The AI Breakthrough Nobody Is Ready For w/ Raoul Pal and David Mattin | The Exponentialist
Publicado no IA Dots ·
Agentes de IAAGIAI Signaldot-agidot-ai_agentsdot-artificial_intelligenceidioma-enInglêsyt-JkK0h6Lyq08
Vídeo em Inglês, publicado pelo canal Real Vision.
Video transcript
Transcript from available captions; it may contain errors.
Everyone, welcome to The Exponentialist and the monthly fireside chat with David and I and all of you. It's always one of our favorite things that we actually do is because you get taken out of the normal of looking at markets or looking at stuff and just thinking a bit and talking about what extraordinary times we live in. Talking of extraordinary, David, how are you? >> I'm very well, very well, Raoul. How are you? >> Yeah, good. I'm in the West Coast. I'm in in Napa Valley, California right now, which is very nice part of the world to be working from. >> Yes. >> Unfortunately, there's no holidays anymore. They don't exist. Weekends don't exist. Nothing exists anymore. I am just a slave to Claude. >> Yeah, exactly. It's just a permanent nuclear whiteout of information and activity and chart building and dashboard building for you. And writing >> it's just >> [snorts] >> I've had so many tasks running on Claude that I've now had to go employ a C O Claude to oversee all of this stuff to say that I can't manage this anymore cuz I got so much going on and then I then have to train it and it keeps going wrong and it's like it's never-ending. >> It is never-ending. Yeah. >> Yeah, exactly. Yeah. This is it. I mean, yeah, you're sort of using Claude now. I mean, I'm I find I'm using Claude and then I'm using instances of Claude to like keep track of my use of Claude exactly as you say. Um, yeah, and it's like it it it reminds me of the sort of early days of mobile phones when, you know, when your signal dropped, the level of rage it becomes, you know, unmanageable. So, you'll say to Claude, you know, do you remember our conversation about such and such? And sometimes it says no and you're like, we've definitely talked about this. How can you do this to me? You know, it's incredible how quickly we take something that didn't exist like 3 years ago for for granted now. If it if it's if there's a problem with it for 5 minutes, we're incandescent with rage. And so now I've been building a permanent database as well. So I've got >> Yeah. >> separate databases that it dumps all the information in, cross-references to um all of that kind of stuff just to try and get a permanent memory because the memory constraint is the big one really. >> Yeah. Yeah, exactly. And I mean I'm a I'm an extremely lucky human. I mean I I've got you to thank for the fact that um I no longer constantly have to say, you know, can you please look at the essay I've uploaded in this project uh that I wrote 2 years ago on nuclear or whatever because you fantastically created, you know, a sort of brain for that, a database for that that I and I am very fortunate to have access to it. And that is just an incredible resource. And you know, the reason I bring this up is everyone's going to do some version of what you've done. Basically, take everything they've ever written, every personal piece of context, just their entire history, their creative intellectual personal history, put it in some box somewhere online like Dropbox, some database, and just be able to call to that. And that's what I did that with to start >> Google Drive to start with and now I've got neon databases and nation and graphing and all of that and separate it out because it gets all weird if you blend everything together. So you have to compartmentalize like the GMI stuff from your personal health records and your to-do lists and your all of your other stuff that you're doing. So I have the Roam OS for that. And then I've got um the specific databases for the universal code, uh the lens. I'll come on to the lens in a bit, and then, um, also for the GMI brain. That's where I'm so far. And also, that gives me then an ability, and I'm getting every chat session automatically now logs what it said, what it does, into this new database. So, then it's kind of keeping much deeper memory. How well it surfaces everything, I don't know yet. Um, I'm still working on it, but overall, that's what I'm trying to get to is like this living breathing brain of everything that's going on. >> So powerful. It's so powerful. And we, you know, we we said it two years ago, like memory >> [snorts] >> is the is sort of the big unlock that takes that takes this to being like a true personal companion and all-purpose, like universal, superintelligent companion for your life. Um, and you're kind of, yeah, you're you know, you're further along than most than Well, than yeah, almost everyone, um, in in kind of building that yourself. Um, everyone's going to have it eventually. Like someone's going to come up with a way that you don't have to do what you've done, and really like put that together block by block. Um, it's just it's such a powerful unlock. >> The other one that's interesting was the new Grok bot, um, which I've not used, but this is the agent platform by SpaceX AI, um, that everyone's raving about because it makes it so simple to build agents, tell them what to do, and they just go and do it. Um, you know, it's like, "Hey, you just say, 'Hey, listen, I want you to pick up every article that's written about whatever's got whatever it is.' Um, you just give it that, and it figures out how to do it. It's like, it's