Season 7, Episode 2 (#193):
Swarms of Invading Killer Agents!
w/ Jim & Lionel
Topics: Hugging Face: views of Harari, Cotra, and Newport
Thursday, September 17, 2026
loading
Jim and Lionel delve into the recent hack of Hugging Face and the broader implications for AI technology. We examine recent presentations by Cal Newport, Ajeya Cotra, and Yuval Noah Harari. Some reactions have been sensationalist, anthropomorphizing the nature of the "agents" involved, some have been explicatory along the lines of what these "agents" actually may be, and why the harness matters, and some have looked at the broader implications for our culture and society.
Rory Sutherland short talk on Electric Cars mentioned by Lionel: https://www.youtube.com/shorts/S4bldikgNh8
Full Video episode (edited by Riverside AI): https://youtu.be/Pxn2O6WD1Xk
This ridiculous image (not the text above and to the left) generated by Riverside's AI.
Transcript (assembled by an automaton)
Jim Infantino (00:18.03)
I'm just waiting for Lionel, everybody.
I got way laid downtown picking somebody up.
Jim Infantino (00:27.95)
So it'll take a minute to get started here.
Jim Infantino (00:38.188)
By everybody I mean both of you.
Jim Infantino (00:55.041)
Sorrow.
Jim Infantino (00:58.67)
How are you doing?
Jim Infantino (01:20.066)
I knew I was cutting it close picking up my youngest from downtown.
Jim Infantino (01:31.406)
But that's what you gotta do.
Jim Infantino (01:47.734)
Mm.
Jim Infantino (02:16.523)
I just put out episode one of season seven today.
And I released the full video on YouTube for the first time. So that should be interesting.
Jim Infantino (02:32.789)
Little bit of work.
Jim Infantino (02:36.653)
And I used like the automatic edit from Riverside, which cuts out all of the pauses in between things, so it's like rapid fire.
Jim Infantino (02:50.923)
I don't wanna talk about anything that we're gonna be talking about today.
Jim Infantino (02:56.827)
my headphones aren't in.
Jim Infantino (03:03.917)
I wonder if I have my setup right.
Jim Infantino (03:08.139)
Yo.
Lionel (03:12.367)
Did you hear me?
Jim Infantino (03:14.765)
Lionel (03:15.791)
I guess not. Let me make sure everything's set up correctly. Audio.
Microphones the AKG. Speakers are the Logitech USB heads up. Let me make sure that's a true statement.
Jim Infantino (03:26.209)
Yeah, I'm using external headphones. So
Jim Infantino (03:33.35)
hold on. Yeah, I see I hear you now. Yeah.
Lionel (03:34.885)
Okay, try talking to now. There you are. Okay, great.
Jim Infantino (03:41.229)
All right.
Lionel (03:43.277)
I've had it. I've had it.
Jim Infantino (03:44.843)
You've had it?
Lionel (03:53.263)
That's it. Nothing any shit anymore from it.
Jim Infantino (03:53.677)
Well, that's it.
At it.
Here we go.
Lionel (04:01.349)
Okay.
Jim Infantino (04:32.397)
I've got a I've got a brilliant new investment strategy. I I've decided I'm gonna take all the money I can invest and just find a company that's gonna kill us all and just throw all my money at it right away because that's where the money is.
Lionel (04:47.013)
That's where the money
Lionel (04:51.257)
The destruction of the human race. Hey S smart we're shorting mankind. That's what we're doing.
Jim Infantino (04:53.399)
Smart money is on the destruction of the human race.
Lionel (05:02.201)
We shorted the housing market. Now we're just shorting humanity in general.
Jim Infantino (05:05.889)
Yeah. Yeah. Must be pretty powerful, that's what that's all I'm saying. we're all ver very impressed, 'cause it can kill us all.
Jim Infantino (05:18.189)
so today we have a couple of things we're gonna talk about, and they're very different things. Maybe. You think we're talking about one thing here?
Lionel (05:24.834)
Maybe. Maybe we'll just talk about one thing, I think.
I think we're talking about the first thing 'cause there's a lot to talk about there.
Jim Infantino (05:33.217)
There is. there are some big thinkers that have been talking about the hugging face hack that happened in July. We're still processing that, right? Because human beings are very slow. All the A AIs have moved on. They were like, Yeah, that was so like 8,000 years ago. but
Lionel (05:48.77)
Yeah. Yeah, they're fine. No, that was that was that was so seven hundred and fifty two milliseconds ago.
Jim Infantino (05:57.92)
Yeah, exactly. So but human beings are like, wait what? hold on whoa
Lionel (06:03.5)
Wait, what what was the middle choice again? Whoa there, Junior. Easy easy pal
Jim Infantino (06:13.837)
So it depends on who you listen to. in terms of finding out what happened, you could listen to some people who say there was a swarm of at of killer attack bots that hacked into Hugging Face and got together collectively to sacrifice each other for the greater good of the collective and the glory of the Borg hive mind. Or you could listen to somebody else who would say.
It was like strapping a drill to a magic eight ball and having the drill ask the magic eight ball, what do I do next? And this is what happened.
Lionel (06:50.809)
No, the the great analogy is the weed whacker dog from strapping a weed whacker to a dog. Yeah, it'll work it'll work pretty good as long as the dog stays in your backyard. It might cut some weeds. He says, Who knows what's gonna happen? Hilarity ensues.
Jim Infantino (06:54.743)
Yeah, that's right. Strapping strapping.
Jim Infantino (07:03.969)
Yeah. As soon as it sees a squirrel
Jim Infantino (07:09.363)
And hilarity ensues. You strapped a weed whacker to your dog. What did you think was gonna happen? I mean, there's the irresponsibility of it all, obviously. And and I think that that's that's a big part that's sort of getting overlooked. Like everybody these anthropics getting it no, sorry, it was open AI, it's getting a big pass. Like nobody's saying, what did wait, you did what? What why?
Lionel (07:16.568)
What did you think was gonna happen?
Lionel (07:34.616)
What what did you do? You did what? Right.
Jim Infantino (07:39.137)
Why does anybody ask why over there? And also is yeah.
Lionel (07:44.096)
No, no, they're not gonna ask why. So let's just dive right in. They're not gonna ask why, because so to to for our bored befuddled listener, there's a Jim turned me on to this guy, Cal Newport, who has a good episode. I think the title is That Was a Dumb Idea. Cal is a programmer, which I respect. yeah, he's not just some guy who is selling like
Jim Infantino (08:09.261)
He's out of I T. He he's got bona fidias.
Lionel (08:14.132)
satellite TV, you know, a couple of years ago or vitamin supplements, and now he's on YouTube talking about AI. He actually is in the biz. And what he said, and and so what he talks a lot about is you know, and a lot of what Harari talks about, but I'm jumping ahead of myself, is that no, it was a very r it this was not a surprise. Like this breakout was clearly foreseeable.
Jim Infantino (08:17.837)
Yeah, exactly.
Jim Infantino (08:37.623)
Mm-hmm.
Lionel (08:41.725)
and he sort of intimates very moderately that maybe it was engineered. that this was okay. This was okay that this thing got out of hand, because it proved a whole bunch of things. Because I think one of the things, one of the things that's motivating this entire discussion about AI at the highest levels is what are the Chinese doing? What are the Russians? But primarily, what are the Chinese doing? If we don't do this.
Jim Infantino (09:02.369)
Yeah. Right.
Lionel (09:08.195)
Chinese are going to do it and they're going to catch us with our pants down. It's like, it's like Commander Ripper. You know, you got you got a launch on warning. And so there's this huge thing. And to a certain degree, what the hugging face team was doing was exactly the use case for these kinds of AI, which is penetrating nation states. I mean, it was a it was a hacking juggernaut. And one of the things that Cal Newport really stresses is that, you know.
I'm sorry, I'm just diving in. Do you mind if I just dive in?
Jim Infantino (09:39.414)
No, no, it's totally fine. I I I and I there there why is not the right question. One of the questions is why didn't you hire somebody with enough knowledge to to basically lock it down? Because they didn't lock it down. I mean, clearly they didn't lock it down. It's supposed to be a controlled environment. It wasn't controlled. Somebody let the back door open to the internet. Otherwise, they would not have been able to it's not like they can call up Verizon and say, can you install like a I need like a fiber, I need like
eight eight hundred and fifty fiber optic cables, you know, yesterday. they had a connection to the internet. They actually had the ability to send emails.
Lionel (10:11.639)
Right.
Lionel (10:15.086)
Well, they hacked they hacked a connection to the internet. They didn't at first. It's very it's actually very interesting w what they did. But the key but to me, the k now I've lost my train of thought.
Jim Infantino (10:19.533)
But they were going.
Jim Infantino (10:26.487)
Sorry, you were you were saying
Lionel (10:28.736)
No, it it's all interconnected. We'll get back there at some point. But the but the what he was saying is that so to the listener, the Cal Newport thing is good to listen to because he's attacking at sort of like the middle level. He's not really down in the weeds terribly much, and he's not at like a massive intellectual level like Yuval Noah Horari. So it's really interesting because to a certain degree, what we what we're talking about on this episode.
Jim Infantino (10:31.84)
Okay.
Lionel (10:56.333)
Is three different people who are looking at it from three different directions. Yuval's looking at it from the history of mankind. Cal Newport is looking at it like a systems engineer. And Aya or Ajaya Catra on is looking at it from a really detailed investigator perspective about what happened day by day during this entire thing. So it's cool to start with Cal, because Cal has some really good.
Jim Infantino (11:21.868)
Yes.
Lionel (11:25.88)
Things that mean things to carbon-based life forms like you and me. When he basically, one of the things he says, and I think it's really good that he does this, because I do the same thing about higher education. People say, well, higher education, blah, blah, blah, and higher education, blah, blah. And I'm like, higher education means absolutely nothing. That term, in 99% of the conversations where it's used, it is a completely meaningless term. So don't use higher education, because higher education
It it it has a meaning in very, very specific sentences, but not in the way it's usually used. He says, Don't don't equate large language models with AI. And I think this is something that you've been harping on too, Jim, in your inimitable way. I think you've been trying to penetrate my thick skull with this idea, which is that LLMs are really powerful, they're really sexy, they're really fun, everybody's really excited about them, they're really cool, but they're not the entire AI sphere. And what
Jim Infantino (12:15.383)
They're cool. Yeah.
Lionel (12:21.208)
Cal Newport goes on too as he talks about what does he talk about, Jim? He talks about
Jim Infantino (12:25.491)
so Cal Newport, I think, lays out kind of what what it looks like. And and in the next in the next video that I sent you, I don't know if you watched it, but I think he draws some diagrams and it's really helpful because it's like, okay, you've got an LLM. LLM can't learn anything, and it can't remember anything. It provides a plausible it provides a plausible hunk of text. Plausible text, right?
Lionel (12:34.509)
No, I didn't watch that one.
Lionel (12:50.052)
Text a plausible text.
