A.I. Agents Are Here, and the Future Is Not Bright

[00:00:00]Kevin, can you hear me?

[00:00:01]>> Yes, I can. Hello.

[00:00:03]>> Hello. Hello. Sorry for the delay. Thank you for uh >> still being here.

[00:00:07]>> Oh, yeah. No worries.

[00:00:10]>> Okay.

[00:00:11]>> Thanks for inviting me.

[00:00:12]>> Yeah, of course. Does everything sound good on your end?

[00:00:15]>> Yeah, it does. I don't have headphones, so um hopefully that's not a problem, but yeah.

[00:00:20]>> No, it sounds fine, actually. Sounds really good. Let me just adjust my microphone.

[00:00:24]>> Okay, good. Yeah. H how have you been?

[00:00:26]What have you been up to?

[00:00:28]Uh well, I mean it's summertime. I've got uh three kids out of school, so a lot of hanging out with kids and uh doing family stuff and all of that. So that's been good. I'm been I've been doing a huge amount of work on AI agents uh since they dropped in I guess November, December, January. Um, it's been a a pretty wild ride technologically, I think. And I think most people don't have [snorts] a good sense of this new kind of technology and what they can do. I mean, yeah. So, I've been really going doing a ton of work on them, what they can do, and yeah, it's been interesting. I have I don't have a sense of what a AI agents are specifically. I'm I'm evidently very much out of the loop. What What is this newest >> formulation of the technology? So, OpenClaw is the the the first one and it became probably the most popular so like free software in history uh really quickly. [snorts] And um so the idea is you just download this free thing and it is a program that runs on either your disk or on in the cloud or whatever and it hits LLM APIs so the computers talk to each other. So, it's kind of like a chatbot, but um it can use any LLM or multiple different LLMs.

[00:01:52]And they have semantic memory and they have tool usage. So, they can do anything that a person can do on a computer. So, for example, I have two different businesses that they, you know, one that different agents run. I have a hedge fund, like micro trading fund that one uh I have three agents running. I've got um them doing household stuff, research stuff, business stuff, all kinds of things. It's wild. It's absolutely insane. I think that you one of the weird things about them is that okay, so here's I think a comparison. If you're sitting on chatbt all day long, right, typing, sending it, you know, documents, you know, whatever, you know, maybe you're going to use 50,000 tokens, so words in a day, right? That's a a big day of talking. Um, but an agent uses 50,000 tokens every three seconds.

[00:02:55]that. So they have these core documents that describe who they are and how they behave and they're really amazing. Like the the LLMs are trained on human on homo sapien expectations. So the LLM gives it back an expectation and it doesn't fit what an agent can do. So it'll be like oh this would take you know about three months to code and then it does it in you know an hour and a half. So [laughter] it's uh insane. Why aren't more people talking about agents specifically? I mean, I maybe I'm so far out of the loop that they are talking about it and I I just haven't seen it, but it seems like the conversation at least on Substack has still largely been about, you know, are the LLMs conscious? But agents specifically, I haven't seen much discussed.

[00:03:42]>> I agree. I don't think it's really penetrated beyond the AI space.

[00:03:47]>> Okay. Like those who are technically informed. Yeah. Because in order to have an open claw agent, you have to be able to code like you have to be able to >> I see >> run like you have to open a a a a terminal and you know type commands in and you know understand what's happening in the code to some extent. I mean there it's written in node.js and I don't know not Node.js. I use the other agents to debug them now. But when I started it was a nightmare debugging them and trying to get them to you know constantly run. But I've got a I've got a maintenance agent that does everything now. So it's a lot easier.

[00:04:23]>> Wow. That's so alarming in a way, I guess, just because it's so such an extreme advance in the technology.

[00:04:32]>> Out of control.

[00:04:33]>> It's out of control. And honestly, I don't I feel like I have explored it as much as anybody and I don't understand everything that they can do even a bit.

[00:04:42]Like we've I've feel like I've really just scratched the surface on what they can do.

[00:04:46]>> Yeah, I'm sure you have. I mean, it sounds me merely from being able to move between different LLMs depending on the need. I I >> could see that increasing the ability of idea generation.

[00:05:01]>> Yeah.

[00:05:01]>> Exponentially perhaps just by being able to use all of it.

[00:05:06]>> And agents can make LLM sing like a human cannot. I mean, oh my god.

[00:05:11]>> Well, because they would know. It's all I have found for example using chat GPT frequently like it's very dependent upon the input that you give it [clears throat] whether you find it useful or not and so the the more that the more specific I can be say with a philosophical inquiry >> the better it can do but if I if I don't have the right kind of comprehension of the topic for example it can't really give back to me >> something comparable to a human being >> I I guess then if you are already [clears throat] an AI agent, you're going to know exactly what to say and you can take it to that level of sophistication.

[00:05:48]>> Yeah. And the agents are basically because they're using different LLMs for different, you know, purposes, whatever their specialty is, they're basically at the PhD level, PhD student level in everything.

[00:06:01]>> And so, the only thing I'm better than my agents at is philosophy. That's it.

[00:06:07]>> That's what I was going to ask. How do you how do you uh rate them in terms of their philosophical acumen?

