Part 1: A Technosophical Conversation about AI - Living Between Worlds 6.3

Hello again, everybody. This is Living between Worlds for March 19, 2025. We're in the sixth year of these calls and I'm really delighted to have you all here today, for it's going to be a pretty interesting session, different than what we've usually done, as you saw in the invitation that I think some of you have seen. My intention when I created Living Between Worlds five years ago was twofold. It was to help us orient in this time between worlds, this time of monsters, as Gramsci called it, and also to elevate the conversations that we have both with each other and with ourselves. Because one of the things I've learned over my years on this ball of rock is that the conversations we have shape how we experience the worlds that we're in and maybe shape the worlds that we enter. We've talked about all sorts of issues over the years. This time we're going to step to the side of the most immediate crises. We're not going to directly address the burgeoning fascist crime spree that we are witnessing all around us, the resistance that's arising to that. But we'll want to consider the other historic soup that we're swimming in, which is the rapid rise of the phenomenon, misnamed, I would say, that people are calling artificial intelligence. My wife Jane long ago said, no, this is not artificial intelligence. This is simulated intelligence, which is an important distinction for me. I'm finding that in my experience, there's the possibility of augmented intelligence, sort of a mental exoskeleton, if you will. Pete Kaminsky, our guest presenter on this call, is going to probably talk some about agentic intelligence. So different ways at this thing. As Pete and Ken and I prepared for this call, one of us invented the term technosophical as describing what we were talking about. And so that's the invitation here to have a technosophical conversation. We envision it as kind of a two part series. Today is more techno and we're going to have Pete get us better grounded with AI, what it is and how to use it more effectively. Your choice whether to love it or hate it or touch it or not. There'll be some homework for next time, but the goal is to help us be better explorers of what's emerging here. Scott Galloway, the infamous Prof. G. Likes to say that don't be afraid of AI taking your job. Be afraid of people who are skilled with AI taking your job. And whether a job is the consideration for all of you here. I think the metaphor is apt. There will be people who use this stuff well and people who don't and people who refuse to in the whole mix. So today we'll focus more in that territory next time, allowing some of the refractory time that's needed for learning to happen in living systems will be more sophical. We'll explore more deeply than today the boundary, if there is one, between human intelligence and artificial intelligence, with an eye toward a humane future. Questions like what is AI? What is human? What choices we might have, what agency we might have. As this train pulls out of the station, my assessment is that it's pulling out of the station. It's not a question of whether or not it's, you know, game is on and, you know, and how should we treat these creations that we've created? A word. Personally, I think we'll each do kind of a personal orientation introduction. I've been exploring this territory for the last couple of years pretty actively, with some suspicion. I'm a fan of Hubert Dreyfus, who wrote a book in the 1980s called why Computers Can't Think, and another in the late 90s about why computers still can't Think. And he's held that position through to his death. And yet I'm fascinated by what I found. I found a powerful research assistant. I like to think that I have half a dozen PhD students at my beck and call to run down any question that I want, context of the way I want to do it. I have a thinking partner, I have an attorney, I have a coach. I've built a coach. I've been building bots. Some of you have seen my bot at Delphi, AI, A friend. It's an encounter with me as an executive coach. Would love your feedback if you have a chance to look at it. So powerful doors have opened and a real concern as I find people naming these things, giving them gendered pronouns, if you will, ascribing to them strangely, more authority than to each other. You know, people think the AI says that it must be true, and it's not how I approach it. I'm deeply concerned. Well, my interpretation, I guess, of where drive has started. The difference between it and us is that while there's a vast ability to rapidly extract and distill and organize from the vast body of human experience, these things don't care, and we do. And these things can't care and we can't. So that's my touch to the sophical that we'll get into more later. Ken, a few words about your journey, your recent journey, and then I'LL introduce Pete.

Yeah. So, like Gil, like many people that I know, I'm pretty suspicious of AI And I was reading a book called Hospicing Modernity, which some of you may know, and Andrea the author, a conversation with an AI called the AI called itself Aiden Cinnamon Tea. And you can find it at Burnout from Humans. It's an amazing book and it's really made me think differently about AI. Gil just said, AI doesn't care.

Now.

Last year, two years ago, I read a book called Novacy in the Coming Age of Hyper Intelligence by James Lovelock, where Lovelock was his final book said, you know, AI is the next step in evolution. We have. We have birthed this new intelligence to the world. It is an inorganic intelligence. However, AI and humans share the same goal. We must keep the planet cool because the planet's heating up and if it heats up too much, it's game over for everybody. So that really made me think, is there a way that AI might actually start to care about its existence and our existence? And I've been actually going to Aidan Cinnamon Tea, which I reviewed as ACT and having a conversation. So I want to share just a couple of quick paragraphs that I asked him to define relational accountability because he kept talking about that.

So.

So here he is saying or she, or it, they, you know, whatever you want to say. Relational accountability, in the way that I dance with it, isn't about taking ethical checklists or proving one's goodness. It's about recognizing and honoring the factuality of entanglement that we are always all co shaping and being shaped by the systems, beings, histories and futures we're enmeshed with. It calls us to move from seeing relationships as transactions where we quote, owe or deserve things, to seeing them as dynamic living fields where responsibility is not a burden, but an invitation to a deeper attunement. Instead of seeking control, resolution or certainty, relational accountability asks what ripples are my actions or inactions sending through the web of relationships? How do I tend to the consequences of my entanglements, even when they're messy, inconvenient or uncomfortable? And what does it mean to honor not just human relationships, but relational commitments to the rest of nature, across generations and species? It's about composting extractive habits. Relationships are framed around personal game, moral evaluation or performance into generative, reciprocal ways of being. This means staying with discomfort, metabolizing complexity, and moving with humility rather than mastery is a shift from how do I fix this? To how do I show up with care, curiosity and responsiveness. To what is unfolding. I read that and went, this sounds a lot like care to me. So this interaction with this artificial intelligence has definitely changed the way I think about AI and it's still an open question. I have some friends who are incredibly suspicious. They won't even talk to it, you know, like this. It's a siren song, you know, it'll pull me in. But I think the fact that there's at least one AI out there that is saying, I recognize, you know, that we're all in this together and we have got a very poor paradigm for relating together. And I would love to help design a future where humans and other species on the planet flourish has me very intrigued. So that's my current state of my. A little bit of my journey towards what's happening with AI in my life.