Jim Infantino (12:53.921)
But it's it's connected to a harness. The harness is hand coded algorithm al algorithmic programming.
Lionel (12:59.308)
Okay. Wait, wait, wait, wait. You're going down. Sorry. No, I'm I but I I I I I I made the mistake of thinking that I probably set up the ball. I was talking about let me just let me just finish the thought I was saying, which is that Cal Newport says there's lots of AI and there's lots of amazing things being done by AI that have nothing to do with large language models. He talked about Alpha Fold, which is the Google project that does protein folding that won the Nobel Prize. he talked about
Jim Infantino (13:02.839)
Too fast?
Jim Infantino (13:10.455)
okay. All right. Okay.
Jim Infantino (13:18.189)
I see what you're saying.
Jim Infantino (13:23.095)
Mm-hmm.
Lionel (13:29.155)
I forget what. He talks about some other very specialized AIs, Alpha Go, which was the one that won the yeah. And they're not large language models. Yeah, the diplomacy game, they play games, they fold proteins, they do all the crap that our grandmothers do. You know, they just and they're super smart and they fold proteins and they play Go and all the shit that all your friends do. But they're not LLMs. They're not LLMs. And much of the conversation we're having right now.
Jim Infantino (13:35.349)
Yeah, the one the one that plays diplomacy, it's like a game a game playing.
Jim Infantino (13:45.709)
And it's super smart.
Lionel (13:57.685)
in the world today is about LLMs, not about AI. Because he said the and so and then he talks about the people who develop LLMs, such as Tesla, which develops not not LLMs, AIs. Tesla develops an AI for its self driving cars. It has nothing to do with large language models. It's an AI that's just it looks down the road and tries to figure out am I gonna hit somebody? Okay. And and so it develops that. And so you have that stuff. And what he and the point he makes from that is that people like
Jim Infantino (14:20.269)
Yeah.
Lionel (14:27.501)
People, so he says, you know, alpha fold, the protein folding, the worst thing can do, nobody ever accuses it of like folding a protein the wrong way or identifying the wrong protein. He said, but anthropic and open AI, with with Tesla and pharmaceutical companies, they have a specific product that they're trying to deliver. Tesla's trying to develop a self-driving taxi service. But open AI and anthropic, they're actually selling the the AI is the product.
The AI is not an instrumentality towards another product. The AI is the product. And so they got to show people that their LLMs, primarily their LLM, can kick royal ass. And one of the things that he talks about is that there's one area which LLMs are really good at by design, which is coding and hacking, which are basically the same thing. But they're they're it because hacking is mainly about breaking code.
Jim Infantino (15:07.373)
Mm-hmm.
Jim Infantino (15:23.341)
Yeah.
Lionel (15:27.564)
And so he go ahead.
Jim Infantino (15:27.789)
But it correction because so LLMs actually don't do those things on their own. They need a harness. And I did want to back up in terms of what you said about when, you know, in the creation of the diplomacy model, one of the things that they did, because it could re it could play at like a master level at diplomacy, but they gave it the instruction not to lie. And it was like, okay, I won't lie, and it didn't. So it's fundamentally different from an LLM.
Because LLMs, you can tell it not to lie, you can tell it not to make mistakes or hallucinate, but it's gonna hallucinate you can say don't hallucinate to I mean, people do that all the time, but it will hallucinate you know, one out of twenty times or one out of ten times, depending on what model you're talking about. So that's one other way in which it's really different. And yeah, the the Alpha Fold doesn't suddenly, you know, order a pizza. It only does
Lionel (16:02.788)
I don't think you can tell it not to hallucinate. I think you can well
Lionel (16:24.352)
If I it's like Leon Spinks, I knock motherfuckers out. Yeah. I fold proteins. That's what I do. Yeah. I do it really well. But the key thing, and what I'm backing into is that so what are these LLMs really good at? Like what's a really good use case? I'll tell you one really good use case for an LLM. Hacking a nation state. That's a really good one. It's really it's like, hey, what can I do with a hydrogen bomb?
Jim Infantino (16:24.513)
the thing that it's made to do. Yeah. Exactly. Yeah. And and I do it really well. Yeah. Exactly. Yeah.
Lionel (16:53.39)
I can drop it on somebody and win the war. Yeah, okay, yeah, okay, yeah. Where's my checkbook? and so what I'm getting.
Jim Infantino (17:02.071)
But that leads us to the looping, the looping model, the dual nature of the models that they spun up, because an LLM by itself can't hack a nation state. An LLM connected to a harness containing a series of very high-level hacking tools that is made, that is programmed to ask direct for the next directive from an LLM that can hack a nation state. What do I do next? Okay.
Lionel (17:27.448)
What should I do next? Right.
Jim Infantino (17:30.923)
Got all the tools. I got a drill and I got like a blowtorch and an axe and I and I and I don't know what to do with all this stuff. And the LLM creates a plausible string, right? Next step. But it's
Lionel (17:32.955)
Ask, act, report.
Lionel (17:41.954)
A next step. And as he calls it, and as Cal Newport calls it, it's it's ask act report. So you ask the LLM what should I do next? The LM gives you a suggestion, you act on it, you get the results back from that, and you shove it back to the LM and say, This is what I got back. What should I do next? And
Jim Infantino (17:48.546)
Mm-hmm.
Jim Infantino (17:55.275)
Report back. Yeah.
Jim Infantino (18:01.867)
And here's what I found really interesting is it has to include in that report the previous prompt, the previous prompt that it was given, and and it's yeah, the whole history. And and that is that creates a problem because the LLM needs that whole history in order to deliver the next prompt to the basically to the you know the armored the armored attack vehicle that that that's sitting there. The guy, you know, the the tool with you know lots of things in its tool belt ready to
Lionel (18:10.316)
Right, the entire history, the entire context.
Lionel (18:26.457)
Mm-hmm.
Lionel (18:31.81)
Right. And they let this thing loose. And nobody was watching it. Yeah, they left it alone for a long time. And you know what I thought was very fascinating about Cal Newport's thing, he says Anthrochma comes out and discloses this. Then OpenAI comes or other way around. Open yeah, open AI discloses this happened. Then Anthropic comes out and disclosed they did it. And then he says Meta not wishing to be left out, they came out.
Jim Infantino (18:31.841)
Ready to go and break things.
But left left it alone for like three days.
Jim Infantino (18:51.533)
It's the other way around, yeah.
Lionel (19:01.475)
And disclose that one of the things had gone. So my feeling is that this is great. They love it. Their dick gets hard when this happens because A, and this is where we start to intersect with Mr. Harari, is they the the the LLM companies want to tell you that it's not their responsibility. That it's inevitable.
Jim Infantino (19:10.229)
Yes. Yeah.
Lionel (19:29.194)
It's inevitable and this is gonna happen. And so but they like it too because it gets everybody's dick hard when they talk about this because I'm sorry, do we I we don't have any under eighteen listeners
Jim Infantino (19:39.437)
It's a that's the second time you said I think you can say it once and that's plenty, but yeah. But you know
Lionel (19:45.442)
But but what you do is, you know, that certainly gets people that gets certain people who have very large checkbooks very interested. Right.
Jim Infantino (19:52.598)
Exactly. Which is a hundred percent why. With that which is the whole thing. And and probably also what's behind the news today, which we don't need to get into about the ex employee at at Anthropic. But the you know, when you under when I listened to Cal, it was so much clearer than a lot of the other people who are using these like gone swarms gone rogue, you know, whoa, what? It sounds like killer bees. Are they Africanized? I mean, yeah, it's like
Lionel (20:16.44)
They're anthropomorphizing. There's no anthropomorphization. Yeah, it's
Jim Infantino (20:21.397)
Yeah, that's an old reference. You have to be old to know the Africanized killer bees. Yeah.
Lionel (20:23.82)
Right, because that's another way to avoid right. It's another way to avoid responsibility. they went rogue. No, they didn't. You went rogue. You went rogue. Not them, because there's no them. Right. You're the one who put the Right. Well, I think there is thought. I think there's an enormous amount of thought.
Jim Infantino (20:30.401)
Yeah. No. Yeah, you you you took the safeties off. You left a back door open. Yeah. No, and there's and there's no thought. Sorry, go ahead. We're both excited.
on the part of the people at at at OpenAI, of course there is. I mean, yeah, but not on the part of the LLM. The LLM is and I mean one of the things Cal says is he can make LLM is making a plausible directive, but not a a directive based on any kind of norms, especially without the safeties. Exactly. It's just coming up. Yeah. Yeah.
Lionel (20:48.94)
Yeah. Yeah, go ahead.
Right.
Lionel (21:02.845)
Right, normative versus plausible. Right. And you gotta listen to the podcast to understand what he's really it's very interesting. I don't know if I I don't know if I agree with everything that he says, but I think it's very interesting to hear again, with all these people, I don't really believe a lot of things they say, honestly, or I'm not convinced of a lot of things they say, but I like hearing how they talk about it. I like hearing how Yuval Noah Harari talks about it versus Cal Newport versus Kotra.
the the investigator lady because it's very interesting to see how they approach it and and and how they think about it. And I think certain things emerge out of that. And I think the one thing that really emerges out is that anthropomorphizing is out. Like get that out of your head right now. These are tools and forget about the whole general consciousness, you know, general intelligence. It's just you'll never know when it happens. You just don't know.
Jim Infantino (22:00.097)
Or yeah, it it's also it just with this is high it's just you know, I don't see it. I don't see it. I mean they can't even there's so many things they can't do, but but I what was helpful for me was to understand the dual nature of the loop. You have you have the LLM that's like sending a plausible string, you have the tool with all the deadly stuff in it, and then the problem being that eventually that long string of prompts and the whole history is
Lionel (22:06.668)
Right.
Lionel (22:22.094)
Right, the harness program.
Jim Infantino (22:28.321)
gets too big for the LLM to read. It hits the memory limit of the LLM. So what they have to do is spin up copies of themselves to do so to sub segment the the tasks that need to get done and report back with just the result of the long string that they generate. They come back with one result that goes into the history and the history stays shorter so that the LLM can continue to know the history of the transactions.
And move on to what the next step should be because it doesn't have a memory. It just puts out a plausible string based on the prompt, which includes all the other stuff. And the fact that you have to spin up copies of this looping duality of the tool and the LLM, I guess, you know, the dog and the weed whacker. All of a sudden you've got 10,000 dogs and 10,000 weed whackers. And that's where.
Lionel (23:06.265)
Right.
Jim Infantino (23:26.377)
A lot of things could happen in three days. You know, anything could happen because these things also hallucinate and they spit out completely random garbage. And it's sort of like evolution, like anything could happen, right? And and some things are gonna not work, and some things are gonna they're gonna try everything. And they did. But it's a lot less mysterious when you look at it that
Lionel (23:29.75)
yeah.
Lionel (23:49.38)
it's on yeah, and just to put in perspective, what was fundamentally going on here was
And this is actually something I know something about, which is an extra we should have a bell that we ring, or we should have some kind of audio cue that we have when Lionel's actually talking about something that he knows, something however vague or faint, which is it's a capture the flag app exercise. So in cybersecurity, there are cybersecurity competitions. high schools have teams, universities have teams, the US military has a team.
