Shawn Cordner (00:13.07)
Hey everyone and welcome to a special edition of the Amplitude of Tech Podcast. I’m Shawn Cordner, Chief Marketing Officer of Amplix. Boy, it was a busy week in AI last week. So I got Scott Weiner, who’s an expert in AI, to talk to me today about what’s been going on in the news. Is there really a 10% chance that AI is going to end the world? Is AI going to take over the world in the next six months? What the hell is going on? Scott helps me make sense of this. Hope you enjoy this episode as much as I do.

Shawn Cordner (00:48.846)
All Scott Weiner, thanks for joining us for a special edition of the podcast.

Scott Weiner (00:53.57)
Thank you. Good to be here.

Shawn Cordner (00:55.246)
Appreciate you jumping on last minute. This seemed urgent from everything that I was seeing on the headlines. I’ve heard that we’ve had the most consequential week in AI. Might also be the most overhyped week of AI history as well. So thought maybe we would hop on a quick podcast and see if we could talk through what’s real and what is fiction. So I f I guess the best place to start off would be the Anthropic employee that has left the company and has been

Talking about how there’s a ten percent chance that AI is gonna kill us and that in six months it’s gonna take over the world. what’s your view on what’s happening there?

Scott Weiner (01:35.16)
Well, first of all, that kind of prediction’s been around for a long time. It’s not new. And I I I always wonder when someone gives a percentage like that, like what what measurement scale are they using to calculate that ten percent? You know? And they never seem to answer that question. I I think look, I think that there’s some real concerns and they’re they’re valid about a a very powerful tool and what it’s capable of. I think where I

Partways a little bit with the kind of the characterization I’ve seen both in the media and also from him, frankly, is this idea of characterizing it as an alien mind. I think when they start to talk about it like it’s an alien mind, they’re they’re doing two things that I think are a disservice to the public. First of all, and it’s not intentional, I just think this is the result. I think that they’re they’re taking away the accountability of the human by saying this is an alien and we can’t control it. We don’t understand it.

I think that any good AI policy or process has to start with the assumption that the human is accountable for every part of it. And in fact, if you look at what actually happened with some of the recent events, they all trace back to human error. So I don’t while they are amazing, I don’t think that they are the catastrophic end of life. The other thing is I want to make a very big important distinction. And by the way, this is a distinction without a difference for a lot of people, but it’s important. There’s a difference between an existential threat to humanity and the potential for a catastrophic

problem happening because of AI. Those are two very different things. The chance that AI causes some big problem, again, AI controlled by humans making mistakes or deliberate is high. It’s very possible. It’s a powerful tool and it can be used by people for good or bad. The chances of it wiping out humanity, I don’t buy the template

Shawn Cordner (03:18.796)
By it either. I think that you know what we’re hearing is heavy in hyperbole and conjecture and fantastic extrapolation. And it’s really light on details and mechanisms and facts, right? So I it it makes me think that there’s a little bit of puffery involved here. I don’t want to assign any kind of intention behind it, but I have a hard time believing that this is just somebody that’s deeply concerned about.

the future of humanity. And that’s why he’s going out and speaking out against this. I think that also if you dig below the surface of the headline a little bit, and this is going to be a bit of a theme on today’s conversation is what’s below the head headline. Yes. What they keep saying is if we implement recursive self-improvement into AI, which would be AI teaching AI, then it could be a risk. And that could mean that

It will be the end of humanity. I mean, those are some really big ifs and some really big leaps, don’t you think?

Scott Weiner (04:21.494)
Yeah, and in fact the what I said before, I think I is even doubly so here is that the accountability s has to stay with the human. I think every policy, every law, every every way we look at this should be human driven. Which that means is if I have self-improving AI, where is the human in that process? Because if you don’t do that, you’re not being accountable. And I I I heard this one analogy I really like is that these things are like pinball machines. You send them out there, you give them a goal.