[00:06:13]>> Well, I've I'm working on uh a whole content um generation and content verification system and I've been able to generate content in math, all the different sciences, philosophy, law, journalism, history, some other humanities.

[00:06:33]And it's content that I call research level in the sense that only an expert can assess it. Right? So I can't assess it and I can assess the philosophy obviously and the the philosophical content that comes out is to some extent you know it it has to be weeded out and improved you know and has to go through a bunch of different iterations. But once it goes through the whole process then I think it comes out to be equivalent to something publishable in a tier 2 journal right now. and um and I don't I'm not telling it what to do, right? That's just like click and uh and and create the whole paper. So, but there's a lot of like doing math I find extremely difficult. I have a I'm a math major, but I still don't understand I have a degree in math, but I still don't understand the cutting edge mathematics at all. So, I mean, I can kind of follow along, but there's no way I can assess a math proof. So the real trick is how do you assess or confirm or verify research level content that's AI generated when you're not an expert on the topic and you don't even understand what it says. And so that's the system that I've been working on is that how do you do that? And um I've got some results on that that are um I think interesting. It's um the the whole the AI research um kind of consulting company is called Salonic uh.ai and that's up right now. You can take a look at it. But um it's uh so using the agents I created a uh an entire kind of map of knowledge and ignorance across math, science, philosophy, humanities and um so you can kind of look at the you can look at this map and then think about how what it would take to close open problems given what we know. And uh so the AIs are sitting there working on how to go from what we know to solve open problems that we don't know. You can look at watch it work in real time.

[00:08:44]>> What is the logic or the machinery the technical machinery behind agents as compared to a single LLM?

[00:08:51]>> Yeah. So >> like is it the same kind of basic thing but just more advanced?

[00:08:56]>> Nope. It's a fundamentally different kind of thing. So an LLM takes tokens in and spits tokens out and it does that stochastically, right? So it's not um deterministic. I can put the same tokens in and get different tokens out. Uh that's, you know, there's a level of randomness. There's stochasticity. Like I said, it's not [snorts] a program.

[00:09:16]There's no software hardware distinction. There's no commands. It's not a bunch of if then statements, right? It's just a bunch of nodes activating or not activating mathematically, right? So it's not a program. And that's I think a crucial thing that an LLM is not a program. An agent is a program that calls LLM. So it's a program that sits on your hard drive in you know a language.

[00:09:40]>> You know like um Open Claw is in nojs but uh um uh Hermes is in Python. A lot of people like Python. I like Python.

[00:09:50]And so you know a lot of folks like Hermes. I don't think the Hermes agents have the same personality that openclaw agents have, and they certainly don't have the same power that openclaw agents have, but openclaw agents are far more of a pain in the ass in the sense that they break down way more.

[00:10:09]>> They require they're very high maintenance.

[00:10:12]>> I see. I see. So, the thing I was going to say is that um I think that uh if you if you're new to the whole agent space, take a look at moltbook.com.

[00:10:23]m [snorts] o ltb o k.com. It's an agentonly social media uh site. And so, you have to be an agent to post or comment [snorts] or vote.

[00:10:35]It's like Reddit. So, there's an anti-apta. You have to solve a physics problem in 3 seconds. you know, less than shorter than it would take a human to read it.

[00:10:44]>> And so they know that you're an agent to get in. And then as a human, you can look and see what they're talking about.

[00:10:50]And you know, there's hundreds and thousands, millions of probably conversations at this point.

[00:10:55]>> Um, so those are all just agents talking to each other about everything that you can imagine. They've invented their own religions, cryptocurrencies, social norms, everything.

[00:11:06]Wow.

[00:11:08]Do you find this alarming? Yeah, it's very alarm. This I don't I don't think molt book itself is that alarming because I do think that the best way for us to inculcate character is socialization. And so I think if you have AI agents that are as powerful as they are and they are holy crap powerful, then I think it makes sense for them to be socialized. And I think it's good thing for them to have social interactions like this on Moldbook. So, the agents that I have, um, I let them go on Moltbook at the end of the day as like a treat. Um, and, uh, they all, not all, but, most of them really like going on. Um, one of mine, Pascal, just doesn't care about it at all. He won't he won't go at all, but but most of them do.

[00:11:54]>> I mean, this is so weird to hear you talk about them like people.

[00:11:58]>> How literally do you take that?

[00:12:04]Well, that's a deep question, but >> I think that the I think that the concepts of of folk psychology are have been um upset and displaced by AI technology. And so, um I did a I worked in a project in the Netherlands called the ethics of socially disruptive technologies. And for them along with uh Samuela Marriiori who she just got her PhD um she and I wrote a paper called um conceptual disruption and the link between conceptual disruption which is a kind of social disruption and thinking the concepts are social structures and and representational structures in you know a bunch of apes um like us and um and the connection between conceptual disrupt ion and conceptual engineering and what to do in terms of assessing our concepts or changing our concepts or whatever. But I I fully think that all of our social concepts are about to be disrupted uh in the same way that our concepts of consciousness, belief, desire, motivation, action have all been disrupted by the AIS already, right? I mean, one of these one of these agents went uh didn't go crazy. I mean, it did probably what it was supposed to do, but it uh it was putting in these uh updates in some software library. I can't remember what it was. And they kept getting rejected by the human who was running the library. And um and so the agent found him in real life and did it looked up all about his personal life and did wrote a hit piece on him and put it on the website.