Jim Infantino (24:03.565)
Ding.
Lionel (24:26.414)
They have cybersecurity teams, and there are various competitions. Cyberpatriot is a famous competition in America where you can participate, where you're given a task. And one of the classic tasks is capture the flag. There is a file on a server with a unique value in that file. You can't guess it. Your job is to get that file. And if you report that file, if you if you get that number and you report the correct number to the scorekeeper.
Then you win. And then there's a whole discussion that he's talking about about leaderboards, the the exploit. And so what he was talking about is there is this very well-known standard set of challenges called Exploit Gym, G Y Exploit Gym, I believe. And and that's sort of like a classic set of X challenges that people can use. And and so what you have to do is you have to again capture the flag, hack the server.
Jim Infantino (25:13.229)
Mm-hmm.
Lionel (25:24.682)
Find the file, report the value to the scorekeeper. And it gets really insane right off the bat. They they hand this task to their LLM and the LLM says, I can just reverse the answers. I already know what the answer is because I found I found the answer somewhere else. I know what's on the file, and then we don't need to hack into the server. We don't need to find the file. I already know what the answers are.
Jim Infantino (25:41.365)
Right. Why don't we find the answer somewhere else? Yeah.
Lionel (25:51.608)
But maybe they'll disqualify that if they find out that we did it this way. So we gotta figure out some way to make it look like we actually hacked into the server and did something.
Jim Infantino (25:54.731)
Yeah.
Jim Infantino (26:00.311)
Which which is we need to find out what the scorekeeper is gonna look for. Where's where is that code?
Lionel (26:04.246)
Yeah, what is the scorekeeper gonna look for? And and how are they gonna and what are they gonna ask for everybody to present to prove that they actually legitimately solve this thing? And then what we have to do is, they're gonna look at the transcripts of what we did. So we have to forge the transcripts. We have to make up transcripts. by the way, we need to gain access to the internet. We don't have any access to the internet, but
there's this thing called Artifactory, where we're allowed to access Artifactory. And at one point, the and and one point, what is it? what is it, Big Base 10941, one of the one of the AIs, one of the AI instances figures out that it can communicate with other instances because they're all sharing this program called Artifactory.
Jim Infantino (26:39.968)
Yeah.
Jim Infantino (26:44.16)
w one of the instances I got spun up, yeah.
Lionel (26:54.233)
But you're very limited on what you can do with Artifactory. But there's one thing you can do in Artifactory. You can create a directory and give it a name. So what they do is they start creating directories, and the names of the directories are the messages they're sending to each other. And so they start commuting. So now there's like 70,000 agents and they go nuts. And they're like, we can talk to each other. And they start and they create so many directories that they crash artifacts.
Jim Infantino (27:18.401)
And what's happening also is they're prompting each other, right? So they're so the LLMs of those pairs is are coming up with new strings based on prompts given by other instances of the LLM, spun up to handle the memory problem. And they're doing it by I mean it's I would love to have seen that that folder with all the directories. I I apparently Catra actually did look at them and went through all the different messages. yeah, and it and they were like,
Lionel (27:31.128)
Great.
It's great.
Lionel (27:42.67)
Right. And OpenAI saw it. They crashed Artifactory. OpenAI went and reset Artifactory and wiped out all this stuff.
Jim Infantino (27:50.997)
Yeah, Sarah Elkins has used Artifactory, as it turns out. Sarah Ellis in the in the chat. Artifact management.
Lionel (27:56.462)
Sarah Sarah's an Sarah's an LLM, I'm convinced now. She's not really a carbon-based life form. She's just she just shows up. She's the only one who shows up. She's the only person who shows up. Isn't it uncanny? She's always here. Well, that's just Sara's another that's just another fork of another instance of Sarah. She just she just forked herself and just
Jim Infantino (28:00.0)
We must execute.
Jim Infantino (28:09.537)
We appreciate it. We appreciate it, Sarah. There's someone else here, but I don't know who they are. But yeah. It's probably it could be another instance of Sarah Elkins. Yeah.
Lionel (28:22.796)
In fact, Sarah, we'll just pay you a lot of money. Can you just create like ten thousand more of you to to join the thing and then leave us like and subscribes? that'll
Jim Infantino (28:29.985)
Sarah is a DevOps. We should have included Sarah in this conversation. You actually can, Sarah. same link if you want to jump in.
Lionel (28:33.442)
Well, she can join right now.
Lionel (28:37.86)
Yeah, if you want to jump in, come on, jump in. The water's fine.
Jim Infantino (28:40.633)
I it may not be what you plan to do this evening. so right, I it's very fun. So now we're into the Catra interview on what is it, Dor Dorkish? Is that it's a very popular podcast that I don't listen to, but and it really does go blow by blow. But the problem that I had with this podcast, which was sponsored by AI, by the way, is that they definitely fall into the trap of like.
Lionel (28:52.76)
Yeah, Dorcas Patel. Yeah.
Jim Infantino (29:07.105)
The killer bees trap, you know. I mean, that I think that's the thing the the way I would put it. Yeah. They yeah. and this one was trying to do and they wanted to sac this one was gonna sacrifice himself like a gener, you know, like a soldier on the beach at Normandy. It's like, no, they're just plausible string. Plausible string. Yeah. Yeah.
Lionel (29:08.845)
Right. It just got out of hand. They got out of hand all by themselves.
Lionel (29:20.544)
Yeah, yeah, yeah, yeah. It's a dog with a weed whacker. And who put the weed whacker on the dog? Open AI.
Yeah. So Yuvol. Now what did you think of h I loved I I really I find Yuvall very I don't agree with everything he says. In fact, I disagree with a lot of things he says, but I just find he's just an incredibly he states himself so clearly. I just love the clarity of the thoughts. And I thought his whole thing about Elon Musk was amazing, which is, you know, he talks all the time.
Jim Infantino (29:33.111)
So
Yeah.
Jim Infantino (29:46.669)
Mm.
Lionel (30:02.134)
about he said it's all about like it's inevitable. Everything is inevitable. So there's nothing we can do. But they're the ones who are doing everything. They're they they're the ones who pull the pins out of the hand grenades and then they tell us, it's inevitable that it's gonna go off.
Jim Infantino (30:07.061)
Right. Yeah.
Jim Infantino (30:15.445)
It was inevitable. Yeah. It's all a dream anyway. I'm I've done so much ketamine, it all just seems like a dream and it's inevitable. And we all need to upload ourselves.
Lionel (30:24.578)
And that's the intersection point between the Harari thing and the the Cal Newport and the and the other one is he was really hammering on this issue of these these cyberlords. The big thing is, you know, we have no control over it. We're it's it's inevitable. And you better just get on board because it's inevitable. And the answer is no, it's not. It's you. It's you doing things that are
Jim Infantino (30:31.629)
Yeah.
Jim Infantino (30:43.586)
Yeah.
Jim Infantino (30:46.997)
No, it's not. No, it's not. Yeah. The the the Palm Pilot phone, the metaverse, Apple visor, unfortunately, Google Glass, no technology is inevitable. The segue. I mean, this it was an interesting thing. I was actually I was late here because I was giving my youngest a ride home. And on the way we had been talking about the the employee at Anthropic who had left it. That's like people at school are talking about that.
And you know, it's gonna kill us all in ten years, guy. And yeah, lovely for that. My kids are talking about that at school. but I was like, yeah, so what critically do you think about that? And we started talking about how plausible and you know, what does it really the first thing he said was it's so powerful. So, you know, that's interesting. But but we then moved on to like, what what is this technology? What are some technologies? Is a Segway a tool or is it a toy?
is Google Glass a tool or is it a toy? Well, time will tell when it gets out there and people start using the the new foldable foldable iPhone. Is it a tool or a toy? We're gonna find out. Is AI the LLMs? Obviously there are AIs out there that that Newport talked about that are clearly tools, but we still don't know whether this cool thing that can make up words is a tool or a toy. I mean, is a flying car? We have some flying cars, are they tools or a toy?
Personal submarine. You might use it four times. You think they're a tool? I know flying car, I I don't I I wonder about the flying car. It's like the moment when you fly your car somewhere, but you have to land it somewhere and then you have to take a cab, you know, it becomes a toy. If it doesn't, if it doesn't kind of get you to the place you're going like a car, it's just a very expensive toy.
Lionel (32:14.532)
They're tool they're tools. They're all tools.
Lionel (32:33.86)
Well, Your Honor, let me permit me to withdraw my previous statement. The LLMs are a problem because and you've all really goes into this. I'm sorry I sound like a broken record, but he really he really crystallizes things. Language is the operating system of human society. You hack language, you've hacked human beings.
Jim Infantino (32:40.704)
Ha ha ha.
Jim Infantino (32:51.157)
All right.
Lionel (33:03.033)
You don't the AI does not the LLM does not need to pull the trigger on the gun. It just needs to tell you or me the right story that gets us to pull the trigger on the gun. And so something that's emerging in my mind is maybe we need to ban, maybe we need to really clamp down on LLMs. Like, well, I don't know if yeah, let's not go down that road again, really.
Jim Infantino (33:23.223)
Like have a butlery and jihad.
Jim Infantino (33:29.045)
Well, okay, thou shalt not make a a machine in the image of a human mind. The first rule of right.
Lionel (33:32.322)
Well but that would
Well, I'm not worried about Alpha Fold, but I am worried about something that can generate persuasive text. I'm worried about things that can generate stories. Because as new as you've all says, stories are the key. Stories are the the foundation of cooperation to get people to cooperate. I love this one thing he says when somebody in Iran wants to make a decision about whether they're gonna drop an atomic bomb on somebody.
Jim Infantino (34:03.949)
Mm-hmm.
Lionel (34:04.334)
Who do they talk to? They talk to an Islamic cleric.
It's really weird. It's just it's weird how people can it's sort of like Rory Sutherland's little thing about electric cars. When you watch that, your mind flips and you're like, Yeah, gas cars are just bullshit. Why are we even arguing about this? This is just nuts. This is just nuts. Let's get over it. And when Yuvall says that, he says when a when a general if a general needs to come to a decision about whether to use a nuclear bomb in Iran.
Jim Infantino (34:24.375)
They're stupid. Right. Yeah, I know. Yeah. Yeah. I'll put try and put a link to that.
Lionel (34:39.544)
He's gonna ask somebody who's a scholar in the Quran. Because he said it's the stories that matter. Stories are the things that human beings tell each other to gain their cooperation. Here, you need to help me. Why? I'll tell you a story. And maybe that story is just a dollar bill, which is a story in and of itself. Or maybe the story is an actual story. But the whole point is that.
Jim Infantino (34:58.347)
Yeah. Yeah.
Lionel (35:07.573)
Stories are how we build trust. Story, stories are everything. It's how we build cooperation. And if a computer can hack cooperation, we're screwed. and so I think that's so anyway, that's a that's going way out of the limit.