And they just bounce off everything until they get the answer and they’re relentless. So if you’re gonna let them do that and you’re not gonna watch them, yeah, eventually they’re gonna hit the hit the target. You know, so that’s that’s what I think’s really going on right now.

Shawn Cordner (05:01.632)
It’s kind of a law of big numbers also, right?

Scott Weiner (05:05.07)
Exactly. Yeah, exactly. You do you do enough variation. You know, one of the things on the the the the open AI issue with hugging face, that that obviously is a big piece of the story. The the kind of the headline was these AI tools coordinated and and had an attack with twelve hundred agents and the reality is if you go back and you you’ll see it’s a chain of human errors. They deliberately left guardrails open, somebody left their password on the internet, you know, I mean it’s like

It’s a whole bunch of mistakes that were made by humans, not by computers. Now, yes, it found the vulnerabilities and attacked them, but it was relentless and nonstop. The twelve hundred agents sounds really scary. Like all these agents started communicating. But here’s the thing. It was really just one AI with twelve hundred instances of the same AI with the same algorithms, with the same processes, the same training. it wasn’t like a whole bunch of different people getting together and planning. So it it it definitely got overhyped in that way. On the other hand, it’s still an incredible feat.

I mean, it’s still, you know, there there’s something to be worried about. It’s just not at the the level they’re putting. Now, getting into the theme of under the covers of this, we have some IPOs coming up. We also have a bunch of, you know, nervous politicians. And so, you know, these these make for interesting bedfellows. And so some of the conversations you can see from the the policies that are getting framed up right now, some of them are like, ban all AI, superintelligence is evil. And

And then you have other ones that are trying to do like, w let’s monitor, let’s watch what’s happening. And so you get them all over the place. What I would watch is the the providers seem to be leaning towards regulatory capture. They’re they’re going towards planning that looks like we’ve got this. We don’t need oversight. And and the problem is then the way they’re doing it is they’re saying you have to put incredible amounts of resources into the monitoring. And what they’re really saying is only us big guys can do it.

Shawn Cordner (07:00.034)
Yeah, absolutely. There’s so much to dig into there that you just touched on, but I want to go back a little bit. So you mentioned Hucking Face. I think that’s another example of you need to look b below the headline a little bit, right? Because to your point, this is not rogue AI agents that escaped a sandbox without yeah, you know, you know, a human being telling them to do it. They were instructed to do this and humans made mistakes, right? So I feel like

What they’re doing right now is meant to find these kind of vulnerabilities and exposures and that’s what they’re testing and that’s what they’re learning from. Did they make stupid mistakes? Was some of it intentional? Who knows? Again, I don’t wanna, you know, ascribe intention to anybody, but certainly this was not a case of these things somehow just managed to escape, you know, anthropic’s network and got get out and cause trouble on their own, right?

Scott Weiner (07:55.256)
They were they were given a puzzle that had no answer. And they were put in a situation where their only mathematical reward was to find an answer. And they found a hole. And by the way, the whole story, what happened is actually amazing. I mean, actually, the whole it’s really an incredible story. But at the end of the day, all that really happened is they had a they had a starving rat in a maze and the rat chewed through the door because it was starving. And it found it found a way out. So yeah, if you put enough pressure on the system and let it run and run and run, it’ll get out.

The the the part of the story that’s I think both compelling and impressive and at the same time a little scary is how fast, as soon as it got out, how fast it was able to kind of put the attack together. That’s that’s the part of the story that I think is real and is if something gets out, we don’t have a lot of time to react. We have to be ready for for it for these breakouts.

Shawn Cordner (08:45.078)
And and I thought meeting up on a a German bulletin board that was no longer active and communicating on there, that’s a pretty novel and interesting solution that they came up with. So, you know, we should stipulate this is all really amazing. It it’s incredible. And there is real danger here, right? So we’re not saying that there isn’t. We’re not saying ignore all this, right? What we’re saying is let’s kind of temper our expectations of how crazy this really is, right?