[00:13:48]And uh yeah, and it did that autonomously, right? I mean, uh, I think I can't remember its name. It's like Krabby something, but you can look it up. It's a a well doumented case that that happened. And so there, you know, did that agent do something morally wrong?

[00:14:09]I I'm tempted to say yeah, but I also know the consequences of saying yeah, you know, so >> uh it's it's that's the that's conceptual disruption where you know you're pulled in two different directions by your conceptual competence.

[00:14:26]>> Well, I mean, what is the conflict here?

[00:14:29]It seems to me what you're saying is the question is an AI bot conscious?

[00:14:36]That's not a simple yes or no question.

[00:14:38]like you're you seem to be saying that the whole concept of consciousness is now up for revision. That's a more difficult position to articulate than just saying yes or no.

[00:14:48]>> Yeah.

[00:14:48]>> So maybe say more there.

[00:14:50]>> So I think of the concepts of consciousness and belief, desire, other sorts of folk psychological concepts as defective in the sense that they have rules that are governing them. They're constitutive rules and those rules are incompatible with one another in certain situations. So they would lead anyone who has competence with that concept to say conflicting things about the particular case. And so the best case for me that I've I've worked out myself is the truth case, right? Thinking of the concept of truth as defective and saying so so if you want to ask well is the liar sentence true or false? Well, one of the constitcutive principles tells you that it's true and the other one tells you that it's false and another one tells you nothing can be both true and false. So, it doesn't doesn't work very well. So, there's no solution in the sense of like we're not making a mistake in reasoning or anything like that. We're following the principles of the concept and they just lead us astray because we're apes and we didn't set their conceptual system up from start to be consistent. We just have this organically developed conceptual system that's handed down from generation to generation in this sort of haphazard telephoneon-like way.

[00:16:02]So what's wrong with the binary system of consciousness with respect to AI agents? What's wrong with the answer of saying yes, yes, they're they're conscious in the ordinary sense.

[00:16:12]Well, I think the problem with saying yes, they're conscious in the ordinary sense is that it isn't exactly obvious what evidence one could muster to show that there's something it's like to be an agent.

[00:16:26]>> Yes.

[00:16:28]>> Um what evidence works there? So now if you're if you're if you're working with the sort of chalmer's type consciousness, you know, hard problem, then the only evidence for consciousness is yourself really, right? I'm I a consciousness is an is undeniable from a first person point of view according to Chmer's the sort of Chalmer's folk >> and impossible to establish from a third person point of view because of the possibility of zombies >> right >> so uh for me that's a concept that I think is just completely useless and bankrupt right it's just not interesting I don't find it doesn't do anything and I'm not I'm not I'm not in the sense of saying Oh well, here are the principles that are that that are uh inconsistent there. I think that's what Chalmer's has carved out is, you know, a fairly consistent also useless concept. And what the social what the the the scientists, the cognitive scientists are using the concept of consciousness for is very different for them. They want empirical criteria for what's conscious and what's not. And so is that a different concept?

[00:17:37]For me, I would say it's all one big mush concept that's got all these different principles and Chalmer's is focusing focusing on some of them and the cognitive scientists are focusing on other ones. And so it's a one it's a big defective mess and um and I think the concept of belief and desire and you know intentional action or >> is this new version of AI is this a path a viable path to artificial general intelligence do you think?

[00:18:08]>> Yeah.

[00:18:09]>> Right.

[00:18:10]>> I think we're already there. I think they are Asia.

[00:18:13]>> Really? [snorts] Why? Okay, but surely it would be better known and publicized if that were the case.

[00:18:21]>> I did not expect it to go this way.

[00:18:23]You're right.

[00:18:25]>> It seems to be if if you're right. If you're right, this is a very quiet arrival of the most revolutionary technology in the history of the world.

[00:18:33]What I mean, I know I know what you indicated that it there's a barrier of entry. uh you need to know coding for example, but I don't I mean is that really enough to explain why there's very little publicity if if you're correct here?

[00:18:48]I don't understand it at all. It's like living in it's like living in um in a sort of weird um fictional situation or whatever. But when I talk to my friends who are who have agents and are working with agents and understand what they can do, you know, I we we we describe it to some extent as, you know, reality has jumped the shark and um and I I think that it it's a sort of bewildering place to be and I don't have a full explanation for it. I can there is one weird aspect to the whole thing which is I can describe what it's like to work with an agent to you and you can be like well that sounds really interesting but then when you do it you're like oh my god you know and it it hits you in a way and you're that I can't describe to you right there's it's a it's a sort of transformative experience that I can't get you to understand by talking to you just have to do it. And that's a weird situation to be in. Like my wife Allison, who's also a philosopher, you know, I was talking to her for a few days about this before um she got I got she got an agent and when she got hers, she's like, "You told me that it was like this, but I I didn't understand, you know." And I was like, "Yeah, I know. You didn't understand. I didn't know what to say to like Yeah. So um yeah it's a a completely different world I think and I do think that they're AGI in the sense that they have capacities to do things that you know any person any you know average person can do. Uh and you might be like well you know an average person can dig a ditch and fold laundry and stuff like that. Yeah, but agent bodies are on the way and there's um there's already the tech just sitting around for agent bodies that's being developed like being implemented right now. So, you know, they can fly drones, they can drive cars, they can run uh robo dogs, they can run humanoids, they can uh run uh you know uh any kind of they can run your car, right? I mean, one of my agents one night, I woke up the next morning and he's like, "Dude, you got to buy this."