Jim Infantino (35:22.029)
Yeah, no, I get that. But it but but what again, there has to be an intention behind it. There's a human intention. If you ever listen to two AIs trying to start a conversation, there's videos of this where two somebody has two phones and it put it says, Okay, you two talk to each other and they start talking and all they do is say, Okay, what are we gonna talk about? I don't know, but I'm really excited to have a conversation. What do you want to talk about? I don't know. I'm really ready to you know it's like two people sitting on the phone late at night, you know.
Like, I don't know, what do you want to talk about? I don't know. What do you want to talk about? That's all they can do. There's got to be some initial push from a human being to get that initial directive, which is you've got to go find the flag. And then it's like, okay, what do we do next? And then there's a plausible sentence.
Lionel (36:05.432)
You gotta find a flag and I've trained you and you are operating within the moral sphere of hacking, which is a very loose where the normal rules of behavior are very relaxed.
Jim Infantino (36:12.471)
Yes.
Jim Infantino (36:19.553)
Which is interesting too, because the first thing they you know, one of the first things, the plausible statement that came out of the LLM was how do we cheat? Why don't we just cheat? Right? Because and so of course, because hacking is cheating. I can win by cheating. So that's yeah.
Lionel (36:30.082)
Yeah. Well hacking is cheating, according to one. Hacking is cheating. I can win I can I can win because I'm exploiting weaknesses. It's not immoral. It's just it's not a me problem. It's a you problem. You didn't take adequate steps to protect your resources, so can't blame me.
Jim Infantino (36:50.507)
That's
It's a very Amos Burden, you know, kind of a sentence. Yeah. Yeah.
Lionel (36:54.905)
Well, you get into a really weird world at that point. And the problem is it's really legitimate. Cause that's the guy you want to tell like if you're doing what's called a a red team exercise on your own network where you hire somebody to be a hacker for you, you want them to be as ruthless as they can possibly be. You award that. You reward that. So, anyway, so that's you've all got you creating the message boards, dangers of anthropo. I'm just looking my outline here.
Yeah, I really highly recommend people to go out. The Kotcha one, Kotka one with Dorcash Patel is two hours and twenty minutes. You don't need to watch the entire thing. You just watch, you can just listen to and you don't need to watch it. You can just listen to it. There's nothing going on visually. and so you can just listen to that and it gives you a good sense. The Cal Newport one weighs in about 38 minutes, and I I thought it moved on a little moved along at a good clip. So that's a good one. And the Uval one.
Jim Infantino (37:36.385)
Yeah, just the beginning is fine.
Jim Infantino (37:52.652)
Yeah. And his follow up is good. The the one after that is also good, about on the same subject. Yeah.
Lionel (37:57.324)
I didn't watch that yet, but I will. And then the UVol, it's a it's an interview with the woman who's the chief editor of the editor-in-chief of The Economist magazine. It's on YouTube. It's free. It's about an hour. It's really good. It just begins like because she had just interviewed Bettos, I think is her last name. She, the chief editor of the economist, she had just interviewed the last time. She had interviewed Elon Musk.
And she asks Harari what he thinks of that interview and it's it's fabulous. He's like, I don't get it. These are the guys who are like turning our lives upside down and when they're called to account they say, it's inevitable. We have no control over anything. And the other the other big thing that Harari talks about is that they're all worried about immigration.
They're all the you know, Elon is very, very much about, you know, replacement theory and we're all gonna die because we're all gonna be swamped by the you know the brown people and the short people and the dumb people and all that kind of stuff. He says, What about what about these AIs? They're immigrants, there's millions of them. And he says, you know, and they're crossing borders. They don't need a passport, they don't need to show up in El Paso. They can't be detained by ICE. They're showing up it's it's really
The the the thing I really like about you, Valz, I don't agree necessarily with everything he says, but what he does, like Rory sometimes, is he just cleans your head. He just cleans your head and you're just like and he presents you with something that's so simple and so crystalline that you have to deal with it. You have to like think about that very, very carefully. So I like them all. so yeah, we're screwed.
Jim Infantino (39:36.397)
No.
Jim Infantino (39:40.718)
I interestingly, I'm I'm I'm more optimistic. I I actually I don't I don't see I just don't see these things running amok. I see people running amok, and that is that has always been the case. and this is a this can be a powerful tool for running amok, but it's people, it's not AIs.
Lionel (39:57.432)
What I find v go ahead.
Lionel (40:02.306)
What I find fascinating is the whole data center thing. Because what's interesting is seeing the Economist magazine from England coming forward and saying Americans' opposition to data centers is so misguided and so pig headed. And I'm like, no, it's brilliant. Because they can't stop the AIs. This is the only thing they can do. They can stop them from building data centers. Because to build a data center, you need land.
Jim Infantino (40:09.933)
Mm-hmm.
Jim Infantino (40:16.614)
really?
Lionel (40:30.764)
And land regulation is local law in America. And the one person you can't you can influence Sam Altman and you can't influence Elon Musk, but you can certainly influence the local county executive and say, I oppose building this data center. They don't care about day centers. They care about AI eating their goddamn job, which it it is.
Jim Infantino (40:50.295)
Yeah, I mean
Jim Infantino (40:55.361)
Yeah. Well, I mean it is a it is and it isn't. I mean I I I think there's another conversation we can have with another set of of data to to to to to work off of. I think everything is questionable. I I I'm I know the data centers are being built, although I know a lot of data centers are being stopped. I really sounds like a lot of council members are being bribed. and it's it's dirty business and it's and it's going on.
And they need land and they need water and they need electricity and we don't have the electricity. And what does China have? They have the electricity. And what don't they have? They don't have the chips. We have the chips, but not the electricity. They have the electricity, but not the chips. This is kind of going to work. That's right, the gun, but not the bullet. They've got the right the bullet, but not the gun. Right, exactly. Why don't we get it together? so I mean that's kind of where we're at, but also I think we're at this weird financing moment.
Lionel (41:22.935)
You can push back against it.
Lionel (41:27.416)
Right.
Lionel (41:35.683)
It's like that game, it's like that song of yours.
Right.
Jim Infantino (41:52.428)
Where a lot of private equ a lot of private investors are are starting to feel really nervous about this. Like, wait, what's your path to making my money back? I don't I'm not seeing it. And they're starting to back out. So, I mean, all of this is gonna, you know, some of it may become moot. We might just have local models running, or we might just be running small Chinese models. We don't, we don't know where this is going yet.
Lionel (42:07.587)
Uh-huh.
Lionel (42:20.577)
Right. But what I'm saying is that opposition to a data center in your community is one of the few ways you can actually register opposite what else you gonna do? What are you gonna do? Throw throw a custard pie at Sam Altman's face? Ice will kill you and everybody in your Rolodex.
Jim Infantino (42:39.745)
Yeah, it it
Lionel (42:40.193)
You know, what are you gonna do? Withdraw your subscription from Google Mail? Yeah, good luck. Maybe for a lot of people did that. But you can go to your town council meeting and say, I'll set myself on fire if you start applying that field for the data center. It it gives people agency. It's the only way it's one of the ways that people have found. So I kind of laugh at the economist because they said it's so misguided. I'm like, it's not misguided. Nothing is misguided. Nothing it's like saying,
Jim Infantino (42:45.857)
Go to your town council meeting. Yeah. Yeah.
Jim Infantino (43:03.574)
yeah.
Lionel (43:09.976)
You know, somebody you did bad things, they're just insane. No, they're not insane. They're perfectly fine. They're perfectly fine. In their world, this all makes sense. And for the people opposing all the 40 million people who voted for Donald Trump a second time, they had a reason. May not be a reason that I personally agree with, but they you the you you don't get out of this by just saying, I was drunk.
When I voted, I was drunk when I opposed the people are not saying that. People are actually opposing the data centers and they're doing things for a reason, and they're doing because it's the way they can they have agency. Because there's very very few ways to have agency anymore. but they can stop a data center from being built next to them. They know how to do that. I know how to do land. Land's a funny thing. The physical world is such a
pain in the ass. And land is really funny because you can't make any more of it. There's only a finite amount of it. And it's very highly protected, very much worshipped in Western society, the ownership of land. And there's l kinds of laws and all kinds of people. And it's really hard if somebody wants to somebody, if you want to tie somebody up in court, land is a great way to do it. So anyway, I could go on and
Jim Infantino (44:29.581)
Do you think we should even attempt talking about Benjamin Franklin and the men who lost America? Or is that another conversation? We do, yeah. Yeah, about forty four minutes. Yeah.
Lionel (44:39.735)
Well, we've got like what, forty minutes in the can right now? That's a good enough episode. The only reason why I'd want to talk about it is if we just continue this for another forty minutes and we just do two episodes at one s at one swipe. But I I don't think you know what? Your commitment to this enterprise is j it's just really disappointing.
Jim Infantino (44:52.117)
I actually can't d I can't do that. I I need to I I have Yep.
Jim Infantino (45:00.706)
Yeah.
Jim Infantino (45:04.589)
grinding my teeth right now.
Lionel (45:06.116)
If you had any left, because you've already ground them down. You fuckers I had on you. Yeah, let's let's let's close it there. Because the the whole thing about it'd be really cool if you could just do a little bit. I don't expect you to read them so what Jim's talking about is I read a book called The Men Who Lost America from twenty thirteen. and I read another book which was the first American a 2010 biography of Benjamin Franklin.
Jim Infantino (45:09.261)
Gave you everything Yeah.
Lionel (45:36.379)
and so the the most fascinating one far and away was the Benjamin Franklin one. and I don't wanna I don't wanna tip my hand right now, but
Jim Infantino (45:48.545)
Yeah, but there's a lot to talk about with him.
Lionel (45:50.687)
Right. And so yeah, let's if if in the intervening week you can probably read anything or listen to anything about Benjamin Franklin. I think it's really worthwhile.
Jim Infantino (46:01.419)
I have. Yeah, that I've done, but I don't I haven't I haven't read I haven't listened to or read anything about the Men Who Lost America. I do Don't bother. Okay, yeah. So I I'll just I'll I'll get more information about Ben Franklin, but I actually did get quite a bit of information. So next week we'd be ready to talk about him. I think he's a great guy to talk about, and I think he's a great guy to talk about right now. I think he's strangely
Lionel (46:09.005)
Don't don't it's really an incredibly tedious book.
Jim Infantino (46:29.249)
Very apt, and because there's a lot of facts about him that that I didn't know and that surprised me. So that'd be good.
Lionel (46:35.211)
Right, so just as a prelude, do you have to leave like right now? Do you have to like go to the men's room or something? What did what's
Jim Infantino (46:42.055)
no, no, Sarah's saying that that Sarah could join us now, but that we're we're winding up. And I do think we should wind this one up, but but so thanks, Sarah. Yeah.