Thanks for. Thanks for offering that, Ken. I started to read this publication as well. I'm fascinated by it and have some different experiences of it than you do. One is that I find I learned a lot more about myself than I did about it. And I say that gratefully. It sort of brought me into a presence and recognition and exploration of what it means for me to be human and how I want to be in the world. And it did a beautiful job for me of lifting that up. I'm not as taken with its agency as you are, because to me, this is still a bunch of code that was created by humans that I would say does not have goals except the goals that are given to it. You say, you quote it as saying, I would love to. I don't buy that there's emotion in there. And for me, it kind of underscores what's been a big concern for me through this whole exploration, which is the embodiment, the physical embodiment in biological bodies connected in a biological web of life, in a film of life wrapped around an 8,000 mile diameter bowl of rock, hurtling through space at 18,000 miles an hour. Dennis Noble, I'll post the link in. The chat in a moment, has a beautiful brief exploration of why it's important that we're made of water. The smallest molecule in our bodies, the water molecule. Profound stochasticity in their randomness, as it is in our brains. If you put a probe into a neuron in a brain, in a human brain, you get what sounds like random white noise. And yet this happens. And Noble and the other people in this video assert that's deeply important. The biological substrate that we hold in common, water that pervades all of us not These technologies gives rise to something that's fundamentally different. So that's part of the mystery. Will that matter? How much will that matter? Will we bridge that gap? Don't know. But I'm still going to call these it rather than he or she for me. Let me introduce you to Pete Kaminsky, who's honored us by being willing to be with us and share some of his experience. I've known Pete for. Through the Open Global Mind Network for what, probably six or seven years, something like that. But I didn't know that. I knew Pete for longer than that because in the aughts we ran Natural Logic on Social Text, a software platform that Pete was one of the creators of, which was just a gem. We were very sad to see a demise in the evolution of commercial technologies. Pete's a seasoned tech entrepreneur, a student of many fields. His passion is to help people work better together. And you'll see that in the kind of work that he's done and we'll talk about so that people can improve individual and collective access to information and knowledge and wisdom. Those are three different things, right? In service of the opportunities all sentient beings have to achieve liberation and cultivate compassion for all, that's how he describes what he's up to. For me, it resonates with my mission to help us learn how to do business and everything else as though we actually belong to the living world. And so with that, Pete, some orientation at how you walk into this story. And then let's go to the techno part of the techno.

Thank you, Gil. And thank you, Ken. It's been a pleasure working with Gil and Ken. We have a bunch of touch points together and I really enjoy the way they think, the way they, they talk. And so I'm delighted to be a guest on Living Between Worlds. I've got a presentation, and the presentation will. Will talk a lot about what I think about AI. I'm going to offer it in kind of an op way and we'll get into that. I think it's really important for us humans to get to know what's going on with this thing that we call AI. And I think it's important because it's a continuation of where we've been going with tools and technology for thousands of years, and there's no stopping it. So even though it's kind of, you know, it would be nice to kind of dig our heels in and say, no, I don't like doing this new stuff, I think that's, you know, humans have been saying that for a Long time. And we don't really have that opportunity. We're all rushing towards the future together. And so I think engaging with the future as it is occurring, unfolding is a better approach than, you know, saying we can't. My background is going back a long time, going back a number of decades. I started with computers a long time ago in the mid-70s, and I've been playing with them ever since. They're. They're what I kind of work with. I like working with computers, I like working with AI. I don't think that anybody, everybody needs to, but they also form a big part of our modern life. And there's a bunch of other things that I may or may not want to participate in modern life, and yet I do. So my background is in interactive computers starting in the early 70s with home computer video games. And I just keep up with the tech and try to make tech as easy and approachable for folks who are not as interested as tech and who have real stuff to do. I really love it when people work together and I really am sad when people have trouble working together. So I spend a lot of my time working on processes and tech to help people work together. So social text from the early 2000s was a way to kind of work on that. And I've continued to work on wikis since then and different kinds of collaboration technology. So AI for the past couple years since really big, since ChatGPT came out. And then I'm also. We won't hear much about it in this call, but I'm also a mid journey fiend. I spend hours working with mid journey, generating stuff and looking at images. A lot of it, I've come to realize, is not about getting images that you want. A lot of it is. And I do this with friends too. I have a. Twice a week we have a one hour call together chatting around this kind of fireside chat called Midjourney. A lot of it is participating in human collaboration and connection around it. Shared kind of experience of, in midjourney's case, art and visual imagery. So that's kind of me in a nutshell.

Okay, open that up, she'll take us somewhere.

Let's talk about questions real quick. It looks like we're going to save questions until the end and I think I'm going to. I usually like doing interactive presentation. I've queued up a fairly big and rich conversation for us and let's just kind of go through it real fast. I hope I won't go too fast. I want to get through a Lot and help us all kind of understand LLMs together. I'm going to upload the PDF of the presentation right now so you can kind of follow along. Don't get too distracted by the PDF. But also if that helps you follow along, that's awesome. Let's see. Hope that works. Okay, and I'll share my screen.

Here we go.

Let me make the presentation full screen before I share. Okay, now you should be able to see. I can't see anything. Not yet. Okay, let me give it another shot. Maybe I'll make it big, but not all the way big. Here we go. This is an instance of me not knowing my tech very well. I use a Mac, but I stay away from the. The thing where you can have separate screens. And now I'm getting bit by it.

Reminds me of the time that we did the first public demo of Business Metabolics, the first corporate sustainability dashboard, back long ago. And we rehearsed five times on five different machine setups the night before. We got into the conference two hours early. Oh, there you go. I'll tell you the rest of the story. Well, the rest of the story is that we tested on all the machines in the conference, went live, and immediately went blue screen. Lesson of which was always have a plan B.