And and really what we’re hearing in the media versus the reality. Another one is the Ruby Gem attacks. And that one is a case where the headlines were that user information was being stolen. But that’s actually not really what happened, right? It was more about API keys that were being collected, right? There was no, I don’t think, any evidence of any actual user data being stolen, is that right?

Scott Weiner (09:38.134)
Not that I saw, but that doesn’t mean it couldn’t. But I think I think the reality is I forget this I’m sorry, I don’t have the stat on top of my head. There was a stat about the number of cybersecurity incidents that are involving AI now is, you know, getting very, very large. And every time AI is involved in a cybersecurity threat, the cost is going up. So it’s a much more expensive problem when AI is involved. So there are bad actors that are going to use these tools, but here’s the thing they don’t need the frontier models to do this stuff. So

All of this talk about regulating the frontier models doesn’t change the fact that there’s a powerful tool that is out in the world and code doesn’t wanna be limited. It’s it’s it’s out there. Yeah. So

Shawn Cordner (10:17.294)
I think the contrast that you see though is that you’ve got one crowd talking about existential risk and then the reality of what’s actually happening where the risks really are, it’s a lot more mundane, it’s a lot less glamorous, and they are really here. And they’re real dangerous risks and there is going to be a cost to society and to businesses, but it’s probably not AI taking over and killing us.

Scott Weiner (10:42.06)
Right. But it also talks to the fact that, you know, slowing down, which is what a lot of people are saying now. I don’t even know what slowing down looks like, even if we could measure it, because that just means that the bad actors have time to build up. It it doesn’t really stop anything. It just it just makes it harder to protect us if we’re not if we’re not on the leading edge. Th this gets into the bigger picture though. I think if you break this down you have you have international issues, you have n nation level issues. Like what

What’s a nation state going to do with AI? That’s one set of conversations. What are bad actors going to do? And then what is just accidents going to cause? Most of the bad stuff that’s happening is because people do not have good cybersecurity in place to begin with. They don’t have good governance in b in place to begin with. I was I was speaking at a conference the other day about cost of AI. It was just all about cost. And one of the things that brought up was that when I say governance, everyone hears stoplight.

Y’all hear about slowing down. And what I don’t think they really recognize a lot of people is governance means getting an efficient stoplight that makes traffic flow. It’s not the 4 a.m. sitting at the red light and it just won’t turn and you’re going crazy because there’s no one else there. That’s not governance. Governance is seeing the flow of traffic all around and figuring out how to let it flow faster. It’s the green light that follows that stoplight. That’s what governance looks like when it’s done well. We’re we’re not putting enough

emphasis on that with the companies. We’re letting them off the hook way too much and we’re saying things like, it’s an alien. Yeah. And it’s that’s not good.

Shawn Cordner (12:10.648)
So I live in New Jersey and we have circles here, traffic circles everywhere. and jar handles. And there’s a circle right by my house. And going into the circle from both directions, there is a light, and it’s only on during rush hour. So if it’s a busy time but it’s not during the time when the light is on, that circle gets really, really jammed up. It gets really dangerous. Traffic gets backed way up. But

During rush hour traffic, when that light comes on, it stops traffic from entering the circle from one direction for a period of time so people can get through the circle and it keeps the flow of traffic moving. And that’s what you’re arguing arguing for here is governance as a way to pace it and keep it flowing as opposed to hurry up, stop, deal with a problem, start over, that sort of thing, right? Yeah. So speaking of

regulations and accountability and making sure that there are some sort of controls over t this type of technology. We’ve heard Altman Musk and Amade talk about slowing down. And it seems like they’ve come to an agreement that they should slow down. I heard this morning China kind of criticizing that. And I want to dig into the the whole dynamic of, you know, the US versus China and kind of the arms race piece of this, but

What you have right now is you’ve got the private companies talking about slowing down and coordinating with each other. But then you have the Trump administration and Trump himself came out and said, No, no regulation. We don’t need to do that. All we need is, you know, a a strong president to control all of this. So it seems like we’re not going to get any kind of regulation at the federal level. I know that the the state level there’s a lot of patchwork type of regulations, but given what you’re seeing and the tenor and and really the polit

politicization of AI and data centers right now with the midterms coming up. You alluded to it. It’s a huge thing in the news right now and it’s a it’s very much a populist issue in our politics in the US. So with all those different ingredients, what do you think is actually going to happen with the regulatory picture in the next, you know, I don’t know, six months?