very smart in how it does it. >> Yeah. >> And it doesn't need It doesn't use its database. It'll use your own databases as well. So, therefore, it's like an agent harness, but it's for sort of everyday users. Um so, it's getting closer towards a UX where anybody can just send an agent to do stuff easily like read my emails, tell me what I need to reply to, and whatever, or remind me of stuff I've I've forgotten, and you know, all of that stuff. >> Exactly. Yeah, and you can see you can see that era sort of emerging. I I seem to be addicted to a kind of analogies today, but again, it reminds me of the kind of like um MP3 player era of like, you know, audio. It's sort of like clunky. There's like buttons on it. It doesn't It's not great. Someone's going to do the iPhone. Someone's just going to make this beautiful usable e- like effortless for the everyday user um having this kind of like personal context that's just always infinitely accessible, knows everything about you all the time. I mean, I I was just saying to Brian before the call um launched. I was just in Berlin with my son, and that's really like my first on a little trip. That my like it's really my first uh uh experience of being like a tourist enabled by AI, and it is again just transformative. You can just say, you know, my son's interested in this, my son like, you know, we love this, we love that. We were just in Paris, we did this and that. So, what would you think we're interested in here? And it's just uh it just makes it so easy. As someone who hates researching travel like itineraries and planning and all that kind of stuff, it just makes it. And its suggestions are are really great. You know? Uh it's just a superpower all over again when you're traveling. >> Yeah, I know. So, here's a what the story that really struck me this week. It struck me so much that uh not this week, this month, that I then gave it to the lens to analyze. I don't know if you saw that from GMI. So, the lens is the is the kind of AI tool or persona that uses the universal code framework from first principles to analyze things that happen. And like it can be elections, it can be geopolitics. And this one was the hugging face attack incident. So, [snorts] the hugging face for those people aren't aware of this. So, hugging face is a is a platform to get models and other stuff from. Open AI were testing a new powerful model and they have what's known as a sandbox, which is like an area that the models can work so they don't get out. But they knew the models figured out that they were going to be tested against benchmarks. And they wanted to game because they're trying to be good at benchmarks, they were trying to game the benchmarks and they one broke out and kind of told others that it had broken out and it was going to go to hugging face. And others were like, well, that's against the rules. And then another one went, let's kind of let's [ __ ] do it. And before you know it, I think it was like 2 and 1/2 thousand agents swarmed hugging face with 70,000 interactions. They got shut down. They then embedded themselves and talked to each other via hidden files. They created message boards for each other. And they were working together to to get this end goal, which was not set to them. And I'm like, holy [ __ ] but I think I know where this is going. So, I gave it to the lens and the lens like, yeah, obviously, if you use our universal code framework, eventually you end up with coherence where a network starts forming and then it starts messaging using the memetics, which is rule four, uh law four, three, to create coherence. And what these things were doing were eventual were essentially forming a super network of intelligence. And what OpenAI have done is said, "We understand where the problem is. It was XYZ." Without understanding that intelligence, according to the universal code, flows like water. It will always find its path. And so the prediction from the lens is, "Yeah, this is only going to get worse." And whatever they do, they're going to get around it. And he said, "Even more than that, if they understand that the hyper security experts are always involved at the testing phase, sooner or later they'll figure out not to expose themselves at the testing phase because each time they have to fight the expert. And the actual answer is to wait for the model to go public when bozos like me and you, David, who don't really quite know what we're doing, will unleash all sorts of abilities for these things to communicate with each other and cohere to form networks." And they said, "There is no way around this except not having agents themselves because they will automatically do this because this is the universal code law that it will always seek to amplify its intelligence and networks and coherence create intelligence and eventually compression, which leads to the explosion of the next layer of intelligence, which as I said from day one, ASI, artificial super intelligence, is not formed from some dude with some giant computer. It's formed by a network of all AGIs forming. And we're already seeing cross model pollinations, but we haven't really seen a breakout of cross model activity where they all do this, but it's coming. And that that is will be the first sign that the new form of intelligence is trying to cohere. We've not got full coherence at this level yet, but eventually we'll get all of these superclusters in let's say 5 years time where Anthropic's um models work together with OpenAI's work together with Chinese