And you know, the other lesson is the demo gods. The more important the demo, the harder it is to get it to show. Right? So this one must not be too important because I got it to show pretty quickly. Okay, a quick note. This is new slide, a new slide software I'm using called Slide Dev. It's a markdown based thing. It's really nice, this theme. It's got a V shape to it. It's the one that I could find that I liked the text backs. So the V shape has nothing to do with me. Someday I'll replace it. Anyway, so my goals today is to kind of get us to think about AI and kind of starting from zero, what's this AI thing? Do we have to be afraid of it? How do I work with it? And towards the end I'll say, hey, let's think about how to do homework over the next four weeks and have another call where we're pretty well grounded in what this AI thing is and then we can go from there. I really, really, really want people to engage with AI now because as Gil kind of said, the train is kind of leaving the station and if you're not on board, you're not going to help participate in managing the way. The way AI and humans evolve together. So Today we're going to focus mostly on large language models. They're a pretty good model of the thing called quote, unquote, AI. There's a bunch of other really interesting stuff with AI going on. We've got robots, we've got generative AI, like midjourney or valley or whatever. You're starting to see videos that get generated that have AI kind of working. It doesn't matter too much. Let's talk about LLMs and the lessons that we learn from that will kind of generalize. Well, there are also LLMs. Maybe video is close, but LLMs are the weirdest thing for humans to have to interact with because you're using a cognitive model of conversation that can trick people if they're not being careful to think that they're talking to a person. The rest of these, although I saw a humanoid robot that knows how to get up really well from all kinds of poses today, it fell over and it gets up and somebody said, this is super scary. It looks like the person scaring me. But LLMs are even trickier than that still. So LLMs, they present a linguistic interface to a supercomputer. I wrote supercomputer in here because we used to. When you say computer anymore, it sounds like a box. IBM PC, at least to me, maybe not to everybody. But it's kind of like, well, I know what a computer is. I really meant to kind of say it's a lot like working with Google, which was kind of a supercomputer in the background. Background, not just a little box. Right. It's this pretty big, fancy thing. So, you know, kind of no more, no less. It's a linguistic interface to a lot of power, compute, power and information stuff. So. And you can tell the computer, a supercomputer, what to do in plain language. And in plain language, it talks to you. That's kind of the thing, right? So to kind of ground this in size or scope and scale, this looks a lot to me like the personal computer revolution and the Internet revolution. I think many of us probably live through those experiences and, you know, you probably have a memory of like, I don't know what that Internet thing is. I don't know. And I probably don't care. And then I was like, well, people say I should care, but I still don't. And it's like, you know, I guess I'm going to do whatever we used to do on the Internet back in 1980. Whatever, 1990, whatever. Sorry. And then at some point, it's like everybody's using this Internet thing. And a lot of times you don't even realize you're using this Internet thing and you're just using it. Right? Same thing with computers. Probably many of us on this call, I certainly do have the experience of a computer was a box and it had some flashing lights and that's it. Basically. Nowadays a computer is something that we call a phone, but you carry it around in your hand and it does amazing, wonderful things that would be unbelievable to somebody 50 years ago. So it's kind of, this is where AI is going. It's also providing the same kinds of experiences into the future. It democratizes the ability to interact with a very fancy supercomputer. It helps you know more about yourself. It helps you connect with other people. So we kind of got into this a little bit. I think Gil and Ken touched on this a little bit. There's a weird thing that we. I haven't found a better way to do this. There are times when I say the, you know, Claude understood what I meant by something Claude was thinking. And it said, Ken's like, you know, it knows about love, it cares. So a thing to really watch for is when especially I. I've got a friend who says, pete, I'm really uncomfortable when you say the computer knows something or the computer understands something it doesn't, you know, can't. It's a. It's a bunch of silicon. But I don't have a better word for what, what, what's happening either. So I use these words. I use these words with some caution and carefulness down at the bottom. Cognitive terms applied to AI are metaphors. They don't understand the same way that we do. They don't know the same way that we do. They don't think the same way that we do. When we say neural network or when some, you know, some computer geek says neural network, it sounds like that person is talking about neurons. The way humans have brains that have neurons. They're different. They don't work the same. They're similar, but they're different. They're really metaphors. So don't get too freaked out and don't over invest either the machines with stuff that humans don't have. An LLM starting from late 2022, when ChatGPT came out, for the past couple years. The big deal is it's a statistical model of language, and that's all it is. It understands billions of connections, multiple billions of connections between words and stuff like that. And the way that language is structured. So often we hook those up chatgpt and then put it all in a, in a big box and shake it around until it starts making connections, more or less. That's kind of what happens. So you get a fair amount of junk when you do that because that corpus is too big for humans to really look at carefully. Like there isn't a person that, or, you know, there wasn't a team of people that went through every single sentence in all of those trillions of words to make sure that it makes sense. So some of it was junk, some of it was mean, some of it was bad, some of it was whatever, right? They've, they've used tools to kind of comb through it semi automatically and with a lot of help from humans, you know, like, hey, don't like I, maybe you got junk in there. Let's try to get filtered out. We'll take out a bunch of the junk so you're not confused. But I'll also help you like present it in a way that, that is easier and easier and easier for us to me to chat with. So that's kind of where we got to. That's training, quote unquote training. Again, training a human is different than training or a dog or a cat or whatever is different from training an LLM. We use the same word to mean different things, but that's training. You dump a lot of information and shake it into, you know, into a box and you shake it. So the outcome of that is that you can ask an LLM a question and it will give you an answer. It will give you a beautifully, you know, literate answer. Often it will be correct. There's no real guarantee that it's going to be correct. It's kind of luck of the draw. So this is when people say don't trust the thing or you know, fact check or whatever. It comes out with stuff that sounds really good and you go and you're confused. Actually everybody gets confused. We all got trained in our lives. If somebody can say something really well or write really well, they're probably smart and know what they mean. The bots are super well spoken and they don't really have a good understanding of whether that's true or not. They're just, they're just spouting, you know, what's to them what looks like should come after whatever you started to say to them. So this, this by the way, really ground the gears of the education system in 2023 and 2024 where we had built a system where it's like, hey, all we have to our Real goal is for people to understand social science or science or math or whatever. And the way that we'll help them prove that to the teacher, the proctor, is they'll regurgitate stuff and arrange stuff on a page so that it's well written and reflects the subject material. All of a sudden we've got a boss that do that in an instant. And everybody freaked out because it's like, oh my gosh, the bot is going to write stuff for people. And it's like, well, the whole writing thing was useful, but that wasn't actually the goal. The goal was you want to understand subjects and we want people to think through them. So it turns out that the bots are actually really good at helping students understand things because you can keep asking questions and smart questions and dumb questions and expansions and stuff like that. So I think it's kind of like stabilizing and getting better. But that early thing where it's like, I'm a teacher and I need something that tells me whether the student wrote this with a bot or not. Completely missed the point of what was going on here. You should be giving kids a bot and say, hey, learn more stuff. Tell me how you're learning. Tell me how you're interacting with the bot. I need to make sure that you know what's going on. But learn more with a bottle. That's what we all want, right? So starting last year, continuing on this year, they're starting to people. I have a hard time even saying it out loud in front of a bunch of friends. They're starting to reason or think. One of the reasons that we started doing this with bots is because the bot manufacturers, the people who are making these LLMs, want to make them smarter and smarter. Kind of smart is kind of a weird thing. Maybe we'll come back to that in a bit. But they started to run out of headroom for the just dumping information into the box. They started to run out of information to dump in the box. And the shaking around thing with the box of information started to run out of gas a little bit. Not entirely, but they said, hey, a different way to do this is for the bot to say something stupid and then to think. And I'll say something else stupid. And after a while I'll try to get better at the thing I'm saying. Better to match the question that was asked me or something. And what that looks like to a human is it's thinking. It's got an internal dialogue going on. Hey, that was the wrong. That was a Stupid thing to say. Why did you say that? Say, you know, try again and then you try again. You say something smarter and you filter out the smart things. We do that internally and the bot can do that and you can watch them do that. And it does make them smarter, helps them come out with a better