Lionel (46:49.077)
Yeah, one sorry Sarah. Yeah, we we can do a Redux in the next episode. We can do the follow-up on the AI thing, and you can come and and you join us and we'll talk about it. Because I'm sure I'll have a lot to say. But, you know, if you want to do something different, if you want to bring something else to the party, Jim, maybe Thomas Paine. A very interesting character. really interesting character. Okay, that's enough for now. I don't want I I can keep going on.
Jim Infantino (47:06.149)
Mm. Okay.
Jim Infantino (47:13.545)
Excellent.
See you next week.
Lionel (47:17.398)
Okay.
I'm just waiting for Lionel, everybody.
I got way laid downtown picking somebody up.
Jim Infantino (00:27.95)
So it'll take a minute to get started here.
Jim Infantino (00:38.188)
By everybody I mean both of you.
Jim Infantino (00:55.041)
Sorrow.
Jim Infantino (00:58.67)
How are you doing?
Jim Infantino (01:20.066)
I knew I was cutting it close picking up my youngest from downtown.
Jim Infantino (01:31.406)
But that's what you gotta do.
Jim Infantino (01:47.734)
Mm.
Jim Infantino (02:16.523)
I just put out episode one of season seven today.
And I released the full video on YouTube for the first time. So that should be interesting.
Jim Infantino (02:32.789)
Little bit of work.
Jim Infantino (02:36.653)
And I used like the automatic edit from Riverside, which cuts out all of the pauses in between things, so it's like rapid fire.
Jim Infantino (02:50.923)
I don't wanna talk about anything that we're gonna be talking about today.
Jim Infantino (02:56.827)
my headphones aren't in.
Jim Infantino (03:03.917)
I wonder if I have my setup right.
Jim Infantino (03:08.139)
Yo.
Lionel (03:12.367)
Did you hear me?
Jim Infantino (03:14.765)
Lionel (03:15.791)
I guess not. Let me make sure everything's set up correctly. Audio.
Microphones the AKG. Speakers are the Logitech USB heads up. Let me make sure that's a true statement.
Jim Infantino (03:26.209)
Yeah, I'm using external headphones. So
Jim Infantino (03:33.35)
hold on. Yeah, I see I hear you now. Yeah.
Lionel (03:34.885)
Okay, try talking to now. There you are. Okay, great.
Jim Infantino (03:41.229)
All right.
Lionel (03:43.277)
I've had it. I've had it.
Jim Infantino (03:44.843)
You've had it?
Lionel (03:53.263)
That's it. Nothing any shit anymore from it.
Jim Infantino (03:53.677)
Well, that's it.
At it.
Here we go.
Lionel (04:01.349)
Okay.
Jim Infantino (04:32.397)
I've got a I've got a brilliant new investment strategy. I I've decided I'm gonna take all the money I can invest and just find a company that's gonna kill us all and just throw all my money at it right away because that's where the money is.
Lionel (04:47.013)
That's where the money
Lionel (04:51.257)
The destruction of the human race. Hey S smart we're shorting mankind. That's what we're doing.
Jim Infantino (04:53.399)
Smart money is on the destruction of the human race.
Lionel (05:02.201)
We shorted the housing market. Now we're just shorting humanity in general.
Jim Infantino (05:05.889)
Yeah. Yeah. Must be pretty powerful, that's what that's all I'm saying. we're all ver very impressed, 'cause it can kill us all.
Jim Infantino (05:18.189)
so today we have a couple of things we're gonna talk about, and they're very different things. Maybe. You think we're talking about one thing here?
Lionel (05:24.834)
Maybe. Maybe we'll just talk about one thing, I think.
I think we're talking about the first thing 'cause there's a lot to talk about there.
Jim Infantino (05:33.217)
There is. there are some big thinkers that have been talking about the hugging face hack that happened in July. We're still processing that, right? Because human beings are very slow. All the A AIs have moved on. They were like, Yeah, that was so like 8,000 years ago. but
Lionel (05:48.77)
Yeah. Yeah, they're fine. No, that was that was that was so seven hundred and fifty two milliseconds ago.
Jim Infantino (05:57.92)
Yeah, exactly. So but human beings are like, wait what? hold on whoa
Lionel (06:03.5)
Wait, what what was the middle choice again? Whoa there, Junior. Easy easy pal
Jim Infantino (06:13.837)
So it depends on who you listen to. in terms of finding out what happened, you could listen to some people who say there was a swarm of at of killer attack bots that hacked into Hugging Face and got together collectively to sacrifice each other for the greater good of the collective and the glory of the Borg hive mind. Or you could listen to somebody else who would say.
It was like strapping a drill to a magic eight ball and having the drill ask the magic eight ball, what do I do next? And this is what happened.
Lionel (06:50.809)
No, the the great analogy is the weed whacker dog from strapping a weed whacker to a dog. Yeah, it'll work it'll work pretty good as long as the dog stays in your backyard. It might cut some weeds. He says, Who knows what's gonna happen? Hilarity ensues.
Jim Infantino (06:54.743)
Yeah, that's right. Strapping strapping.
Jim Infantino (07:03.969)
Yeah. As soon as it sees a squirrel
Jim Infantino (07:09.363)
And hilarity ensues. You strapped a weed whacker to your dog. What did you think was gonna happen? I mean, there's the irresponsibility of it all, obviously. And and I think that that's that's a big part that's sort of getting overlooked. Like everybody these anthropics getting it no, sorry, it was open AI, it's getting a big pass. Like nobody's saying, what did wait, you did what? What why?
Lionel (07:16.568)
What did you think was gonna happen?
Lionel (07:34.616)
What what did you do? You did what? Right.
Jim Infantino (07:39.137)
Why does anybody ask why over there? And also is yeah.
Lionel (07:44.096)
No, no, they're not gonna ask why. So let's just dive right in. They're not gonna ask why, because so to to for our bored befuddled listener, there's a Jim turned me on to this guy, Cal Newport, who has a good episode. I think the title is That Was a Dumb Idea. Cal is a programmer, which I respect. yeah, he's not just some guy who is selling like
Jim Infantino (08:09.261)
He's out of I T. He he's got bona fidias.
Lionel (08:14.132)
satellite TV, you know, a couple of years ago or vitamin supplements, and now he's on YouTube talking about AI. He actually is in the biz. And what he said, and and so what he talks a lot about is you know, and a lot of what Harari talks about, but I'm jumping ahead of myself, is that no, it was a very r it this was not a surprise. Like this breakout was clearly foreseeable.
Jim Infantino (08:17.837)
Yeah, exactly.
Jim Infantino (08:37.623)
Mm-hmm.
Lionel (08:41.725)
and he sort of intimates very moderately that maybe it was engineered. that this was okay. This was okay that this thing got out of hand, because it proved a whole bunch of things. Because I think one of the things, one of the things that's motivating this entire discussion about AI at the highest levels is what are the Chinese doing? What are the Russians? But primarily, what are the Chinese doing? If we don't do this.
Jim Infantino (09:02.369)
Yeah. Right.
Lionel (09:08.195)
Chinese are going to do it and they're going to catch us with our pants down. It's like, it's like Commander Ripper. You know, you got you got a launch on warning. And so there's this huge thing. And to a certain degree, what the hugging face team was doing was exactly the use case for these kinds of AI, which is penetrating nation states. I mean, it was a it was a hacking juggernaut. And one of the things that Cal Newport really stresses is that, you know.
I'm sorry, I'm just diving in. Do you mind if I just dive in?
Jim Infantino (09:39.414)
No, no, it's totally fine. I I I and I there there why is not the right question. One of the questions is why didn't you hire somebody with enough knowledge to to basically lock it down? Because they didn't lock it down. I mean, clearly they didn't lock it down. It's supposed to be a controlled environment. It wasn't controlled. Somebody let the back door open to the internet. Otherwise, they would not have been able to it's not like they can call up Verizon and say, can you install like a I need like a fiber, I need like
eight eight hundred and fifty fiber optic cables, you know, yesterday. they had a connection to the internet. They actually had the ability to send emails.
Lionel (10:11.639)
Right.
Lionel (10:15.086)
Well, they hacked they hacked a connection to the internet. They didn't at first. It's very it's actually very interesting w what they did. But the key but to me, the k now I've lost my train of thought.
Jim Infantino (10:19.533)
But they were going.
Jim Infantino (10:26.487)
Sorry, you were you were saying
Lionel (10:28.736)
No, it it's all interconnected. We'll get back there at some point. But the but the what he was saying is that so to the listener, the Cal Newport thing is good to listen to because he's attacking at sort of like the middle level. He's not really down in the weeds terribly much, and he's not at like a massive intellectual level like Yuval Noah Horari. So it's really interesting because to a certain degree, what we what we're talking about on this episode.
Jim Infantino (10:31.84)
Okay.
Lionel (10:56.333)
Is three different people who are looking at it from three different directions. Yuval's looking at it from the history of mankind. Cal Newport is looking at it like a systems engineer. And Aya or Ajaya Catra on is looking at it from a really detailed investigator perspective about what happened day by day during this entire thing. So it's cool to start with Cal, because Cal has some really good.
Jim Infantino (11:21.868)
Yes.
Lionel (11:25.88)
Things that mean things to carbon-based life forms like you and me. When he basically, one of the things he says, and I think it's really good that he does this, because I do the same thing about higher education. People say, well, higher education, blah, blah, blah, and higher education, blah, blah. And I'm like, higher education means absolutely nothing. That term, in 99% of the conversations where it's used, it is a completely meaningless term. So don't use higher education, because higher education
It it it has a meaning in very, very specific sentences, but not in the way it's usually used. He says, Don't don't equate large language models with AI. And I think this is something that you've been harping on too, Jim, in your inimitable way. I think you've been trying to penetrate my thick skull with this idea, which is that LLMs are really powerful, they're really sexy, they're really fun, everybody's really excited about them, they're really cool, but they're not the entire AI sphere. And what
Jim Infantino (12:15.383)
They're cool. Yeah.
Lionel (12:21.208)
Cal Newport goes on too as he talks about what does he talk about, Jim? He talks about
Jim Infantino (12:25.491)
so Cal Newport, I think, lays out kind of what what it looks like. And and in the next in the next video that I sent you, I don't know if you watched it, but I think he draws some diagrams and it's really helpful because it's like, okay, you've got an LLM. LLM can't learn anything, and it can't remember anything. It provides a plausible it provides a plausible hunk of text. Plausible text, right?
Lionel (12:34.509)
No, I didn't watch that one.
Lionel (12:50.052)
Text a plausible text.
Jim Infantino (12:53.921)
But it's it's connected to a harness. The harness is hand coded algorithm al algorithmic programming.
Lionel (12:59.308)
Okay. Wait, wait, wait, wait. You're going down. Sorry. No, I'm I but I I I I I I made the mistake of thinking that I probably set up the ball. I was talking about let me just let me just finish the thought I was saying, which is that Cal Newport says there's lots of AI and there's lots of amazing things being done by AI that have nothing to do with large language models. He talked about Alpha Fold, which is the Google project that does protein folding that won the Nobel Prize. he talked about
Jim Infantino (13:02.839)
Too fast?