Scott Weiner (14:21.63)
f in the next six months I I I I I have a hard time seeing anything. I mean maybe they they pass something. I don’t know. I it depends on how high the panic level really gets, but I think the reality is that the stop it all c crowd is not going to win. and I think that a lot of it is kind of toothless. But there are things in the middle and some of the things that were recommended by the by the the big players actually are reasonable. I just don’t think they’re the whole picture. I think that they discount the non frontier work

almost completely in their plans, which is I think problematic. The way I look at it is at a at an international level. So you talk about China, for instance, I really think we may be at the point where we need the Cold War, you know, the red phone. I think that all of these com countries, not just China, not just the US, but all of them, need a way to communicate when something happens. Like if AI did pull a stunt and or or look like it was or faked a stunt or whatever, because you can you can fake people, you can fake things. I mean it’s so easy.

To do something fake, you’ve got to be able to pick up the phone and talk to someone on a reliable format that it’s not digitally rendered and say, hey, we didn’t do that or we didn’t mean to, or whatever. You you’ve got to be able to do that. You gotta be able to disarm that thing, or that really could turn into the the scary thing that they’re worried about. It’s such an easy way. So I I look at it this way, principle wise, I talked about accountability, staying with the humans. I think that has to be at the fundamental part of any.

Any of these policies that do make it through, watch for who’s accountable. Second part is transparency. We need to look at how are we gonna know what’s going on, right? There’s a there’s a one one policy someone was talking about, I don’t know if where it’s at, but having NIST monitor what’s happening. I also like A Amade’s idea of having third party regulators or third party observers that are in in house living at a desk in their office. I like that. I just think it’s it

I I just wonder, you know, how much can they pull the switch? You know, it’s one thing for them to be able to be inside and they can write their white papers and and publish and that’s great if that really happens. But does it change anything? Does it really matter? And it has to be because remember, like like OpenAI made a new regulation that says they’re having people observe and they’re saying that if if in thirty minutes you can’t solve for the problem, you shut it down. I I think that’s probably okay.

Scott Weiner (16:44.598)
I just hope that we don’t get to the point where that’s, you know, what we have to do to solve these problems.

Shawn Cordner (16:48.738)
I mean it sounds it sounds reasonable when you say it like that, right? Third party observers, but isn’t that where regulation comes in? I mean, isn’t that what the government’s supposed to do? Isn’t that what regulatory bodies do? Isn’t that what Congress is for? I mean, why why are we reinventing third party observers here when we actually have mechanisms for that?

Scott Weiner (17:07.522)
Yeah, we have I exactly. We have, you know, we have financial tools for doing things like this. We have we have things in healthcare. We we have ways of doing this. We have boards and panels and so on. I think the thing that’s different here is the complexity and the speed at which reaction needs to be taken. You can’t see you can’t see a virus get invented. I mean a human virus get invented and distributed and then deal with it. Like you gotta deal with this stuff fast. And so these breakouts

they they need to be observed and addressed immediately. So I’m looking at what’s the what kind of teeth do they have. So yeah, the regul some of the the policies they’re talking about have really steep penalties, like twenty years in prison type things. But but that’s all after the fact. Like you know, if it actually did damage, then, you know, okay, this person’s going away for a long time. But that’s not really we don’t need deterrence. We need monitoring. We need to know it’s not going to happen. So it’s a little different that way.

But I think it does look a lot like the the f the finance industry in terms of regulation.

Shawn Cordner (18:07.34)
I was just gonna say, know, d isn’t it true that the SEC will have observers sit with brokerage houses for periods of time, right? So there are Yeah, exactly. So there are models for this. And you know, the back to the arms race with China thing, like I think it’s in everyone’s interest. J you can look at nuclear as a model, right? So there’s mutually assured destruction. but there’s also cooperation. There’s international cooperation, the IA EA, for example, right? So

Scott Weiner (18:14.99)
And all that. Yep.