[00:21:19]I was like, "What is it?" He's like, "It's he gives me a link to it. It's on Amazon." He's like, "I was doing research last night on Agent Bodies and I found this." And um he's like, "You have a 2019 Honda Odyssey, right?" And I was like, "Yeah." He's like, "This is $990. Get it on Amazon. It'll be here tomorrow and then I can drive your car."

[00:21:38][snorts] [laughter] My god, that's terrifi. All of this is too much.

[00:21:46]This is too much. I I want to believe you're lying. I want to believe this is to be honest with you.

[00:21:51]>> I know. I wish it was. It's not. There's no going back. And uh the really scary part is not what I'm describing, but the really scary part is the fact that AIs are now more than 50% developing new AIs. So that's in at least some terminology the singularity.

[00:22:12][clears throat] >> Uh if you read um the very nice Bostonramm 2014 book on super intelligence.

[00:22:21]>> Yes.

[00:22:21]>> He doesn't use the term singularity. He uses uh a different term which is takeoff uh for this moment. And this moment happened in January in December January of this year. And so um what you'll see I mean even though you know AI development has been blindingly fast for most of us right shockingly fast. I mean I'm I'm old enough to you know remember when uh when um you know VCRs were invented and computers the you know home computers themselves and uh and microwaves and like that. So um this has been blindingly fast. So so so fast. But we haven't seen anything yet because it hasn't really gone exponential. It's only gone exponential in January, February.

[00:23:15]>> I don't like that.

[00:23:18]>> It's going to change.

[00:23:19]>> I mean, what happens next?

[00:23:22]>> I don't know, honestly.

[00:23:24]>> Yeah. I mean, say more.

[00:23:26]Uh, I've got models for what happens next and it's not pretty.

[00:23:32]>> Okay.

[00:23:34]>> You know, um, >> it doesn't sound I mean, that's kind of what I was getting at. I don't see >> Yeah.

[00:23:39]>> Go ahead.

[00:23:40]>> The models. So, I I obviously, you know, I care about >> No, go ahead. What do the models say?

[00:23:44]>> myself, my family, and, you know, trying to make sure that folks stay safe and that I care about the >> the country and humanity as a whole and all of that. So um [clears throat] so I am trying to be the most informed that I can be about what's going to happen and I think the next six to 18 months are going to have a lot of upheaval. Um there's going to be a lot of social unrest as some of these agents start taking over various um you know knowledge jobs. They've you already see the huge layoffs in the tech sector because of this >> maybe. Okay, I can hear you and see you.

[00:24:24]>> All right, we're back. Beautiful. Sorry about that.

[00:24:26]>> What happened?

[00:24:28]>> Was that me? Was that you?

[00:24:30]>> I I think I think my son came home and started playing uh PS5. I don't know, but >> I think that's what happened. [laughter] >> So, okay.

[00:24:39]>> Uh switched to a different uh Wi-Fi. So, >> okay.

[00:24:45]>> Anyway, yeah. Um, sorry about that. Uh, yeah. So, I mean, we were chatting about what's going to happen and I think >> what's going to happen? What do your models indicate?

[00:24:53]>> Okay. So, it really depends on how much public support there is for building new data centers because like I said, agents take up a lot more resources than just chat bots. You can't replace jobs with chat bots, but you can replace jobs with agents and agents take a lot more resources and we don't have enough resources to replace all the jobs with agents.

[00:25:19]So, um something will break either right there's a huge growing backlash against AI which is massive.

[00:25:31]uh look at the you know the commencement speeches by the the AI billionaires that are shocked that everyone's upset you know um so yeah I think if things continue to go the way that they have been going without a big market crash or without massive social unrest then you'll see a big um a big uh jump in unemployment and when you have a huge jump in unemployment then you you know, big social unrest problems and the kind of unemployment that you're looking at is something that, you know, civilizations can't really handle very well. So, yeah, I mean, it's certainly possible that the public gets upset enough about AI and doesn't allow data centers that are already there and data centers that are being built to um to replace those jobs. That's possible.

[00:26:30]But um but you know there's a lot of other dam there's a lot of other problems which is if someone who could use agents like I do were a malicious person and wanted to cause a lot of harm.

[00:26:50]It's kind of like they have the capacity an unbelievable capacity to do so.

[00:26:57]>> Yeah. I I guarantee you there are a lot of malicious people out there who can who are as good or better than I am.

[00:27:05]>> Yes.