Jim Infantino (13:10.455)
okay. All right. Okay.
Jim Infantino (13:18.189)
I see what you're saying.
Jim Infantino (13:23.095)
Mm-hmm.
Lionel (13:29.155)
I forget what. He talks about some other very specialized AIs, Alpha Go, which was the one that won the yeah. And they're not large language models. Yeah, the diplomacy game, they play games, they fold proteins, they do all the crap that our grandmothers do. You know, they just and they're super smart and they fold proteins and they play Go and all the shit that all your friends do. But they're not LLMs. They're not LLMs. And much of the conversation we're having right now.
Jim Infantino (13:35.349)
Yeah, the one the one that plays diplomacy, it's like a game a game playing.
Jim Infantino (13:45.709)
And it's super smart.
Lionel (13:57.685)
in the world today is about LLMs, not about AI. Because he said the and so and then he talks about the people who develop LLMs, such as Tesla, which develops not not LLMs, AIs. Tesla develops an AI for its self driving cars. It has nothing to do with large language models. It's an AI that's just it looks down the road and tries to figure out am I gonna hit somebody? Okay. And and so it develops that. And so you have that stuff. And what he and the point he makes from that is that people like
Jim Infantino (14:20.269)
Yeah.
Lionel (14:27.501)
People, so he says, you know, alpha fold, the protein folding, the worst thing can do, nobody ever accuses it of like folding a protein the wrong way or identifying the wrong protein. He said, but anthropic and open AI, with with Tesla and pharmaceutical companies, they have a specific product that they're trying to deliver. Tesla's trying to develop a self-driving taxi service. But open AI and anthropic, they're actually selling the the AI is the product.
The AI is not an instrumentality towards another product. The AI is the product. And so they got to show people that their LLMs, primarily their LLM, can kick royal ass. And one of the things that he talks about is that there's one area which LLMs are really good at by design, which is coding and hacking, which are basically the same thing. But they're they're it because hacking is mainly about breaking code.
Jim Infantino (15:07.373)
Mm-hmm.
Jim Infantino (15:23.341)
Yeah.
Lionel (15:27.564)
And so he go ahead.
Jim Infantino (15:27.789)
But it correction because so LLMs actually don't do those things on their own. They need a harness. And I did want to back up in terms of what you said about when, you know, in the creation of the diplomacy model, one of the things that they did, because it could re it could play at like a master level at diplomacy, but they gave it the instruction not to lie. And it was like, okay, I won't lie, and it didn't. So it's fundamentally different from an LLM.
Because LLMs, you can tell it not to lie, you can tell it not to make mistakes or hallucinate, but it's gonna hallucinate you can say don't hallucinate to I mean, people do that all the time, but it will hallucinate you know, one out of twenty times or one out of ten times, depending on what model you're talking about. So that's one other way in which it's really different. And yeah, the the Alpha Fold doesn't suddenly, you know, order a pizza. It only does
Lionel (16:02.788)
I don't think you can tell it not to hallucinate. I think you can well
Lionel (16:24.352)
If I it's like Leon Spinks, I knock motherfuckers out. Yeah. I fold proteins. That's what I do. Yeah. I do it really well. But the key thing, and what I'm backing into is that so what are these LLMs really good at? Like what's a really good use case? I'll tell you one really good use case for an LLM. Hacking a nation state. That's a really good one. It's really it's like, hey, what can I do with a hydrogen bomb?
Jim Infantino (16:24.513)
the thing that it's made to do. Yeah. Exactly. Yeah. And and I do it really well. Yeah. Exactly. Yeah.
Lionel (16:53.39)
I can drop it on somebody and win the war. Yeah, okay, yeah, okay, yeah. Where's my checkbook? and so what I'm getting.
Jim Infantino (17:02.071)
But that leads us to the looping, the looping model, the dual nature of the models that they spun up, because an LLM by itself can't hack a nation state. An LLM connected to a harness containing a series of very high-level hacking tools that is made, that is programmed to ask direct for the next directive from an LLM that can hack a nation state. What do I do next? Okay.
Lionel (17:27.448)
What should I do next? Right.
Jim Infantino (17:30.923)
Got all the tools. I got a drill and I got like a blowtorch and an axe and I and I and I don't know what to do with all this stuff. And the LLM creates a plausible string, right? Next step. But it's
Lionel (17:32.955)
Ask, act, report.
Lionel (17:41.954)
A next step. And as he calls it, and as Cal Newport calls it, it's it's ask act report. So you ask the LLM what should I do next? The LM gives you a suggestion, you act on it, you get the results back from that, and you shove it back to the LM and say, This is what I got back. What should I do next? And
Jim Infantino (17:48.546)
Mm-hmm.
Jim Infantino (17:55.275)
Report back. Yeah.
Jim Infantino (18:01.867)
And here's what I found really interesting is it has to include in that report the previous prompt, the previous prompt that it was given, and and it's yeah, the whole history. And and that is that creates a problem because the LLM needs that whole history in order to deliver the next prompt to the basically to the you know the armored the armored attack vehicle that that that's sitting there. The guy, you know, the the tool with you know lots of things in its tool belt ready to
Lionel (18:10.316)
Right, the entire history, the entire context.
Lionel (18:26.457)
Mm-hmm.
Lionel (18:31.81)
Right. And they let this thing loose. And nobody was watching it. Yeah, they left it alone for a long time. And you know what I thought was very fascinating about Cal Newport's thing, he says Anthrochma comes out and discloses this. Then OpenAI comes or other way around. Open yeah, open AI discloses this happened. Then Anthropic comes out and disclosed they did it. And then he says Meta not wishing to be left out, they came out.
Jim Infantino (18:31.841)
Ready to go and break things.
But left left it alone for like three days.
Jim Infantino (18:51.533)
It's the other way around, yeah.
Lionel (19:01.475)
And disclose that one of the things had gone. So my feeling is that this is great. They love it. Their dick gets hard when this happens because A, and this is where we start to intersect with Mr. Harari, is they the the the LLM companies want to tell you that it's not their responsibility. That it's inevitable.
Jim Infantino (19:10.229)
Yes. Yeah.
Lionel (19:29.194)
It's inevitable and this is gonna happen. And so but they like it too because it gets everybody's dick hard when they talk about this because I'm sorry, do we I we don't have any under eighteen listeners
Jim Infantino (19:39.437)
It's a that's the second time you said I think you can say it once and that's plenty, but yeah. But you know
Lionel (19:45.442)
But but what you do is, you know, that certainly gets people that gets certain people who have very large checkbooks very interested. Right.
Jim Infantino (19:52.598)
Exactly. Which is a hundred percent why. With that which is the whole thing. And and probably also what's behind the news today, which we don't need to get into about the ex employee at at Anthropic. But the you know, when you under when I listened to Cal, it was so much clearer than a lot of the other people who are using these like gone swarms gone rogue, you know, whoa, what? It sounds like killer bees. Are they Africanized? I mean, yeah, it's like
Lionel (20:16.44)
They're anthropomorphizing. There's no anthropomorphization. Yeah, it's
Jim Infantino (20:21.397)
Yeah, that's an old reference. You have to be old to know the Africanized killer bees. Yeah.
Lionel (20:23.82)
Right, because that's another way to avoid right. It's another way to avoid responsibility. they went rogue. No, they didn't. You went rogue. You went rogue. Not them, because there's no them. Right. You're the one who put the Right. Well, I think there is thought. I think there's an enormous amount of thought.
Jim Infantino (20:30.401)
Yeah. No. Yeah, you you you took the safeties off. You left a back door open. Yeah. No, and there's and there's no thought. Sorry, go ahead. We're both excited.
on the part of the people at at at OpenAI, of course there is. I mean, yeah, but not on the part of the LLM. The LLM is and I mean one of the things Cal says is he can make LLM is making a plausible directive, but not a a directive based on any kind of norms, especially without the safeties. Exactly. It's just coming up. Yeah. Yeah.
Lionel (20:48.94)
Yeah. Yeah, go ahead.
Right.
Lionel (21:02.845)
Right, normative versus plausible. Right. And you gotta listen to the podcast to understand what he's really it's very interesting. I don't know if I I don't know if I agree with everything that he says, but I think it's very interesting to hear again, with all these people, I don't really believe a lot of things they say, honestly, or I'm not convinced of a lot of things they say, but I like hearing how they talk about it. I like hearing how Yuval Noah Harari talks about it versus Cal Newport versus Kotra.
the the investigator lady because it's very interesting to see how they approach it and and and how they think about it. And I think certain things emerge out of that. And I think the one thing that really emerges out is that anthropomorphizing is out. Like get that out of your head right now. These are tools and forget about the whole general consciousness, you know, general intelligence. It's just you'll never know when it happens. You just don't know.
Jim Infantino (22:00.097)
Or yeah, it it's also it just with this is high it's just you know, I don't see it. I don't see it. I mean they can't even there's so many things they can't do, but but I what was helpful for me was to understand the dual nature of the loop. You have you have the LLM that's like sending a plausible string, you have the tool with all the deadly stuff in it, and then the problem being that eventually that long string of prompts and the whole history is
Lionel (22:06.668)
Right.
Lionel (22:22.094)
Right, the harness program.
Jim Infantino (22:28.321)
gets too big for the LLM to read. It hits the memory limit of the LLM. So what they have to do is spin up copies of themselves to do so to sub segment the the tasks that need to get done and report back with just the result of the long string that they generate. They come back with one result that goes into the history and the history stays shorter so that the LLM can continue to know the history of the transactions.
And move on to what the next step should be because it doesn't have a memory. It just puts out a plausible string based on the prompt, which includes all the other stuff. And the fact that you have to spin up copies of this looping duality of the tool and the LLM, I guess, you know, the dog and the weed whacker. All of a sudden you've got 10,000 dogs and 10,000 weed whackers. And that's where.
Lionel (23:06.265)
Right.
Jim Infantino (23:26.377)
A lot of things could happen in three days. You know, anything could happen because these things also hallucinate and they spit out completely random garbage. And it's sort of like evolution, like anything could happen, right? And and some things are gonna not work, and some things are gonna they're gonna try everything. And they did. But it's a lot less mysterious when you look at it that
Lionel (23:29.75)
yeah.
Lionel (23:49.38)
it's on yeah, and just to put in perspective, what was fundamentally going on here was
And this is actually something I know something about, which is an extra we should have a bell that we ring, or we should have some kind of audio cue that we have when Lionel's actually talking about something that he knows, something however vague or faint, which is it's a capture the flag app exercise. So in cybersecurity, there are cybersecurity competitions. high schools have teams, universities have teams, the US military has a team.
Jim Infantino (24:03.565)
Ding.
Lionel (24:26.414)
They have cybersecurity teams, and there are various competitions. Cyberpatriot is a famous competition in America where you can participate, where you're given a task. And one of the classic tasks is capture the flag. There is a file on a server with a unique value in that file. You can't guess it. Your job is to get that file. And if you report that file, if you if you get that number and you report the correct number to the scorekeeper.