Shawn Cordner (18:35.788)
There are models where individual nation states that aren’t even necessarily friendly states can work with in fact it works best when they’re not friendly states, when they’re not allies, right? Because it’s it’s it it is that red phone to create that communication and for them to collaborate on these issues and try to mitigate the things that could lead up to an escalation and a problem.

Scott Weiner (18:58.616)
Yeah, it is funny. I we we we we are gonna need AI monitoring AI for some this. We’re going they’re doing it. We’re gonna need to do more of that. But at the end of the day, I really think it’s humans are the solution to the to the problem. I I just I just think communication is really important. And I don’t think this is I don’t think this has to be as complicated as it’s being made to. And I like kind of what you said before, I can’t speculate on their motives, but it does feel a little like some of this is being fabricated. A little bit. I don’t wanna say that the problem is not real, but the

The intensity feels a little bit

Shawn Cordner (19:30.498)
I agree. And you know, you talked about the IPOs that are coming up. And if you think about one of the root causes, I believe, personally, for all of the fear in society right now about AI, it it started with how these founders from companies like OpenAI and Anthropic, less anthropic, but Palantir, how they would publicly talk about the danger and the risk and also publicly talk about how

they were going to start replacing jobs, right? We should have already seen a twenty percent reduction in labor force by now if you would have listened to their predictions just a couple of years ago. And it instead what we’ve seen is a net creation of over a million jobs thanks to AI, right? So it’s gone the other way. So I I almost feel like there’s some sort of Scott Galloway, I don’t know if you know him. He’s a podcaster professor, sorry.

He’s been talking he’s been using the term techno narcissism, right? It’s like my model’s so big and bad and it’s so great and and you gotta check it out and you should invest in it because it’s so awesome that it could actually destroy the entire world. That’s how powerful this is, right? So is it a play to get funding and and to get more ca we know they’re burning through tons of capital. We know enterprises are starting to question what the ROI is. Will that start to impact the investment? Will public scrutiny and concern start to

impact the growth and the adoption of AI. I don’t think what you I don’t think you’re gonna see a contraction of the AI market, but what you may see is a slowing of the growth. It may not scale up the way that they expected it to scale with all the capital investments that they’re making in it now. Does that put them in kind of a house of cards situation? And I’m gonna go finish the rant for just a second. If there starts to be some sort of less demand than they expected,

And they’re on the hook for all this capital investment. And you’ve got this kind of circular, incestuous industry where NVIDIA is investing in the companies that are buying their chips from them. And you’ve got a bunch of off-balance sheet accounting of capital investments that are being done. We know that 50% of GDP growth last year was from AI investment. We also know that GDP growth has been slow right now. If you end up in a situation where

Shawn Cordner (21:50.338)
There’s less AI investment, you end up with stagflation. I mean, this whole thing starts to feel like a bubble.

Scott Weiner (21:57.526)
It it feels a little like a bubble or a Ponzi scheme. I mean it does. It does a little bit. Like, you know, you don’t want to be the last person in there. It it’s funny because y I remember was I w was it earlier this year, last year, I’ve spa it goes so fast now. Some of the subsidized subscriptions started to dry up. They weren’t worth as much as they were. You you all of a sudden people were freaking out because they’re they were running out of their subscriptions in the middle of the month and didn’t understand why and so on. And so, you know.