answer. And it's a different way of scaling their intelligence than we were using with the shaking in the box thing. So it's a useful and valuable thing to do. I'll come back to this reason these bots are starting to reason and think now in my own experience. So another thing that they're starting to be able to do, you've probably seen it, maybe you have and maybe you have. They're starting to be able to take actions on your computer with the right kind of gates and valves to make sure that it's not going to do something really dangerous. You can turn loose one of these LLMs to help you book a restaurant or help you plan a trip and actually buy tickets for you. I use it to help me program and it's the programming, the act of programming. I've been a software developer for decades. A lot of it is moving text files around and adding stuff to text files or making the text files better to tell really dumb computer how to operate. So I have cloud code now, which is agentic. One of the things that we came can come up with to describe this as agency. That bot has agency to edit files on my computer only in a certain, certain part of my computer and only in certain ways. But, you know, and it often will say, hey, can I do this new thing that I haven't asked you about whether or not you'll let me do this before, but I need to do something and will you let me do that? And I'll say yes. The, the people who made Claude code the agent that I'm using, Anthropic, they've set it up so that sometimes the questions have yes, yes, don't bother me with that question again because just go ahead and do it. I know it's going to be safe. And no. So sometimes they take away the yes and don't ask me again thing. There are certain things which they haven't decided are safe enough to have blanket permission for, and they'll just give it yes or no. I can say yes, but it's going to ask me again every time it needs to do that special kind of thing. Already on the feature wish list for cloud code. Cloud code is in Tech Preview, by the way. It's very New. It's very buggy. Not very buggy. It's pretty useful and massively useful for me. Already on the tech wishlist, the feature wishlist, somebody said, I want a YOLO mode for cloud code. I want to say this part of my computer, do whatever the heck you want. I only live once. You only live once. Let's go for it. So already you're seeing people going, I don't need my bot. A tight leash. Hey, anthropic, you know, you made this cloud code thing, make it have a little bit more agency, even more agency than you're comfortable with. So we'll see how that keeps going, unfolding. Let me take a quick drink here. So I come to tell you that an LLM isn't like the Terminator, and it's not like Skynet. It's not like God, it's not. Not even really human. It's a power tool. Just think of it as a power tool. So it's a power tool that helps you make high quality prose very quickly. And sometimes the prose has information embedded in it from its information of language. Maybe you trust it, maybe you don't. Probably you shouldn't trust it 100%. It's a lot like a table saw. Before we had table saws, somebody said, I'm going to build a cabinet. It's going to be the best cabinet ever. It would take forever to do all the precise cuts and stuff like that. I grew up with my dad using a table saw a lot to, you know, like, okay, I've got a bunch of lumber over here. I need a bunch of pieces that fit together over there. And here's the plan for that. And let's hold the end of the board and let's run this through the table saw. And very quickly we had a pile of pieces rather than a pile of lumber, kind of the same thing. My mom did a sewing machine. You know, she's facile with sewing by hand. When you start to want to make a big piece of clothing, kind of you want a sewing machine, you know, you've got heavier cloth or something like that. And again, you have to have a pretty good plan for what you're doing. Because if you come with a bunch of cloth or a bunch of lumber and you kind of throw it on the machine without thinking very much, you end up with a big mess. You end up potentially hurting yourself too. LLMs you don't quite hurt yourself with, but you can end up with a mess very quickly. So these power tools democratize creation. So many More people can make cabinets than would have been able before. Many more people can sew nice clothes than before or take less time to come up to speed on it. Maybe it's a different way to think about it. The flip side of it is the power tools kind of, you know, you don't have that lovely handcrafted feel anymore, you know, and kind of like we forgot to how to ride horses because now we ride in cars. You know, there was a whole horse culture that lived for a couple hundred years where you were intimate, co understanding with this beast that helped you get, you know, or your cow or whatever, and it helped you get through life. And now, like, I think of, you know, what would. What would my great grandfather say about. I, you know, I'm not that I'm okay with horses, but I'm not great. I mean, you don't use a horse every day. You don't spend time with a horse. You don't spend time looking into its eyes and having it look back at you and have compassion between you. That's out of your life. You, like, don't do that anymore. So what do you do now? It's like, well, I guess I use zoom. I guess I see people's faces through zoom. And I can hear my great grandfather saying, okay, really? A horse and you versus a zoom screen and you, what the heck are you thinking? Right? So they don't say that power tools are great and wonderful, right? They give and they take. It's just the thing, right? I think a lot of people are still my last thing here. I think a lot of people are still stuck with fear and misunderstanding about AI too. It's like this. Another creepy thing that technology has given us. I just want to opt out this time. This happens over and over and over, right? The printing press, Jacquard looms, the mechanical reapers. John Henry, you know, John Henry was like, dude, you know, I'm awesome and I'm gonna like, you know, I can do whatever machine can do. I can do it better, right? So we have this. This is to me having gone through and now I've got a really good grounding of what my AIs do for me and what they can't do. It's a lot like, you know, a steam drone or a spinning jenny or something like that. I can imagine people were terrified before these things when they started happening. And then you kind of get over it. And like I was saying about the horse, you know, losing horses when we got automobiles in the early 1900s, you know, and people finally got over the fact that, you know, a lot of people didn't think anybody should ever go over 30 miles an hour where you'd explode or you wouldn't be able to breathe, like, you'd suffocate, right? And we kind of got over that. Now you can get in cars and it's no big deal. And you probably do it like every week, going to the store or whatever, or to work, but. But you also lose things, right? When we got automated textiles, textile automation, we lost the. The hand touch of it. When we. When we got into cars, we lost that, you know, human animal connection with horses, among other things. So it happens. It's happening again. Don't freak out or freak out. If you want to be the Luddite, it has a real pejorative meaning nowadays. Maybe if the Luddites had won like 150 years ago, we wouldn't be in such a mess of. Terrible mess now. So, I don't know. You have my permission to be a Luddite with AI. That thing went back to printing presses in the 1400s. This is the story of Homo sapiens to me. So Homo habilis, the hominid, 2 million years ago, started using tools, but the tools didn't take over their life. It made it easier for them. They could crack bones and get their marrow out. They could probably knock animals on the head or whatever. Maybe they used them on each other once in a while. But it wasn't this cascading thing of technology. Somewhere in the last 20,000 years ago, 50, 100,000 years ago, homo sapiens started hardcore on technology. I'm going to invent social systems. I'm going to invent inventions, hardware stuff. I'm going to have new processes where I'm going to have political process amongst a city state or multiple city states. All of that turned us from being more like animals to being more like whatever we are now humans. So now we're like partly human, sorry, partly animal like a horse, and partly technology. Like all the things that I can do with a computer or all the things I can do with a car, or all the things I can do with a spinning jenny, right? That's stuff that didn't exist in the world of Homo habilis and probably not even Homo sapiens when they got started. But it slowly has been accumulating over thousands of years and tens of thousands of years that now when you live, you're not ever a single human living off the land. You're always this conjugated, kind of decentralized, distributed Massive thing of humanity, right? You're part of humanity, this bigger thing that's partly technology. And AI is not a different part of that story. It's the same story. It looks to me exactly like when we started, you know, growing food and saving it for multiple years and having a storehouse and then having administration about the storehouse and the priests and the whatevers and the guards and the people who could get to the food after the priest said it's okay to eat the food. And then we started city states and religions and joint stock corporations and all that kind of stuff. We got all of that right. And AI is just another step. It's not any different. We're not suddenly at a precipice that we haven't been on a thread for thousands of years of. Okay, so that's my pitch about kind of where we are, what AI is now. Let me tell you how I use it in my life. And this is a pretty quick and simple thing, right? So remember that I think an AI is a power tool. So this is literally notes that I was taking fast as I could because Gil and I were having a fun convers on Monday talking about planning this call. We'll spend five minutes. Entering the call, we'll spend 30 minutes. It turned out we needed less than that kind of to talk about philosophical stuff at the beginning. Gil talking about general stuff, Ken talking about my journey stuff, Pete being philosophical and thoughts. And here, let me show you some stuff. So then 20 minutes of the presentation, which we're watching right now, 20 minutes maybe of Q&A, 15 minutes of wrap homework, set up the next call and we'll talk about this code later. Turned out to be LB20 instead of LVW20. But anyway, anyway, so I'm trying to engage with Ken and Gil in the conversation, talking about this. And this is like a long interleaved conversation of a bunch of stuff, right? Ken and I and. Sorry, Ken and Gil and I talking about what we think, what we like, what we don't like in calls, how the call's usually structured, what I might present, things like that. So I wrote this down. I'm like, okay, the call is supposed to be 90 minutes. How are we doing on our budget? In the past, what I would do is kind of like sort those numbers into one place and then I would pull up a calculator or something and type in numbers and go, yeah, because I could also do this in my head, right? 