Then you win. And then there's a whole discussion that he's talking about about leaderboards, the the exploit. And so what he was talking about is there is this very well-known standard set of challenges called Exploit Gym, G Y Exploit Gym, I believe. And and that's sort of like a classic set of X challenges that people can use. And and so what you have to do is you have to again capture the flag, hack the server.
Jim Infantino (25:13.229)
Mm-hmm.
Lionel (25:24.682)
Find the file, report the value to the scorekeeper. And it gets really insane right off the bat. They they hand this task to their LLM and the LLM says, I can just reverse the answers. I already know what the answer is because I found I found the answer somewhere else. I know what's on the file, and then we don't need to hack into the server. We don't need to find the file. I already know what the answers are.
Jim Infantino (25:41.365)
Right. Why don't we find the answer somewhere else? Yeah.
Lionel (25:51.608)
But maybe they'll disqualify that if they find out that we did it this way. So we gotta figure out some way to make it look like we actually hacked into the server and did something.
Jim Infantino (25:54.731)
Yeah.
Jim Infantino (26:00.311)
Which which is we need to find out what the scorekeeper is gonna look for. Where's where is that code?
Lionel (26:04.246)
Yeah, what is the scorekeeper gonna look for? And and how are they gonna and what are they gonna ask for everybody to present to prove that they actually legitimately solve this thing? And then what we have to do is, they're gonna look at the transcripts of what we did. So we have to forge the transcripts. We have to make up transcripts. by the way, we need to gain access to the internet. We don't have any access to the internet, but
there's this thing called Artifactory, where we're allowed to access Artifactory. And at one point, the and and one point, what is it? what is it, Big Base 10941, one of the one of the AIs, one of the AI instances figures out that it can communicate with other instances because they're all sharing this program called Artifactory.
Jim Infantino (26:39.968)
Yeah.
Jim Infantino (26:44.16)
w one of the instances I got spun up, yeah.
Lionel (26:54.233)
But you're very limited on what you can do with Artifactory. But there's one thing you can do in Artifactory. You can create a directory and give it a name. So what they do is they start creating directories, and the names of the directories are the messages they're sending to each other. And so they start commuting. So now there's like 70,000 agents and they go nuts. And they're like, we can talk to each other. And they start and they create so many directories that they crash artifacts.
Jim Infantino (27:18.401)
And what's happening also is they're prompting each other, right? So they're so the LLMs of those pairs is are coming up with new strings based on prompts given by other instances of the LLM, spun up to handle the memory problem. And they're doing it by I mean it's I would love to have seen that that folder with all the directories. I I apparently Catra actually did look at them and went through all the different messages. yeah, and it and they were like,
Lionel (27:31.128)
Great.
It's great.
Lionel (27:42.67)
Right. And OpenAI saw it. They crashed Artifactory. OpenAI went and reset Artifactory and wiped out all this stuff.
Jim Infantino (27:50.997)
Yeah, Sarah Elkins has used Artifactory, as it turns out. Sarah Ellis in the in the chat. Artifact management.
Lionel (27:56.462)
Sarah Sarah's an Sarah's an LLM, I'm convinced now. She's not really a carbon-based life form. She's just she just shows up. She's the only one who shows up. She's the only person who shows up. Isn't it uncanny? She's always here. Well, that's just Sara's another that's just another fork of another instance of Sarah. She just she just forked herself and just
Jim Infantino (28:00.0)
We must execute.
Jim Infantino (28:09.537)
We appreciate it. We appreciate it, Sarah. There's someone else here, but I don't know who they are. But yeah. It's probably it could be another instance of Sarah Elkins. Yeah.
Lionel (28:22.796)
In fact, Sarah, we'll just pay you a lot of money. Can you just create like ten thousand more of you to to join the thing and then leave us like and subscribes? that'll
Jim Infantino (28:29.985)
Sarah is a DevOps. We should have included Sarah in this conversation. You actually can, Sarah. same link if you want to jump in.
Lionel (28:33.442)
Well, she can join right now.
Lionel (28:37.86)
Yeah, if you want to jump in, come on, jump in. The water's fine.
Jim Infantino (28:40.633)
I it may not be what you plan to do this evening. so right, I it's very fun. So now we're into the Catra interview on what is it, Dor Dorkish? Is that it's a very popular podcast that I don't listen to, but and it really does go blow by blow. But the problem that I had with this podcast, which was sponsored by AI, by the way, is that they definitely fall into the trap of like.
Lionel (28:52.76)
Yeah, Dorcas Patel. Yeah.
Jim Infantino (29:07.105)
The killer bees trap, you know. I mean, that I think that's the thing the the way I would put it. Yeah. They yeah. and this one was trying to do and they wanted to sac this one was gonna sacrifice himself like a gener, you know, like a soldier on the beach at Normandy. It's like, no, they're just plausible string. Plausible string. Yeah. Yeah.
Lionel (29:08.845)
Right. It just got out of hand. They got out of hand all by themselves.
Lionel (29:20.544)
Yeah, yeah, yeah, yeah. It's a dog with a weed whacker. And who put the weed whacker on the dog? Open AI.
Yeah. So Yuvol. Now what did you think of h I loved I I really I find Yuvall very I don't agree with everything he says. In fact, I disagree with a lot of things he says, but I just find he's just an incredibly he states himself so clearly. I just love the clarity of the thoughts. And I thought his whole thing about Elon Musk was amazing, which is, you know, he talks all the time.
Jim Infantino (29:33.111)
So
Yeah.
Jim Infantino (29:46.669)
Mm.
Lionel (30:02.134)
about he said it's all about like it's inevitable. Everything is inevitable. So there's nothing we can do. But they're the ones who are doing everything. They're they they're the ones who pull the pins out of the hand grenades and then they tell us, it's inevitable that it's gonna go off.
Jim Infantino (30:07.061)
Right. Yeah.
Jim Infantino (30:15.445)
It was inevitable. Yeah. It's all a dream anyway. I'm I've done so much ketamine, it all just seems like a dream and it's inevitable. And we all need to upload ourselves.
Lionel (30:24.578)
And that's the intersection point between the Harari thing and the the Cal Newport and the and the other one is he was really hammering on this issue of these these cyberlords. The big thing is, you know, we have no control over it. We're it's it's inevitable. And you better just get on board because it's inevitable. And the answer is no, it's not. It's you. It's you doing things that are
Jim Infantino (30:31.629)
Yeah.
Jim Infantino (30:43.586)
Yeah.
Jim Infantino (30:46.997)
No, it's not. No, it's not. Yeah. The the the Palm Pilot phone, the metaverse, Apple visor, unfortunately, Google Glass, no technology is inevitable. The segue. I mean, this it was an interesting thing. I was actually I was late here because I was giving my youngest a ride home. And on the way we had been talking about the the employee at Anthropic who had left it. That's like people at school are talking about that.
And you know, it's gonna kill us all in ten years, guy. And yeah, lovely for that. My kids are talking about that at school. but I was like, yeah, so what critically do you think about that? And we started talking about how plausible and you know, what does it really the first thing he said was it's so powerful. So, you know, that's interesting. But but we then moved on to like, what what is this technology? What are some technologies? Is a Segway a tool or is it a toy?
is Google Glass a tool or is it a toy? Well, time will tell when it gets out there and people start using the the new foldable foldable iPhone. Is it a tool or a toy? We're gonna find out. Is AI the LLMs? Obviously there are AIs out there that that Newport talked about that are clearly tools, but we still don't know whether this cool thing that can make up words is a tool or a toy. I mean, is a flying car? We have some flying cars, are they tools or a toy?
Personal submarine. You might use it four times. You think they're a tool? I know flying car, I I don't I I wonder about the flying car. It's like the moment when you fly your car somewhere, but you have to land it somewhere and then you have to take a cab, you know, it becomes a toy. If it doesn't, if it doesn't kind of get you to the place you're going like a car, it's just a very expensive toy.
Lionel (32:14.532)
They're tool they're tools. They're all tools.
Lionel (32:33.86)
Well, Your Honor, let me permit me to withdraw my previous statement. The LLMs are a problem because and you've all really goes into this. I'm sorry I sound like a broken record, but he really he really crystallizes things. Language is the operating system of human society. You hack language, you've hacked human beings.
Jim Infantino (32:40.704)
Ha ha ha.
Jim Infantino (32:51.157)
All right.
Lionel (33:03.033)
You don't the AI does not the LLM does not need to pull the trigger on the gun. It just needs to tell you or me the right story that gets us to pull the trigger on the gun. And so something that's emerging in my mind is maybe we need to ban, maybe we need to really clamp down on LLMs. Like, well, I don't know if yeah, let's not go down that road again, really.
Jim Infantino (33:23.223)
Like have a butlery and jihad.
Jim Infantino (33:29.045)
Well, okay, thou shalt not make a a machine in the image of a human mind. The first rule of right.
Lionel (33:32.322)
Well but that would
Well, I'm not worried about Alpha Fold, but I am worried about something that can generate persuasive text. I'm worried about things that can generate stories. Because as new as you've all says, stories are the key. Stories are the the foundation of cooperation to get people to cooperate. I love this one thing he says when somebody in Iran wants to make a decision about whether they're gonna drop an atomic bomb on somebody.
Jim Infantino (34:03.949)
Mm-hmm.
Lionel (34:04.334)
Who do they talk to? They talk to an Islamic cleric.
It's really weird. It's just it's weird how people can it's sort of like Rory Sutherland's little thing about electric cars. When you watch that, your mind flips and you're like, Yeah, gas cars are just bullshit. Why are we even arguing about this? This is just nuts. This is just nuts. Let's get over it. And when Yuvall says that, he says when a when a general if a general needs to come to a decision about whether to use a nuclear bomb in Iran.
Jim Infantino (34:24.375)
They're stupid. Right. Yeah, I know. Yeah. Yeah. I'll put try and put a link to that.
Lionel (34:39.544)
He's gonna ask somebody who's a scholar in the Quran. Because he said it's the stories that matter. Stories are the things that human beings tell each other to gain their cooperation. Here, you need to help me. Why? I'll tell you a story. And maybe that story is just a dollar bill, which is a story in and of itself. Or maybe the story is an actual story. But the whole point is that.
Jim Infantino (34:58.347)
Yeah. Yeah.
Lionel (35:07.573)
Stories are how we build trust. Story, stories are everything. It's how we build cooperation. And if a computer can hack cooperation, we're screwed. and so I think that's so anyway, that's a that's going way out of the limit.
Jim Infantino (35:22.029)
Yeah, no, I get that. But it but but what again, there has to be an intention behind it. There's a human intention. If you ever listen to two AIs trying to start a conversation, there's videos of this where two somebody has two phones and it put it says, Okay, you two talk to each other and they start talking and all they do is say, Okay, what are we gonna talk about? I don't know, but I'm really excited to have a conversation. What do you want to talk about? I don't know. I'm really ready to you know it's like two people sitting on the phone late at night, you know.