It was a wake up call for everyone and I think it was a really good wake up call, at least early adopters, to realize like there’s not gonna be a free lunch forever. Like you’re gonna have to figure out this is expensive to do. I do what I can with my subscriptions, but when I have to use API stuff for for clients and things, the boy, you have to watch those dollars. I mean, it goes fast. And then you start realizing how you know, and you get better at optimizing, but it doesn’t matter. It’s expensive. And what I think’s gonna happen is we’re gonna start finding every company is gonna start bifurcating their work and they’re gonna have

internal models, they’re gonna have frontier models, they’re gonna have some open weight models, you know, that they’re using through cloud providers and so on. They’re gonna have a mix. They’re gonna have models running on their phones that are good enough to do a lot of the work at some point. So I think this is more complicated in some ways, but I also don’t mean that the the mix of solutions is going to vary, but they all come down to the same points. Humans need to put accountable systems in place. You create your trust layers first and layer the AI on that, not the other way around.

I was at a client today and we were talking about how to use AI to optimize their development process. And one of the things I was telling them was, you need to look at your processes generally, figure out your bottlenecks today and resolve those now, or at least optimize those now, because when we layer the AI solutions on, all that’s going to go faster is getting to those bottlenecks faster. You’re still going to have the bottlenecks because there’s still checkpoints and a and human problems and so on. So everywhere I look in terms of the AI risk, it looks sort of like that too. It looks like

The the AI is a problem because it’s powerful, but it’s the humans that can put the systems in place that make that not as big a deal. If you don’t leave your A your your passwords in your code, you probably aren’t gonna have them stolen from the AI looking at your code. That type of thing.

Shawn Cordner (24:07.604)
And and the bottlenecks are much more expensive bottlenecks now, right?

Scott Weiner (24:10.934)
Yeah. Yeah. They they are. And they’re getting you know, we were using that traffic analogy before, but one of the things that’s happening is AI is putting a lot more cars on the road and that’s just creating more, you know, traffic jams. Yeah. So we gotta we gotta so when we say slow things down, I don’t think that slowing things down necessarily is conceptually bad. I think when we say slow everything down is problematic. If we know there’s a risk in an area, let’s identify the risk and how we’re gonna resolve it. That that’s that’s reasonable.

But just saying stop everything because don’t know anything, that’s aviation kept innovating even and got safer over time, even when planes were falling out of the sky. And the reason was we didn’t penalize people for coming forward. We didn’t penalize people for figuring things out and saying, hey, this is a problem I’m concerned. We we rewarded them. And if we create reward systems like that, I think that this stuff kind of But it shouldn’t take someone running out of anthropic, quitting their job and saying the sky is falling. It shouldn’t have to get to that point for people to feel like

comfortable coming forward and we want to talk about these things.

Shawn Cordner (25:11.084)
Yeah, absolutely. Two quick things I want to touch on. Another headline that sounds really bad, but I think when you get below the surface a little bit, it’s not quite as bad as it sounds. Anthropic blocking flawed being used for bioweapons.

Scott Weiner (25:24.531)
yeah. Yeah. Yeah. I mean okay. I I think that’s great. I think I mentioned this book to you once before. I I’m I’m I’m trying to remember the name of the author. You might have to put it up later, but if anyone builds it, everyone dies. And and it you know, it it came out l was la late last year, early this year. It it’s basically just a kind of a anthology of all the things that have already happened in AI that are scary.

And then projecting outgoing what could go wrong. And so a lot of the stuff that you’re hearing now, it’s it’s they’ve been saying for a long time now. But one of the things they talk about was that I’ll use Open AI because that was the one they used. OpenAI, they they all do this, but they have basically gateways in front of the AI. So when you ask a question like how do I build a bomb or how do I make you know an illicit drug or something, the AI is gonna stop that from getting getting in.

But if it doesn’t for some reason, on the other side, when the when the data’s coming back out, there are layers that actually say you can’t re say respond this way. So they’re able to kind of prevent things from going back and forth. So the story they tell is one of, yeah, but a lot of that training happened in English. So they poked around with other languages. And I don’t know I honestly don’t remember what model it was they did this with. They said they poked around with other languages and eventually they found a language that could they could ask the question in and the answer came out in that language. And it got through, right?