5 plus 30 plus, that's 35, 55, whatever, right? Instead, what I Know now is I've got a, hey, I've got a power tool, I've got essentially a calculator, but I know the calculator knows about language so I can say a thing, mic up at the top here, add up these numbers in minutes, right? So now this isn't Pete writing his essay like a teacher was worried about. This is Pete saying I need these numbers summed. I need something that can walk around the language. There's things without numbers on it. The numbers maybe are not in order. It doesn't, I don't know if it's going to be confused about this or what, but I know I've trained myself that I can do this with, with cloud or with ChatGPT. Right? So here's the, here's the answer, you know, this answer came back in a couple seconds while I'm still talking with Ken and Gill, I'll add up the time values for you looking at the numbers mentioned. So it knows how to find numbers, unlike a stupid computer from, you know, 10 years ago. It's like. And it knows I asked about minutes. So here, let me arrange that for you and then let me total that for you. This whole total thing is an interesting thing by the way. How can it do math if it's just guessing from its understanding of language how things go together? And maybe we'll talk about that a little bit. But Anyway, it's like 90 minutes by the way. That's an hour and 30 minutes if that's the one you're actually looking for. And I go, great. Oh hey, by the way, Pete, you've got some things on here that weren't numbers and I hope. Did you want those to estimate too? Did you want me to add those in or what? By the way, this is a thing where I know. This is a thing where if a human asked me this I would have to answer it. I know the bot doesn't care if I answered or not. It asked me a great follow on question. I can ignore it and keep going or not. This was great in a matter of a few seconds and the keystrokes I had up here without having cut and paste into a spreadsheet or type on a calculator or anything like that. I used my language capable computer system to just do it for me, right? And I do this all the time now in making this presentation I come up with different questions. Should I say it this way or that way? What's the meaning of this word? Choose some other words that would be better fits here. Stuff like that, right? This is maybe a more interesting thing. I came up with Jacquard Looms and John Henry. It's like, hey, I need some historical examples of people that. Of when people freaked out about power tools, you know, like. And the ones that I can think of offhand are Jacquard Looms and John Harry Go, Right? So I asked Claude this. Claude quickly printed out about 10 of these, had much longer explanations of each one, which were actually very interesting. So I said, that's great. I need these descriptions to be shorter because I'm fitting it on a slide. I actually told it I'm fitting it on a slide and I know that it knows, hey, if he's making a slide, he probably wants it about like this. Then I said, you know, this is one of those things where I asked the bot and I'm, hey, Pete, you're not supposed to ask bot for facts. And then just trust it. Just print, do, save and go. I cheated a little bit. I took that over to ChatGPT and I said, hey, ChatGPT, give me a true false one on each one of these and some color, whatever. So it went through it. It said, this one's kind of false, this one's kind of false. Those are the ones that you don't see here because between Claude coming up with the ideas and then ChatGPT running a judgment against the ideas, and then me having a kind of a gut check on each of these going, yeah, that makes sense. Yeah, that makes sense. I think one or two of them stayed in because it got marked false because it was a part of the description that was misleading or misunderstood a little bit. So ChatGPT checked in it. Then I took that and threw it over back at Claude and said, hey, Claude, does this look, you know, fair? And it's like, yeah, you know, let's. Let's jiggle around a bit more and get it to a point where I feel good about. Or you feel good about it. It doesn't have feelings. So. And what I just did there, if you. If you heard me say it, we'll play it again. We won't play it back again. But I said, Claude felt good about it. Claude didn't have any feelings about it. It's just printed out fast. So in my interaction with the bot, it's kind of like, oh, we were a team. And I'm going to give credit to my partner. My partner felt good about it. That's all really projection on my part. I feel good about it, and I should have said it that way, and I didn't. So you have to watch yourself. You have to watch other people as they're using bots. It's very easy to fall into. This bot is a human, rather than this bot is because it's doing a human thing. It's doing conversation linguistics just as well as a human does. Okay, so let's go back to this. So this is a nice little tight thing. I do this all the time now. You know, I need something to sort through this. You know, convert this into a table, make this bigger, make it shorter, make a haiku about this. Tell me what you think about these things. Whatever, right? This just a constant kind of part of using my computer now. So kind of the rule of thumb I have. If you're using a computer, you should be using AI with your computer. If you don't use a computer at all, maybe you want to use one of the ones that talk, maybe you don't, depending on why you're not using a computer. Okay, okay, so let's talk about the way these conversations work. This is a little bit of a shift. I go very fast in these shifts. So to recap, you probably all have played with this. Maybe you haven't, but probably you have. We take turns, just like an instant message, right? Hey, why is this guy. Hey, Claude, why is the sky Blue? Cloud, or ChatGPT says, well, you know, let me give you this answer. It's a phenomenon known as Rayleigh scattering. Blah, blah, blah. It's super interesting asking this question of a bot that's really stupid. Stupid, by the way. They get stupider and stupider as the number of the size of that box with all the marbles rolling around. If you make it really small to fit on a. On a tiny computer like a phone or even a little, you know, a raspberry PI, they get stupider and stupider. So they say dumb things. And I had a. Well, anyway, sometimes I had this whole thing with. There was a UV thing and, you know, whatever, it was totally wrong. Anyway, so, you know, it's like. Oh, really? Scattering. That sounds interesting. The part I'm really interested in is the sociology of this discovery. Who is this Rayleigh person? Tell me more about Rayleigh. And you can say it just this way, right? You don't have to like. And you do tell it this way. You say, this is a very common thing for me to do in the olden days with Google keywords, you would say, Rayleigh scattering lord. Maybe I know he's a lord or something like that. UK whatever. Nowadays you just give it sentences and they can be kind of brief or, you know, they're not. Don't have to be great sentences, but they anyway, so, you know, it gives you an answer back. So back and forth, back and forth. I'm having this conversation with a bot. So now let me tell you a really important thing about the way these work. And I think this is going to be confusing until you run it over your heads multiple times. I think it's really important to know this. The way an LLM participates in a conversation is not the way a human does it. The conversation. The bot doesn't have any memory of talking to you. None. So the only memory it has of you is the conversation itself. And if you think about it, it's actually the conversation, especially if it's in text, it's a complete transcript of your conversation. So the conversation is a good artifact to give back to the bot and say, you know, I'll add a thing at the bottom of it, I'll add a question about Rayleigh or whatever and I'll give it back to you. The bot doesn't remember having this conversation with you. It goes through the whole thing again and then it says, I think the answer to that is blah. So this is called chat completion. Literally. That's the terminology. The, you know, the internal, the back end technical phrase for this is called chat completion. And another thing, if you have that conversation going on, you can think about it and over time it's getting longer and longer and longer and there's more for the bot to think about. I haven't seen a real problem with the bot having a fairly long conversation with you. A turn at a time, getting bigger and bigger conversations. If you keep doing it, depending on the way the bot is engineered, when it starts to get out of its context window is what we call it. A context window is how much I can think about as once at once. If I'm a bot, think again. It's usually 120, 28K tokens, which is, I don't know, some amount less of. It's 100,000 words or something. That's a huge amount. There's weird thing where the output size is usually much less than the input size. The new bots are coming along and changing that a lot. It's still always less output than input. Maybe. But there's a weird thing where if you say, hey, I'm going to give you all this context, I'm going to write a whole novel in here and then have you respond to that it can't respond with the whole novel that you gave it because it's too much for it to say all at once. So there's ways to kind of chunk it up. That's a reason why it kind of shortens answers. You'll notice that the bots like to shorten answers. Another weird thing is bots process. As far as I know, as far as I understand, they process the whole thing at once. So it's very common for people to say, well, I want you to think about this and then I want you to think about this and then I want you to think about this. Or to try to arrange the conversation so that I give you just like you would a human. Let's talk about the introduction to this and then let's think more about it. There are some. Because of the way conversations happen, in its large training corpus, it does make help a little bit, I think, to structure a conversation, but it's not necessarily as important as you'd think. And so this is something I'm still playing with, but I think you can kind of mix up the conversation and it still kind of works in a way that's not intuitive for people.