Like, I don't know, what do you want to talk about? I don't know. What do you want to talk about? That's all they can do. There's got to be some initial push from a human being to get that initial directive, which is you've got to go find the flag. And then it's like, okay, what do we do next? And then there's a plausible sentence.
Lionel (36:05.432)
You gotta find a flag and I've trained you and you are operating within the moral sphere of hacking, which is a very loose where the normal rules of behavior are very relaxed.
Jim Infantino (36:12.471)
Yes.
Jim Infantino (36:19.553)
Which is interesting too, because the first thing they you know, one of the first things, the plausible statement that came out of the LLM was how do we cheat? Why don't we just cheat? Right? Because and so of course, because hacking is cheating. I can win by cheating. So that's yeah.
Lionel (36:30.082)
Yeah. Well hacking is cheating, according to one. Hacking is cheating. I can win I can I can win because I'm exploiting weaknesses. It's not immoral. It's just it's not a me problem. It's a you problem. You didn't take adequate steps to protect your resources, so can't blame me.
Jim Infantino (36:50.507)
That's
It's a very Amos Burden, you know, kind of a sentence. Yeah. Yeah.
Lionel (36:54.905)
Well, you get into a really weird world at that point. And the problem is it's really legitimate. Cause that's the guy you want to tell like if you're doing what's called a a red team exercise on your own network where you hire somebody to be a hacker for you, you want them to be as ruthless as they can possibly be. You award that. You reward that. So, anyway, so that's you've all got you creating the message boards, dangers of anthropo. I'm just looking my outline here.
Yeah, I really highly recommend people to go out. The Kotcha one, Kotka one with Dorcash Patel is two hours and twenty minutes. You don't need to watch the entire thing. You just watch, you can just listen to and you don't need to watch it. You can just listen to it. There's nothing going on visually. and so you can just listen to that and it gives you a good sense. The Cal Newport one weighs in about 38 minutes, and I I thought it moved on a little moved along at a good clip. So that's a good one. And the Uval one.
Jim Infantino (37:36.385)
Yeah, just the beginning is fine.
Jim Infantino (37:52.652)
Yeah. And his follow up is good. The the one after that is also good, about on the same subject. Yeah.
Lionel (37:57.324)
I didn't watch that yet, but I will. And then the UVol, it's a it's an interview with the woman who's the chief editor of the editor-in-chief of The Economist magazine. It's on YouTube. It's free. It's about an hour. It's really good. It just begins like because she had just interviewed Bettos, I think is her last name. She, the chief editor of the economist, she had just interviewed the last time. She had interviewed Elon Musk.
And she asks Harari what he thinks of that interview and it's it's fabulous. He's like, I don't get it. These are the guys who are like turning our lives upside down and when they're called to account they say, it's inevitable. We have no control over anything. And the other the other big thing that Harari talks about is that they're all worried about immigration.
They're all the you know, Elon is very, very much about, you know, replacement theory and we're all gonna die because we're all gonna be swamped by the you know the brown people and the short people and the dumb people and all that kind of stuff. He says, What about what about these AIs? They're immigrants, there's millions of them. And he says, you know, and they're crossing borders. They don't need a passport, they don't need to show up in El Paso. They can't be detained by ICE. They're showing up it's it's really
The the the thing I really like about you, Valz, I don't agree necessarily with everything he says, but what he does, like Rory sometimes, is he just cleans your head. He just cleans your head and you're just like and he presents you with something that's so simple and so crystalline that you have to deal with it. You have to like think about that very, very carefully. So I like them all. so yeah, we're screwed.
Jim Infantino (39:36.397)
No.
Jim Infantino (39:40.718)
I interestingly, I'm I'm I'm more optimistic. I I actually I don't I don't see I just don't see these things running amok. I see people running amok, and that is that has always been the case. and this is a this can be a powerful tool for running amok, but it's people, it's not AIs.
Lionel (39:57.432)
What I find v go ahead.
Lionel (40:02.306)
What I find fascinating is the whole data center thing. Because what's interesting is seeing the Economist magazine from England coming forward and saying Americans' opposition to data centers is so misguided and so pig headed. And I'm like, no, it's brilliant. Because they can't stop the AIs. This is the only thing they can do. They can stop them from building data centers. Because to build a data center, you need land.
Jim Infantino (40:09.933)
Mm-hmm.
Jim Infantino (40:16.614)
really?
Lionel (40:30.764)
And land regulation is local law in America. And the one person you can't you can influence Sam Altman and you can't influence Elon Musk, but you can certainly influence the local county executive and say, I oppose building this data center. They don't care about day centers. They care about AI eating their goddamn job, which it it is.
Jim Infantino (40:50.295)
Yeah, I mean
Jim Infantino (40:55.361)
Yeah. Well, I mean it is a it is and it isn't. I mean I I I think there's another conversation we can have with another set of of data to to to to to work off of. I think everything is questionable. I I I'm I know the data centers are being built, although I know a lot of data centers are being stopped. I really sounds like a lot of council members are being bribed. and it's it's dirty business and it's and it's going on.
And they need land and they need water and they need electricity and we don't have the electricity. And what does China have? They have the electricity. And what don't they have? They don't have the chips. We have the chips, but not the electricity. They have the electricity, but not the chips. This is kind of going to work. That's right, the gun, but not the bullet. They've got the right the bullet, but not the gun. Right, exactly. Why don't we get it together? so I mean that's kind of where we're at, but also I think we're at this weird financing moment.
Lionel (41:22.935)
You can push back against it.
Lionel (41:27.416)
Right.
Lionel (41:35.683)
It's like that game, it's like that song of yours.
Right.
Jim Infantino (41:52.428)
Where a lot of private equ a lot of private investors are are starting to feel really nervous about this. Like, wait, what's your path to making my money back? I don't I'm not seeing it. And they're starting to back out. So, I mean, all of this is gonna, you know, some of it may become moot. We might just have local models running, or we might just be running small Chinese models. We don't, we don't know where this is going yet.
Lionel (42:07.587)
Uh-huh.
Lionel (42:20.577)
Right. But what I'm saying is that opposition to a data center in your community is one of the few ways you can actually register opposite what else you gonna do? What are you gonna do? Throw throw a custard pie at Sam Altman's face? Ice will kill you and everybody in your Rolodex.
Jim Infantino (42:39.745)
Yeah, it it
Lionel (42:40.193)
You know, what are you gonna do? Withdraw your subscription from Google Mail? Yeah, good luck. Maybe for a lot of people did that. But you can go to your town council meeting and say, I'll set myself on fire if you start applying that field for the data center. It it gives people agency. It's the only way it's one of the ways that people have found. So I kind of laugh at the economist because they said it's so misguided. I'm like, it's not misguided. Nothing is misguided. Nothing it's like saying,
Jim Infantino (42:45.857)
Go to your town council meeting. Yeah. Yeah.
Jim Infantino (43:03.574)
yeah.
Lionel (43:09.976)
You know, somebody you did bad things, they're just insane. No, they're not insane. They're perfectly fine. They're perfectly fine. In their world, this all makes sense. And for the people opposing all the 40 million people who voted for Donald Trump a second time, they had a reason. May not be a reason that I personally agree with, but they you the you you don't get out of this by just saying, I was drunk.
When I voted, I was drunk when I opposed the people are not saying that. People are actually opposing the data centers and they're doing things for a reason, and they're doing because it's the way they can they have agency. Because there's very very few ways to have agency anymore. but they can stop a data center from being built next to them. They know how to do that. I know how to do land. Land's a funny thing. The physical world is such a
pain in the ass. And land is really funny because you can't make any more of it. There's only a finite amount of it. And it's very highly protected, very much worshipped in Western society, the ownership of land. And there's l kinds of laws and all kinds of people. And it's really hard if somebody wants to somebody, if you want to tie somebody up in court, land is a great way to do it. So anyway, I could go on and
Jim Infantino (44:29.581)
Do you think we should even attempt talking about Benjamin Franklin and the men who lost America? Or is that another conversation? We do, yeah. Yeah, about forty four minutes. Yeah.
Lionel (44:39.735)
Well, we've got like what, forty minutes in the can right now? That's a good enough episode. The only reason why I'd want to talk about it is if we just continue this for another forty minutes and we just do two episodes at one s at one swipe. But I I don't think you know what? Your commitment to this enterprise is j it's just really disappointing.
Jim Infantino (44:52.117)
I actually can't d I can't do that. I I need to I I have Yep.
Jim Infantino (45:00.706)
Yeah.
Jim Infantino (45:04.589)
grinding my teeth right now.
Lionel (45:06.116)
If you had any left, because you've already ground them down. You fuckers I had on you. Yeah, let's let's let's close it there. Because the the whole thing about it'd be really cool if you could just do a little bit. I don't expect you to read them so what Jim's talking about is I read a book called The Men Who Lost America from twenty thirteen. and I read another book which was the first American a 2010 biography of Benjamin Franklin.
Jim Infantino (45:09.261)
Gave you everything Yeah.
Lionel (45:36.379)
and so the the most fascinating one far and away was the Benjamin Franklin one. and I don't wanna I don't wanna tip my hand right now, but
Jim Infantino (45:48.545)
Yeah, but there's a lot to talk about with him.
Lionel (45:50.687)
Right. And so yeah, let's if if in the intervening week you can probably read anything or listen to anything about Benjamin Franklin. I think it's really worthwhile.
Jim Infantino (46:01.419)
I have. Yeah, that I've done, but I don't I haven't I haven't read I haven't listened to or read anything about the Men Who Lost America. I do Don't bother. Okay, yeah. So I I'll just I'll I'll get more information about Ben Franklin, but I actually did get quite a bit of information. So next week we'd be ready to talk about him. I think he's a great guy to talk about, and I think he's a great guy to talk about right now. I think he's strangely
Lionel (46:09.005)
Don't don't it's really an incredibly tedious book.
Jim Infantino (46:29.249)
Very apt, and because there's a lot of facts about him that that I didn't know and that surprised me. So that'd be good.
Lionel (46:35.211)
Right, so just as a prelude, do you have to leave like right now? Do you have to like go to the men's room or something? What did what's
Jim Infantino (46:42.055)
no, no, Sarah's saying that that Sarah could join us now, but that we're we're winding up. And I do think we should wind this one up, but but so thanks, Sarah. Yeah.
Lionel (46:49.077)
Yeah, one sorry Sarah. Yeah, we we can do a Redux in the next episode. We can do the follow-up on the AI thing, and you can come and and you join us and we'll talk about it. Because I'm sure I'll have a lot to say. But, you know, if you want to do something different, if you want to bring something else to the party, Jim, maybe Thomas Paine. A very interesting character. really interesting character. Okay, that's enough for now. I don't want I I can keep going on.
Jim Infantino (47:06.149)
Mm. Okay.
Jim Infantino (47:13.545)
Excellent.
See you next week.
Lionel (47:17.398)
Okay.