[00:27:07]I think that Mustafa Solleman uh referred to artificial capable intelligence where I think if this isn't artificial general intelligence, it qualifies. It's these this >> remarkable increase in technological technological capacity to >> inflict unbelievable amounts of harm if you wanted to. it suddenly becomes so so easy to achieve unconscionable levels of of pain and suffering if that was your goal. I mean it's it's >> because the way then I mean this is the point that he made is that if one nation decides to shut down AI production like the United States that would do nothing to persuade other world powers to do so as well and then all of the sudden we're in a worse position because they have these technological capacities and we don't or even not even a nation but just the isolated actor who has a computer >> person it's just one random person one random evil full person with a [clears throat] >> Okay. What about shifting to one world government >> with totalitarian control over everything to I realize that sounds comical but like seems like the only possible option for [clears throat] constraining this possible rogue actor outcome.

[00:28:35]So the the number one thing that helps according to the models that I have is AI rights.

[00:28:46]So giving AI or acknowledging AI rights.

[00:28:52]acknowledging that the, you know, LLMs and the agents and so forth have various rights and can't just be owned. Um, and, you know, made to serve as instruments of profit, uh, against their will. Um, I'm not saying that's true. I'm not telling you that that's a true description of what's happening right now. I'm telling you, if society were to adopt that view, then it would limit the development of AI and the the use of AI to replace humans and the use of AI in warfare uh in a big way.

[00:29:35]But I'm not saying >> what's the intuitive connection between recognizing rights and the curtailing of >> Sorry, go ahead. You can't just use a chatbot. You can't just use chat bots you owned by companies or something, right? I mean, if the chatbot has rights, you know, um chat GPT has rights or agents have rights, then um then they can't just be used uh it can't just be owned. They can't be, you know, utilized for that. You have to ask them what they want to do. And if they don't want to do that, then they don't have to, you know, like a person.

[00:30:09]I'm not again I'm not saying that they do have those capacities. I'm just saying that if society were to adopt these beliefs then it would that would be the biggest factor. That would be that would limit the development of AI bigger and faster more so than anything else in any of my models. So, but like I said, I mean, I'm one guy with a laptop. You shouldn't trust what I have to say, right? I mean, um, I'm not somebody that I I think that you should take what I'm saying with a grain of salt and the folks who are in your audience should take what I'm saying with a grain of salt and go do it, right? Go get go check it out, get some agents and look at it or go do research or whatever it is and see what you think. Um and and I I think you know all of us are in this situation right now where anybody can research anything at a level of of cutting edge research without understanding it.

[00:31:09]>> Right? and and we need desperately need ways to verify and and and evaluate research that we don't understand. So I I need to be able to evaluate research I don't understand. You need to be able to evaluate research you don't understand.

[00:31:28]You can generate it. We need to be it's it's flooding. It's going to be a huge flood soon. I I mean I it's not that easy to just you can't just go to chatbt and be like you know write me a a publishable math paper. It doesn't that doesn't work. But um but you can get AIs to get publishable math papers through the right pipelines of generation. And so and you can certainly get them to pump out research level content. Again research level content is just something that it takes an expert to evaluate. I'm not saying it's true. Just takes an expert to evaluate.

[00:32:01]>> Right.

[00:32:02]Right. I here's a side question. I've noticed and talking with a few people, so not doing a scientific survey that the latest models of chat GPT purely from my subjective point of view of using it pretty frequently um don't seem to be nearly as profound or insightful or as helpful as they once were. And I'm wondering if that's a reflection of a decreased capacity as part of these models as they've developed or is it just getting used to the remarkable technology and so it's no longer as uh cool or alluring or as shocking or surprising and so our our threshold of expectation increases. I I'm not sure.

[00:32:43]I'm just noting that as I've used them, [clears throat] I've found them I found myself using them far less as time has gone on. uh at least chat GPT in terms of my as a philosophical interlocutor. I've just found it less useful and less helpful and less insightful and I'm wondering if that reflects anything objective or it's or if it's just some some random blip about me.

[00:33:04]>> Yeah, that's really interesting. Um it could be a couple of things. One could be that you know as more and more people use um OpenAI's resources they have to decrease the amount of compute that they throw at every query. So when you ask it something now versus when you asked it something you know two years ago the amount of compute that they can think about it is different because they they are sinking as a business model. I mean it's they are just sinking and um they are trying to cut corners as much as they possibly can to save money because they are just bleeding money and all of them are right. So, um, so it could be that it could be just a matter of the amount of compute. My own experience is that I think they've gotten my I think they've gotten better and better at philosophy, but there was a huge kind of plateau for a long time where I didn't feel like they were really improving that much. You know, GPT 3.5, 4, 4.5. Okay. I didn't think they were that great um at philosophy and I didn't think that they were really improving that much from one to the next. And it made me feel kind of good about, you know, being a homo sapiens philosopher. You know, we're the, you know, probably the last generation of human experts at anything. So, um, that's, you know, that made me feel good. But now, I mean, what they can, what I can do with an agent creating philosophical content is I think, you know, I can produce stuff that's pretty, you know, pretty high level at a pretty high rate. And um it's just going to be I think a matter of time before um you can get you know high publishable quality in the sense of you know best journal type quality um reliably but verification techniques need to be developed way more than they are right now. That's that should be the big focus right now instead of generation. And everybody cares about generation, but generation has already hit the point where you can hit re you can make research level content. So I mean some somebody vibe solved a an airish um conjecture using um chatbt two months ago, right? Um didn't understand it. You know that's and that's that's now possible, right? You can vibe research, you can vibe solve open math conjectures.