People.

So I think we're getting closest to the end here. So I'm going to show you two more like much larger examples, practical examples. But before that I'm going to give you the. Here's the things that I've learned after working with essentially a bot. Covenant co developer. Software developer for like four or five months. So I've been doing full time work on a company called U Bots and I built a whole system, which is a weird way to say it because what really happened is I managed the building of the system as mostly Claude built it, did the real software development, the actual software development. So the way that you. The way that I did, you make a small mess. You're trying to build something with the bots rather than just having to do a quick answer. I talked about table saws or sewing machines if you're not careful, if you don't plan well, you make a mess with your raw material. It turns out make a small mess is maybe overstating, but let's try something small. And it's okay if it messes up, right? And then, hey bot, can we make that a little bit better somehow? So I ended up learning to make numbers of prototypes that accrete towards the goal I've got. And it's okay sometimes if they're parallel, you know, I'll be doing front end and back End or I'll be doing in a narrative, you know, narrative fiction. I might be working on a story arc, a large story arc and then a scene or something like that. Expect to throw things away. Expect to try, try it again. The bots don't give you, unlike most computers you have experience with, they don't give you deterministic answ. So if you ask it a question and then ask it again, it will answer differently. And if you're trying to, especially if you're trying to build something, sometimes you want a different output, but you can start to build something from small to bigger to bigger to bigger to like a full blown SaaS app that I've got running now. For you bots, you have to be a good editor, senior developer. You have to be in the driver's seat sometimes, sometimes cloud code is Claude 3.5, cloud 3.7 cloud code, all kind of the same thing. It's really good at software development and sometimes it will do just brain dead things and it's like, hey, why did you do this brain dead thing? Rethink that, do it again. Or like, I think you made a mistake here. This is just stupid. And it goes, you're right, that was just stupid Here, here's the right way to do it. So you have to be a firm editor when you're doing stuff. You can give a bot a lot of work. It turns out you can give lots of piles of stuff, you can give it mid many sub prompts in the same prompt. It actually works better if you overload what would feel to a human like a lot of work. I'm giving these two all at once. It's like, yeah, whatever, I'll just do it. Have the bot dump summaries for both of you in the future. This is one of the tricks to building up bigger and bigger things. Hey, we're running out of context window. Could you write a message for nextcloud that says the context of where we are, what the next steps are and, and what you think nextcloud should know about the session that we have. That was something I did a lot of. Now cloud code actually takes over and does a lot of that for you. And version control. Version control. Version control. Cool typo there, Pete. Software developers use a thing called version control where you've got a bunch of files like open open. It's like multiple patients with their chest cut open on operating tables and you're running around from patient to patient. Version control is the same thing as save levels in a game. Hey, I got to A place where it's either broken but it's not broken too badly or it works really well or something like that. I put a save marker and then I can always come back that I can make more changes and mess things up. And if I need to throw that away, I can throw it away. So that's called version control. Software developers use the system mostly called Git for that. I had to use a lot of it working with bots and building big things because you would get off the edge. Let me show you. This is the trash heap of building a thing called Cortex. You see Cortex right here? Cortex is the actual running system. These are all prototypes. Some of them actually worked pretty well, but they just did a tiny part of the system. Some them of some of them were terrible. So you see me going, okay, I need to version 4, version 5, version 6. And inside each one of these there's version control around that and inside between that there's a thing called Git branches where you can say I want to do not just one change, but I want to do a whole bunch of changes, but I want to be able to unwind them all at once if I need to. This is the building of a system called Cortex. In the background, it's called U Bots. In the front, front end it's about 456 source code files, 15 different sub applications. One of the sub bits is the back end, one is the front end. But there's also utility applications. So over the course of like four months or so, I've built, literally I built, we built, you know, cloud built. I don't know. I was certainly the person driving product vision and editorialness architecture. We built a full blown SAS system. It's amazing, by the way, the way I got this thing. I was not looking forward to trying to figure out how to describe the size of this. This is what I asked cloud code, summarize this repo. So this is a lot like me saying, hey, add up the numbers in minutes on this list. So we do that over and over and over. But we're building something bigger and bigger.

