be general. I recommend everyone has tech skills. Everyone has programming skills. Everyone has AI skills. Everyone has

writing and speaking and presentation skills. These are like the universals, and I'm actually working on a big post around this, which is going to be a member post around this. That's going to be super key of like how to get ready for what's coming, what skills you need. So that's pretty much a teaser for that. Mitra is partnering with Nvidia to create a $20 million AI supercomputer to to make US government operations more efficient, which is cool

and terrifying. Joe Biden is converting the Foxconn debacle, which didn't go well in Wisconsin into a $3.3 billion Microsoft AI center. I feel like this administration is doing well on AI, both in countering China but also building ourselves up. And the acquired podcast is basically like biographies of people, except for it's for companies. So a lot of people are super into that. In the Bay area, Biden is quadrupling tariffs on Chinese EVs. These things are a real

threat to the American car industry. So he's going to tariff them quite a bit. This one said 400% but quadrupling. Yeah, this one said quadrupling. But I've seen it say doubling. So I'm not sure the exact numbers. But he's definitely increasing the tariffs. California is going to start charging your bill for electricity based on your income starting in 2025. Scientists have found all DNA and RNA bases in meteorites, basically hinting that building blocks might be extraterrestrial. This is

not surprising in the slightest. To me, it just seems logical and there's life floating around everywhere. I don't think life is that special. In one characterization of special in terms of special of like, oh, we're we're a snowflake and we're the only ones who have it. I don't think we're special in that way. That does not mean it's not special in terms of having high, interesting, unique value. I think we absolutely have. That doesn't mean it has to be unique. Does it mean it has to be

one of a kind? Okay. Scientists constructed a one millimetre square piece of the human cerebral cortex at a nanoscale resolution. So they basically get to see the actual neurons and I believe synapses as well. I haven't looked at this document fully yet or this image fully yet. I'm not sure I would even understand it, but I love the

fact that we could see that. I love the fact that it's only one millimetre square and we're getting all that, all those, you know, millions of different cells and neurons and everything in there. The study says very large study, 154 million deaths prevented due to vaccines. Streaming is basically cable. Now there's a racket around emotional support animals. That was quite hilarious and disturbing. And I think Mark Andreessen is mistaken about paradox applying to AI. This this podcast here.

He basically says that AI will not make it easier to build companies because of Jevons Paradox, which basically says that you think if you make roads wider, it would reduce the amount of traffic, but in fact it just makes more people drive on them, which actually increases traffic. Now Mark basically said, well, I think that means that AI making it easier to start a company will make it actually harder to start a company because more people will be doing it and the standard will be higher.

And I think he is misunderstanding that. And so basically what I say about that is it's only when it's a resource. Jevons paradox applies to a resource. It basically says if you make that thing cheaper, it won't reduce the total amount of usage across the world because more people will use it. So usage actually goes up. That's the limit of the paradox, at least as I understand it. And the situation with AI helping startups is actually a

friction issue. It's reducing friction to get into the market. It's not about there's only so. Much of X or so much of Y, and if it were, the thing that would be talking about is there's only so much energy. There's only so much compute. So that's a limited resource that would actually apply to Jevons paradox, but not friction of getting into a market. So I think he was

wrong about that. And that basically cleared that up. Courage is everything I've been throwing around this three level system here. Courage is action versus fear. Discipline is courage versus laziness. You see how they build on each other. This one is this one. And then success is discipline versus mediocrity. So it goes from courage to discipline to success. So everything you want is on the other side of courage, whether it's courage against fear or courage against laziness. And

this applies to all sorts of real life situations. So it's like hard conversations, you know, getting more healthy, eating right, not wasting time doing the wrong things, or quitting a soul crushing job. So if you're trying to quit a job, it's like you're scared if you're eating right, it's a matter of discipline becoming fit. It's a matter of discipline. You got to go to the gym. You got to, you know, eat right. You got to exercise. Not wasting

so much time with games. It also discipline, having a hard conversation. That's courage. I guess what I'm saying here is all of these can be thought of as courage. And what I'm trying to do here is build a frame. I want to build a frame that says when I'm being lazy, I'm not being courageous. So I want to frame this to myself as have the courage to not

be lazy, have courage to go to the gym. So if I'm presenting that as there's somebody to go save in a fire and I'm going to get burnt and that inflames my my interest to be a moral person, well, that can motivate me. I will walk through that fire to go do that thing because that's what my identity is. I want that to be my identity. And it is. So that's the thing I want to do now. I don't naturally, and most of us don't naturally have that for going to the gym. It's not a moral issue.

I want to turn it into a moral issue. I want to frame it as a moral issue so that I'm like, look, oh, you don't want to go to the gym. Okay, well, that's because I don't want to use bad language, but I basically want to talk to myself in this bad language of, you know, man up, go do it, have the courage to do the right thing. And I think that can massively help. Recommendation of the week. Know yourself. Think about what someone should say when they

introduce you. So this is Sarah. Sarah does this do you want them to say about Sarah? What do you want them to say about you? Mine is something like he has a company that builds products and services that help people transition to something called human 3.0, so they can survive what's happening and thrive with what's happening with AI. It's something like that, and it's different for different audiences.

But that's the basic vibe. And you basically want to know what this is for you, not only so that you can deliver it, but that other people can do it when you're not there. And the aphorism for the week where your fear is, there's your task. Where your fear is, there is your task. Carl Jung. Unsupervised Learning is produced and edited by Daniel Meisler on a Neumann

U87 AI microphone using Hindenburg. Intro and outro music is by zombie with the why and to get the text and links from this episode, sign up for the newsletter version of the show at Daniel meisler.com/newsletter. We'll see you next time.