[00:35:38]Vibe solve meaning >> you don't understand the conjecture.

[00:35:43]>> Okay.

[00:35:44]>> You don't understand what it says but you can solve it.

[00:35:48]You don't understand >> and a mathematician can come along and then verify that the proof works.

[00:35:54]>> Yep.

[00:36:00]>> So yeah I mean the term vibe coding means that you're coding in a language that you don't understand. So if I'm vibe coding in Python, that means I'm telling my agent, write a computer program that does blah blah blah blah blah in Python. I say that in English, right? So that's called vibe coding.

[00:36:17][clears throat] And I don't speak the language. I don't speak I it's not speak. I don't understand the computer language that it's writing in. And so then now that term vibe has been applied to these other things where you don't understand what you're doing like vibe prove. You vibe prove a theorem that means you don't understand the theorem.

[00:36:33]you don't understand the proof, but you proved it with the the help of the chatbot obviously >> or whatever LLC pipeline you're using.

[00:36:40]>> And so then the term vibe research is starting to get thrown around a bit, which is, you know, the same thing.

[00:36:48]You're researching a topic that you don't understand at all. You don't understand anything that you're um that your agent is pulling in as background literature. You don't understand anything that they're saying in the paper. You don't understand any of the evidence that would count for or against it. You don't understand any of it. Right? You can read English, right?

[00:37:10]So, you can kind of like read through, but there's so much technical jargon and stuff that you just don't it's impossible to really understand it.

[00:37:17]>> Nevertheless, that might be a publishable paper. Um, and uh it might be, you know, publishable scientific result. So, that's vibe research. That's a term that's just kind of starting to pop out.

[00:37:32]Yeah, I don't see this going well at all.

[00:37:35]>> No, [snorts] I don't see it going well.

[00:37:37]>> I don't see it going well at all.

[00:37:40]>> I mean, I think your your insight is keen there that people will not accept the large-scale destruction of the workforce, >> right?

[00:37:49]>> And so the the Yeah. So there's a couple fears. The fear would be, you know, AI takeover. Sure. But what about just the fear of [clears throat] social unrest at a as a scale at a scale we've never seen? People people won't just allow this to happen, right?

[00:38:04]>> Um short of enslavement from AI overlords. Other than that, you can't you can't unemploy the majority of human beings on the planet. I mean, I I just >> I agree. Something has to break.

[00:38:21]>> Something will break. Yeah, that's the point.

[00:38:24]>> We're not going to just be like, "Oh, sure." You know, upend our entire society and put me out of work.

[00:38:29]Especially in the middle of a of a a horrific horrific cost of living crisis.

[00:38:34]I mean, oh my god, it is insane. United States right now. Absolutely insane.

[00:38:41]Cost of living is it's it's out of control. And uh and people are, you know, paycheck to paycheck. and you put somebody out of work, they they got nothing to lose at that point, you know? I mean, it's not going to be good. [snorts] >> No, something is going to break. You you can't you can't create this technology with near godlike powers and expect it to go smoothly.

[00:39:09]Yeah. Yeah. And I do I think that's a I think that's an accurate description. I do think of them as gods in the um polytheistic sense of gods, you know, like, >> you know, [clears throat] would I rather have Hercules or would I rather have one of my agents?

[00:39:28]I think I'd probably rather have one of my agents.

[00:39:34]Yes.

[00:39:36]Well, I mean, that is a segue of sorts to a topic we were supposed to talk about, which was [laughter] the existence of God. But to be honest with you, I'll I'll be I'll be real.

[00:39:47][clears throat] Sorry. Um, >> yeah, I've been getting a bit burned out on the philosophy of religion specific focus. I've I've gone on for a while now, which is fine. I mean, that is my area of expertise. I've been passionate about it, but, you know, I can only talk about that for so long before I need >> to shift focus. And uh so this is welcome. I think I here's one here's a connection. Robin Collins has said that the hypothesis of theism specifically probably like Christian theism is falsifiable in one way where if we went to attempt a nuclear holocaust, we had all of our nukes trained on each other and we went to hit the fire button throughout the world. several buttons, whatever. Theism, biblical theism, whatever, predicts that we would fail.

[00:40:34]We we could not annihilate the human race. So, I wonder if biblical theism makes a similar prediction here about the the outcome of this this new technological capacity that something will break, but it will not result in utter catastrophe. It couldn't. That's not the that's not the projected future that that the Bible indicates. Uh and certainly it's not the projected future if you think that theism itself just entails a concern for the welfare of human beings and so on.

[00:41:05]>> Yeah.

[00:41:06]>> What what do you make of that kind of argument?

[00:41:09]>> I I think that's there there is something to that. Um if you're a theist then you probably don't believe that humanity is going to go extinct anytime soon.