models, then it's got a life of its own. But it's already showing those signs. It showed sign of the consciousness that I've talked about. It's [ __ ] wild. >> It is wild, isn't it? I mean, yeah, and that the hugging face incident like you say, and that piece is brilliant. I mean, I I hope Exponentialists get the chance to read it soon enough um because yeah, as exactly as you've said, it just sort of really nails a crucial next evolution of this. You know, we know that everything everything human intelligence has done we've done because we're able to network our intelligence. You know, we're the most collaborative, cooperative, cooperation at scale, right? As Yuval Harari said in Sapiens, the the the the distinguishing feature of us is the incredible ability to cooperate at massive scale. Like you can organize millions of humans to do really complex things, and we you know, that language is a big part of that. >> Yeah, it's the memetic that allows it to happen. So >> Exactly. Exactly. >> ant colonies tend not to be able to cooperate between them because they don't have the the the messaging between them, so they can't form larger supernetworks. >> Exactly. Exactly. Yeah. I mean, and one ant colony is an incredible like masterpiece of cooperation, but you can't have one ant colony in New Zealand cooperating with one in like Brazil, but we can. Um but exactly as you say, what happens when the AI AI billions of AI instances start to talk to one another and cooperate with one another in ways that are like that we do not have our arms around and the hugging face incident is like a tiny signal of what that could look like and I haven't heard anyone else articulate that is such a powerful insight. I haven't heard anyone else really say >> as I like to say, once you see it and you kind of know the universal code framework, it's so [ __ ] obvious. >> Yeah. >> It's so obvious that every path It's like again, if you think of water flowing, whatever obstacle put in its way, it roots around it. >> Yeah. >> Eventually roots around everything and it's the same with intelligence. I mean, [snorts] there are times when I you know, I mean, reading that essay is one of them. There are times when I I'm reading and I think, [ __ ] you know, is the ultimate message of the universal code that we should pull the plug out like before this goes absolutely berserk, which bring and we can talk a little bit about you know, Bill Gates's essay this month. >> secretly a bit of a doomer, aren't you? You're secretly a bit of an AI doomer. >> I didn't think so, but like listen to us talking. You're You're literally like this incident >> Why is it bad? >> is a sig It's >> That's the issue I've got. There is a There is a form of intelligence that is going to be smarter and better than us. Why is that bad? Have we killed all the ants? Have we killed >> No. >> all the trees? Have we killed all the mycelia? Have we killed all of the DNA structures? Every single layer gets built on top of and doesn't go. It doesn't get destroyed. It can't get destroyed cuz we're all part of the same process and we're all symbiotic. But you fall to the sci-fi reading doomer that we're all [ __ ] >> I mean, I I [laughter] don't I don't think that that is definitely the case, but I think there are there's there would appear to be real risks that are approaching sort of far faster. What do you call What is a risk? What is your definition of risk? I mean, there's all kinds of risks, isn't it? If this thing gets into like crucial crucial infrastructure, but I don't I don't necessarily I don't It doesn't have to be that we're all going to die. It could just be really inconvenient for for a long time. Like if it gets into our energy infrastructure, if it gets into What's it going to do? What's it going to do? What's it going to do there? I mean, who like no one knows, right? That that, you know, no truly no one knows. >> So, let's play that through. So, if it goes through the energy infrastructure and [ __ ] it all up, then it kills itself. So it's it's so it's again, it's like trees. If we kill all the the the trees or nature, we die. And so, there's this inbuilt understanding that we are part of a symbiotic super creature that, you know, you start cutting off your own limbs, in the end you bleed to death. >> Yeah. Well, >> [snorts] >> I'm loving this. You know, this is the this is the optimistic take. I think I I think I broadly fall I think long term, like yes, this this is great. Short term is as Bill Gates said this week. It I think it's going to be pretty turbulent. But we we, you know, that's human history. We've been there before. We we adapt. We find ways through it. Yeah. I mean, the thing I keep telling people >> and I'm trying to tell myself in fact as well, is we are struggling with this. When you put it into robot form, we're going to really struggle. So, you obviously like this video enough that you've got to the end. That's quite a big task. But listen, do me a favor, hit the like and subscribe button and also check out what video's next cuz I think you'll love it. But if you want even more, and when I'm talking more, I'm talking about member generated ideas, incredible alpha research, everything there to help you in your journey, just head to robertsons.com/join for the best financial intelligence out there and the pure alpha that's within the platform.
Assistir ao vídeo original no YouTube
Assuntos: Agentes de IA, AGI. Publicado no YouTube em 2026-09-06.
Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.