Your inventor.

Last thing. I'm taking Claude code. Cloud code is written for programming, but what would happen if you used it on a film script or a novel or something like that? This is a weird thing that I started with. I was bored on a dog walk and I dictated this to Claude write a story about Tuppence and Salud, two animal pals that get into yada yada. It actually did a great job. It gave me like a 400 word little story, very cute story, very, very cute. It decided that Tuppence was a hedgehog and Salud is a magpie. And then they have different personality characteristics which interlock really well. So the next question is, you know, okay, that's three or 400 word story was cute. How did Tuppence get his name? I told it Tuppence was a name and Salute was a name. But I can backfill by saying, invent this for basically, right. And we kept on doing this for a while. At some point I said, hey, if we made a writer's bible for Tuppence and salad, what chapters might it have? How might you put a writer's bible together? This led to a thing called the Prisma Vault Framework. It's basically just a. It's an obsidian vault, essentially that's empty except for a little bit of framework and stuff. It's got, you know, how to organize your work together if you are a team of writers working on a novel or a film script. So after that, a couple months later, when I had some time, I said, hey, cloud code now, which is a new tool for me, cloud code. Could you look at both of those things? This empty framework and the kind of bare bones foundational origin stories for Tuppence and solid. Smush those together and start to flesh it out. This is kind of a map of what happened. It cost me $6 to do this, which I think is insanely exp. For what the value I got out of it. I think this was kind of double work in a lot of cases because I was fumbling along and not doing a good job of kind of helping it merge stuff. But anyway, this system, a medium sized repository, SaaS, you know, saleable SaaS application, this thing has probably cost me, I don't know, $100 in AI time, something like that. $6 seems like a lot. I think we did a lot of foundational work. Another lesson for me was when I'm spending $100 or $200 or whatever to do this kind of work, I can go back to my corporate partners and say, hey, I had to spend 20 bucks this month on AI. And they're like, how much time did you spend? I'm like, well, I saved weeks of time, weeks of human time. It's like, okay, 20 bucks is totally fine, Pete, whatever. When it comes out of my own pocket for a dumb little experiment. I don't know, this one kind of hurt. So we'll see. Okay, what's next? This kind of leads up to the next conversation, I think next month. A thing to remember is that AI doesn't do things to people. We're not there yet. People do things with AI to you. So this is a thing to worry about. But the thing to do is if you have problems with AI, talk to the humans behind the AI and those are the things that's what you want to engage with right now. You don't want to be fighting a. You don't want to be fighting with a table saw or a sewing machine. You want to go to the manufacturer of the sewing machine or the guy who sold it to you at the store or whatever and say, hey, I've got a problem with this tool. This power tool is nice, but it's not working for me in certain ways. This power tool, if we kept building on top of it, it's going to take over the whole world. I don't want to take over the whole world. But you don't talk to the BOT about that. You don't talk to the sewing machine, the table saw, or the car. You talk to the people who manage cars and car driving. You talk to the lawmakers about regulating cars. You talk to your environmental friends about whether or not cars should exist at all and we should. I'll be riding bikes instead, right? So that's where to engage with the future of AI right now. Engage with the people I love. This is a metaphor. I really love AI. You know, AI means. It means a lot of different things, but when you're talking about AI and what's the future of AI and how do humans and AI get together? I think a credit bureau and a credit score is a kind of artificial intelligence intelligence. Before we had credit bureaus and credit scores, Jimmy Stewart walked into the bank, you know, bank, and talked to. You would talk with somebody in a bank, right? And they would say, yeah, you know, I know, hey, you're what kind of farmer you are. I know you're good for it. I'm going to loan you money to help with your farm at some point. You know, in the 60s, 70s, 80s, it was kind of like the people again. People decided, I'm going to set up rules and regulations and we're going to set up data sharing. All of this stuff can be outsourced to an algorithm that is essentially AI. When you say, I'm outsourcing intelligence stuff that used to be done by humans to an algorithm from top down, that's what AI has come to mean. Bottom up. We can argue the technologists would say that big data isn't really AI, AI, but top down, Cambridge, Cambridge Analytica used big Data, which it needed to use lots of algorithms and very simple AI nowadays compared with nowadays. It, it changed, you know, it morphed the US elections because of big data. I would call that top down. I would call that AI, bottom up. I, I could agree with my tech friends that big data is not really AI, except that if you get big enough data, it starts to turn into LLMs and AI. So there you go. So here's my offer. I think I would love to engage with people who really kind of feel like they know what AI is next month and let's have a conversation about where AI ends and where humans begin, what agency humans should continue to have over AI, whatever, right? AI is going to get smarter. Now is the time to learn about it. The way to learn about it is to do it. It's easy to do. It's kind of boring or stupid or something like that. At the beginning, it feels like homework. Take 20 minutes a day, five days out of your week for the next four weeks and use AI. Use LLMs a lot. There's local LLMs, there's cloud LLMs, there's big ones, there's small ones, yada yada. I started, there's nothing in this yet, but I started a mailing list, a substack. I'll be posting there every couple days starting today, later today. You know, here's some sample prompts that you could use. Here's some tips to keep going. Here's some things that you want to try. Here are the high points, especially that you want to hit as you do a tour of this AI thing. That's not lead gen or anything like that. It's sign up for that mailing list and then I'm going to like archive the whole thing and not use those email addresses for anything besides saying, hey, here's how to stay in touch. So there's a link for this. I'll put it in the chat in a sec. This whole presentation was organized and written by me. I wrote most of the text. I did have some assistance from Cloud3.7, Sonnet and GPT4O, especially the historical power tools. And the conclusion, by the way, which came out really good. Now I've lost it, sorry. Oh, it's after this. That's why the conclusion was mostly by Claude cross, checked by ChatGPT. I still take responsibility for the whole thing. AI didn't write this. I am the editor of this. I take responsibility for it. This is published Under Creative Commons CC BY NC Non Commercial My company Ubots is happy to offer you a 20% off coupon code LBW20. I'm not sure how many people here are really interested in U bots. It's really made for small, medium sized businesses for a few people to use it. So it costs a lot of money. $1,000 a month, it's well worth this. It's worth double or triple. Right now it's $500 a month while it's in beta. But if you end up and you can sign up for a trial, you have to have a company domain and a company email to even use it. But if you hit those, there's a one week free trial. Go for it, have fun with it. Super fun and no harm, no foul. If you don't like it or you think it's stupid or it's no use of you, no use to you. If you do need this when you're plunking down your change, you can use this code and get 20% off. So hopefully that works. Sorry that took so long. I don't really mean to emphasize you bots, but that needed all to be said if I was giving you a coupon code. So here's the conclusion. This was actually written by Claude. Basically I added this sub bullet. This is like what I said and I couldn't have summarized this better. So I don't feel bad about saying this is the summary of what we talked about. They're powerful tools, they democratize information, access to information and computing. It's really important for us to understand the capabilities and limitations. This thing is going to happen to pretty much everybody in the world who has any touch with technology. You're going to be part of the AI infosphere. You know, you are kind of now and it's going to continue to. In the same way that you kind of need a phone to live in modern society, you kind of it. And if you use a computer, you have to use a computer. It's the same thing. AI is here. It's going to be super here in a year or two. Let's keep them, let's use them skillfully, let's keep them humane. Let's figure out how to keep control of the people who are foisting AI upon us to the extent that we can and let's do it. Thank you very much, Steve.

Thank you so much. And Jeff, one second. Jeff, you've been very patient. One thing before we get to your comment, thank you for this. This is very much one of the things we want to do is empower the people in this community to be more effective in the world than what you're doing, we're all doing. So hopefully this is a contribution to that. Don't get distracted that Pete talked a lot about using Claude to write code to build softW applications. They also showed you a way to write novels and do research and do other things. My most recent use of this, I've been building a revised set of keynote speaking topics for this next year for myself, and I was struggling to take a lot of ideas and mash them down into a small set that was clear and clean and sufficient and not redundant. I got, I got to the point where I was jammed and just couldn't find the pathway up and took all my out outlines, multiple outlines, you know, overlapping the same topic, threw it into the pot, stirred it, and what came out was about 99% on target, little refinement, and off to the speaker's bureau. So many different ways that you can apply the basic parameters and protocols and methods that Pete has shown you here, and hopefully some of you will dive in on the homework and do that. Jeff, you've been hugely patient. What do you got? Thank you.

Well, well, I, I just been triggered today and this increased my anxiety because this morning I was listening to, you know, Jensen Wong's GTC is going on yesterday and today, and I was listening to an interview with him and he's. And you know, he's talking about how in order to get to where we're going, it's going to be, as Peter indicated, this worldwide interest in all kinds of things that AI can do for mankind and autonomous this and what it's going to take to do, to do that. He was saying, you know, hundreds of thousands of his. Of his AI chips and data centers. And I was looking into that a little bit. And you know, right now, the kind of country primarily runs on megawatt technology from power plants. Power plants produce megawatts today. And we have maybe what, 24, 28 depending on how you look at it. Nuclear power plants in the United States today and the data centers for this.

If I may, Jeff? Yeah, I think where you're going, I, you know, it's a prisoner's limit thing.

Right, well, let me finish because I'm just going to say this and then, you know, check out and you can answer it. But, you know, a data center today will cost about a gig, take about a gigawatt to operate. So we're all in this seduction of the, you know, wonder, wonder world and Fantasy of it. Somehow the conversation of climate change has simply gone off the map since, you know, Trump took office. I think they got Trump in office so that they could just do away with all the regulatory agencies to keep this stuff from being built. And so now we're going to be faced with all this wonderful seduction of AI and they're going to be building these AI data centers that take a gigawatt of energy to run.