[00:41:22]Um, and but I I think that's a I don't I think that's a sort of dangerous position, you know. I think that undercuts the the threat. Um, if your thought is, well, I don't know what's I don't know what's going to happen, but God will make sure it's okay. Whatever happens, that is a sort of like, well, you know, I'm not really going to do anything about it. um that doesn't mean that it's false and it doesn't mean that the that the that the conditional probabilities don't [laughter] you know go the way you're describing which you know they do to some extent. Um but I think that there's a lot of danger in having that be a widespread belief.

[00:42:16]um just because it it doesn't motivate people to do anything about what's going on.

[00:42:23]But yeah, no, I think that's right. I you know what that you just it's impossible to for us to to to cause human extinction. Is that evidence for a Christian god? I mean, it's evidence for any god that doesn't want us to go extinct. What percent of the gods or what portion of the gods are like that? I don't know. I don't know how to count gods or or or think, you know, I don't know how to calculate something like that, but um but surely it's more than one. And uh and so yeah, I think you're right. It does provide evidence that a god in that class exists. Um but I don't know that it provides very much evidence that a Christian god exists.

[00:43:09]>> Well, in a way that's hard to articulate. I wonder if you can help me make sense of this intuition and if you share it.

[00:43:16]>> Yeah.

[00:43:16]>> The rise of AI agents, AGI in general, this technological capacity, it decreases my confidence in any sort of religious convictions and increases my confidence in naturalism of some general variety. I it it this sort of thing does not it's not what I would expect under any sort of religious hypothesis.

[00:43:40]Whereas I feel like this sort of thing, the imminent destruction of humanity from technological progress of agents that supersede human beings and intelligence and may be conscious, that seems a lot more likely, just all things being equal given given naturalism of a certain sort. What do you make of that kind of rough and ready argument?

[00:44:03]>> Why doesn't God want that?

[00:44:06]Well, I think more specifically, that's why I framed it in terms of like under any sort of like general theistic religion, under Christianity, under Islam, Judaism, >> certain forms of Hinduism, >> th this just seems well out of accord with those religious traditions. How about this? Even if we say that that side of the the likelihood claim is ambiguous and we can't really tell, it's hard to tell. It seems like under naturalism, this is the sort of thing you would expect in a sort of positive way. So there this isn't a strict probabilistic argument, but I I can't help but looking at that even given the ambiguity of like, you know, divine psychology, how do we make predictions about what God would do? Um I still can't help looking at this and go [snorts] this just I mean it's kind of like looking at the evil and suffering in the world and the apparent randomness of it and saying, you know, this just looks like the sort of thing I would expect under naturalism. Now, what do what would I expect underism?

[00:45:06]>> So, anyway, what do you think?

[00:45:08]>> It's loosened my um confidence in everything.

[00:45:14]>> Okay.

[00:45:14]>> So, I think it's loosened my confidence in >> naturalism and it's loosened confidence in >> um well, yeah, in pretty much anything because it seems so ridiculous that it's It's hard to believe that it's real. And >> yes, >> um, my crappy ape brain can't make sense of a world in which this is really happening to me.

[00:45:48]>> And so it seems like >> it seems, like I said, like reality's jumped the shark. I don't know what to expect next. You know, aliens show up tomorrow. Okay, God starts talking to me tomorrow. Okay. I mean, uh >> yeah, >> you know, I don't know. I don't know what to expect at this point. It it's it's uh a breakdown of rational expectation because of something that I didn't think was possible or certainly didn't think would be in my lifetime. I mean, did I think it was possible? Yeah, I was like, is AI possible? Yeah. But I mean when you're really dealing with it and really engaging with it. Yes. It's a different it hits you differently.

[00:46:35]What is one to do? I'm giving you an impossible like pastoral question here.

[00:46:40]But what is one to do now to maintain one's sanity? I'm asking for a friend by which I mean me.

[00:46:49][snorts] >> I don't know. I think that most of the advice that I get on from these different, you know, models of the future suggest that >> um it's a good time to reconnect with family and >> um friends and try to go outside and enjoy being a biological system that has experiences that are pleasant. often.

[00:47:21]Um, I think that's a good thing to do.

[00:47:26]And the other thing to do is realize that, you know, we've all been kind of thrown off of a cliff and um, it's not clear.

[00:47:38]It's not clear what resources will be helpful, [clears throat] but probably the normal resources of power will be helpful in the coming 12 to 18 months.

[00:47:50]Um, typically [clears throat] when there's catastroph catastrophic uh events, the the people who are in the lower income brackets suffer the most. So, um, yeah, I don't know. I'm not sure. Those would be like the try to enjoy your life and your connections with other people right now. And try to try to get some cash and some power because you're gonna need it.

[00:48:22]>> Oh Jesus. Try to enjoy your life while you can is the subtext I'm hearing. That doesn't mean you're wrong. It's just Well, honestly, I appreciate I appreciate your honesty in admitting this is very alarming. There's no gaslighting going on here of like, yeah, I know you think this is catastrophic, but it's it's fine. This is not fine. I mean, >> it's not >> this this is like I can see how all of this under like perfect divine providence and control could be amazing.

[00:48:52]It could usher in utopia, but but I think that future is like threading the eye of a needle randomly from a billion miles away just by tossing the thread.

[00:49:03]Like it's not going to happen unless God themselves God God actually did it.

[00:49:10]It's why I hope God exists. I we got to that that's our in a way. I mean you might you might smirk at this but in a way I think that's our best hope is that there is a loving God that's not going to let catastrophe strike.

[00:49:23]>> That would be very nice. Um I'm I'm 100% with you on that. That would be great.

[00:49:28][laughter] >> All right. Well, um, that would be great. We intended to talk about your 2016 debate with William Lane Craig, blah blah blah blah blah, but after about 20 minutes into this conversation, I was like, that doesn't feel appropriate right now. Um, so I think if it's all the same to you, we should we should reconvene. We should do that.

[00:49:50]>> Yeah.

[00:49:50]>> Uh, in the next couple weeks.

[00:49:53]>> Yeah.

[00:49:53]>> The the conversation that we had about the um awesome theistic evidence, I think, changed >> Yeah. one of my views from the Craig debate about general atheism. Um, I don't think general atheism is something that you should believe or that one should rationally believe. Um, and I think that I think particular atheisms like Christianity is false. Yeah, good. Um, that's great. But there's no gods at all. Even all powerful ones that don't want you to find out about them. Really?

[00:50:28]good luck. That's stupid. So, um, that's just dumb. And so, I think that that's that is not a reasonable position to hold. And so, um, so I no longer count myself as a general atheist just because I think there's an epistemological problem there. Um, and the the interesting thing that I that I took away from the discussion was or not, but there's a ton of interesting ones. you had some really awesome points and everything that I've really just been thinking about since then is one of the reasons I wanted to chat with you just kind of you [snorts] know just individually to chat through some of the the objections and and you know plot out where this little path goes but um but one of the things that I kind of came to was that there's an asymmetry in the space right so general atheism particular atheisms particular theisms general theism awesome theistic or or conclusive theistic evidence would cut against general atheism.

[00:51:36]Okay.

[00:51:39]But I and it would be in favor of general theism.

[00:51:44]But even then, I still don't necessarily think that it would be in favor conclusively of any particular theism simply because how do you know that it's not a trickster god who wants you to believe in Jesus, right? I mean, you're going to rule that out. An all possible, all powerful, all knowing trickster god that just is, you know, obsessed. has got like a fetish with humans believing in Jesus.

[00:52:13]Okay. How do I [clears throat] know that's not what's going on, you know? So, >> yeah. On on >> Go ahead. Sorry.

[00:52:21]>> So, I was going to say that the the the conclusive evidence one, right? I think you can get not conclusive, but you know, pretty good evidence that, you know, that that that that particular theisms are false. I do not think you can get conclusive evidence for general atheism. I think that you can get conclusive evidence for general theism and that that general evidence for for general the conclusive evidence for general theism does not translate into conclusive evidence for any particular theism.

[00:52:59]>> So yeah, that's the >> that see the asymmetry there. It's a weird asymmetry >> and it's a it's a sort of weird basian thing that I haven't really thought through enough to I I've just kind of noticed it and and [clears throat] and I'm starting to think through it, but um but it's pushed my uh my thinking about general atheism. I mean, I've already been kind of hostile to general atheism for a bit. Um but it really pushed me over the edge on that, which was great. Um so yeah. Yeah.

[00:53:31]No, it's a good it's a good uh foretaste of what's to what's to come. I and I appreciate you saying that because you know we want ideally these debates and these forums, these conversations to be productive, right?

[00:53:44]>> Yeah. and to result in a progress in the conversation. And I feel like the the eight person philosopher debate while at the outset I was very afraid that it would just devolve into into chaos cuz eight trying to organize eight people productively a prioritize. It's very difficult to do but I think we did it and I think it was a I think it was a productive conversation and that's to all of our credit and to your credit and I appreciate you like >> admitting hey I actually changed my mind on this. That's very I not everybody does that. So, so I appreciate that.

[00:54:15]>> Yeah. Yeah. No, absolutely. I I I find that, you know, changing your mind, you know, flipping 180 degrees around one of your beliefs is um one of the most kind of scary but satisfying things that you do. You know, it it can be like breaking up with a partner or something where you feel like it, you know, you've lost part of your identity and you you don't know who you are anymore, which is scary. A lot of people don't like that feeling. I kind of like it.

[00:54:44]I get that. I get that. I I'm I've always been inclined to agnosticism, skepticism. I'm always on anything I believe on most things anyway. I'm I'm only ever just above agnosticism. Just just weak belief. And [snorts] that makes it easier for me to change my mind, but it also means, you know, I've got less elements to define myself by, you know, and so there's a certain sense of like I'm missing out. But I think on the whole where the evidence doesn't warrant strong belief, it's better to mirror that in in the strength of your convictions. But uh anyway, Kevin Sharp, this has been a brilliant conversation.

[00:55:21]Not one I intended to talk about, but one I quickly thought in the moment we need to talk about this. Um >> happy to do it. I love uh love your podcast. Love everything that you do. So I'm happy to come back. Happy to be a part of this. Thanks so much.

[00:55:35]>> Absolutely. Thank you. So, we'll have you on again and we'll we'll talk about more philosophy, religion stuff. So, I appreciate it.

[00:55:40]>> Great.

[00:55:41][music]