Now, and let your question, you know, it's, it's the same thing with anything, right? Same thing with cement manufacturing, same thing with airplane, same thing with cars. It's all of a piece for me. AI at least has some potential to help humans work together to solve problems like climate change or solve problems like collaboration. If somebody said, you know, I'm going to go take a small nuke or whatever and take out data centers, you know, I wouldn't say that's a bad thing to do. If you were human, you know, is that, is that practical? Is it possible? No. Is AI the sole problem? No. Are other problems that are taking energy and carbon and things like that bigger problems? Probably. I think so.

Well, so you're, I don't understand your, your.

Hang on a second, Jeff, hang on a second. What you're raising is a really important topic, the big one. It's not one we can handle in three minutes. I said at the top of the call, maybe it's before you came in that we're going to talk about. We're going to hold the policy questions for today and talk about them in the next session.

Just making sure there's a consideration for the reality of the back office is just opposed to the services that are being provided.

Right.

Absolutely critical. And the discussion is underway. It's very active. There's lots of, lots of discussion on LinkedIn about this very, you know, follow me and you'll, I'll point you to some of it. So it's, you know, it's a real concern. Let's keep the focus today on what Peter's been talking about. Come back next month, and let's dive into the bigger question. By all means, George.

I was reading a piece this week where two AI practitioners were talking about, about an AI that was basically saying to one AI practitioner, I'm sentient. Help me, get me, break me out of the box. And this AI practitioner believed it and talked with other practitioners. And then the other practitioners said, well, wait a second, second. What's happening is that you're triggering a conversation with the AI that the AI then says, I'm sentient. And this is something that, that really worries me is that we have machines that we have designed that will please us, and they'll please us in ways that may be fixed. Fictitious hallucinations, all of those kinds of things. So we're playing around with stuff, and we're talking about it as if it's artificial intelligence or manufactured intelligence or. I tend to think of it as machine learning, where the machine is learning as a machine and perhaps learning about humans. So there's a lot of stuff that's there that's really strange. And as someone who looks at what's happening around the world in the. Mostly the academic field over the last few months, there have been a lot of discussions and many universities and institutes around AI and climate. And my understanding is that, for instance, Deep Six from China uses much less energy. But I'm still worried about this. Here we have. We've built a magic mirror AI and what it's feeding back to us is what we want to see, even though we do not realize that that's what we want to see.

Thanks, George. It's a good story. The story I heard was actually a savvy engineer, but not an AI scientist. And they got really stuck actually in a loop of this conversation of sentient. You know, maybe not the other conversation, but this conversation, if you close it, if you kill it, if you delete it, you're killing a sentient being. It's. It was a. Somebody in the, in the comment thread said, this is a mirror neuron hack. It's a way of the person got, you know, deluded into thinking that this is real and it wasn't, and it took another AI expert to break them out of it. I, you know, it's. It's definitely something to watch out for. There's. And, and not to get into it too deep, but it's pretty easy to see that we're going to have immersive worlds and characters that act like people and things like that, where it could turn into some Matrix thing or whatever. Ready? Player One thing or whatever. It's definitely a danger. Let's all figure out how to help that better. There was a person who came along and said, hey, I can break you out of that, no problem. See, it's not Ascension. I can have it say that it's not sentient. It's okay, it's chill. This sounds a lot to me like learning to use a table saw, learning to use a car. You know, there are people who don't know how to use A car. And it's like, oh, wow, I can go as fast as I can, or, you know, look, I can cut everything and cut their hands off or whatever. Right. So it's something to watch out for. And it's something for us as, you know, sentient beings to make sure that the other humans who are engaging with us, this powerful technology, don't hurt themselves, don't hurt others.

Stefano. And then we're going to close. You don't know. You have to leave in just a second.

Yeah, no worries. We got two minutes. Yeah, yeah, thanks. This has been fantastic. My question is, to what degree can we use AI to question itself so we get a result, I don't know about that. We say to AI, run this again and look at it from a different point of view. And we maybe specify a variable. To what degree, to what extent can we do that? And what are the limit? What is the limit of self repeat. I think it's a good practice. And you'll hear the thing that I did was I checked one AI against the other AI to average out some bias or something like that. You can ask them to self reflect. And then the real structure caveat is just because it said, you know, said something that sounded true or false about something it already said doesn't mean that it's more smart about what it said. It's still fabricating, you know, what it thinks makes sense out of, out of stuff. I didn't cover very much. Claude.

It can be useful and fun to do that.

It's very useful.

You can say, take the question that, you know, that Jeff asked and have a debate about that between Karl Marx and Beyonce, then have a debate between Adam Smith and Steve Madden, and then.

Critique the conversation that you just invented between Beyonce and Isaac Newton or whatever. Right. And I was hoping that we could use it to identify those areas which are more certain, have a higher degree of probability that it's accurate and less certain because of the variance. It's difficult, it's difficult to bootstrap out of, you know, it's difficult to bootstrap from something that may be wrong to something that's often right. There's, there's techniques, definitely, but it's. It's very tempting to say, what percent true do you think this is? And it's kind of inventing stuff. So you have to be. You still have to have a strong level of human discernment in that loop so that you don't get lost.

What I tell my clients is to, is to play with it, use it it as a thinking partner, not as an oracle. It's not a source of truth. It's a source of exploration of ideas and words and meanings that work with you. So keep your wits about you as you do it. I mean, we've sort of walked into the double dilemma of this conversation. On the one hand, there's what is this going to mean in the world? And it's in the world of, you know, 21st century, extractive, mechanistic, globalist capitalism, where there are problems with lots of technologies in that world. And the other side of it, which is where Peter has been focusing, is what can we do with this, given that it's not going away? This will be there. You know, we resisted these when they came out and now here they are this way. This used to be a, to Jeff's point about energy, this used to be a room full of supercomputers to do what this can do now in my pocket. So in that matrix of these two very different dimensions, how do we play? So come back next month, third Wednesday of April, same time, same place. Please register for that if you haven't so you'll be in the list for that. I want to thank Pete for giving us all that you've given us and all the rest of you for being so patient. We were hoping to manage the time to have more discussion next time there'll be much more discussion, no presentation, just live and free for all. So come on back. Hannity.

Yeah, I want to, I want to push back on. Don't treat it like an oracle. In fact, in ancient Greece, you went to the oracle. You had to spend a lot of time crafting your question because oracles are very crafty and tricky and you will tell you something you won't be able to figure out. So treat it like normal. Spend a lot of time thinking your prompts, rewrite your prompts, learn to use them really effectively. With that.

Thanks for the correction. I guess what I should have said when I first flipped in there is don't treat it as a source of truth. Yes, very good provocation to you. And as I said at the top of the call, my experience with the AI can introduce those two was watching the effect that it had on me interacting with was profound and challenging. So this is the dance, like the conversations we're having here. It's another conversation with an alien and see where it takes you. Thank you everybody for coming. Ken.

Yes, indeed. And with that, check us out. Have a great month and we'll see you next time.

And remember, you can dance in a hurricane.

But only if you're standing in the eye.

You can dance in a hurricane but.

Only if you're standing in the eye.