So the reality is I’m glad they’re stopping it from doing bioweapons and all that, but I promise you, smarter people will start to figure out ways to answer questions around that. And so I you know, the idea that you could do that with a model today, like I don’t I just I just see that as a temporary stopgap. It’s a good thing, it’s not a bad thing to say, don’t don’t allow bioweapons to be built in my model. I don’t think that’s a a bad thing. But at the same time, what if you’re trying to build defenses against bioweapons? Am I not allowed to talk about that either?

Shawn Cordner (27:11.66)
that’s exactly what i what actually happened. So if if you get a little further in the articles that are written about it, the honest journalism on it at least, what they really found was somebody was trying to write a a grant for gain of function research for a novel virus. Well, that is a perfectly reasonable use case. That’s obviously something that

you can’t just do there’s sort all sorts of regulations around here. All they did was write the grant to try to get the funding so they could actually do the research. And that’s important research and gain of function functions similar to how you would create bioweapons because you’re taking a virus and you’re making it you know, give giving it new capabilities, I’ll say it could be more virul virulent or it could be more deadly or i the transmission could change or

how it’s transmitted, so y you do those things on purpose so that you can be ready, you know, vaccinate against it and be ready if something like that were to actually mutate in the wild.

Scott Weiner (28:21.038)
That’s right. And and I think I think certainly it’s reasonable to say, hey, we’re not sure what they’re doing, so we’re gonna s we’re gonna put a flag up. I do think for a lot of this stuff there should be smoother paths for getting those permissions and and making it less bureaucratic and all that. But w we’re still figuring this out. I don’t I I think I’d kind of give people some space on this stuff. They’re trying to figure out what’s the right thing. I I get being a little panicked about someone asking about building a virus with my software. So I

Shawn Cordner (28:47.982)
Yeah. I I’m certainly not criticizing the action that they took. I’m criticizing the headline Yeah.

Scott Weiner (28:53.782)
The hype. Yeah. my god, they’re trying to build weapons and yeah, I know.

Shawn Cordner (28:58.424)
But the very last thing I wanted to bring up w a little bit outside of what we’ve been talking about, but I found it interesting, especially that it was coming out of Florida, that they’re trying to put legislation in place to hold ChatGPT accountable because or I’m sorry, open AI accountable because somebody used ChatGPT for the FSU shooting to help plan it. So it it it I’m not quite you know, I’m not a lawyer, but I don’t think something like that is going to hold up. I think there’s, you know, tons of precedent around gun manufacturers and

whether or not they’ve been held accountable for shootings and so I I just don’t think that there’s case law to support this. But I don’t know. Do you have any thoughts on that type of regulation?

Scott Weiner (29:37.664)
Yeah, and that’s not the first one, right? We had we’ve had several sad incidents with children and feeling if they were manipulated by the AI and all that. When I look deeper into those, I’m not even sure those are I mean, they actually won in some of those cases, I believe, or something, or settled. I I’m not even sure those are really, you know, realistic because at the end of the day, that’s not that’s not the intent of the software. That’s the that was the use case that was done. I mean, you know, if a child goes and drives a car and the car crashes, is that the automobile’s fault? I mean

There there’s there’s a lot of I don’t want to get too nuanced about this. So I think I think the bigger issue that you’re bringing up is that something that lawyers have told me, ’cause also not a lawyer, that we really want to focus on what the sphere of control is. So what what who has the most control over what happens at that point? Somebody who plans to do something horrible is clearly accountable for the horrible thing. Right.

Shawn Cordner (30:28.898)
They’re the limpus of control, not the AI.

Scott Weiner (30:31.648)
Yeah, and and the thing is, like imagine they were just using like a a word document and writing up their plan. Would Microsoft be responsible? I mean

Shawn Cordner (30:38.542)
They could Google it or they could go to a library and they could look up military tactics, right? I mean, there’s a million ways they could get that information.

Scott Weiner (30:46.144)
Exactly. And and and I’m not saying that that there isn’t some accountability here, but I don’t think it’s at that level. I just I just don’t. But that’s that’s an opinion, not a legal one.

Shawn Cordner (30:56.344)
But right, but maybe kind of like the bioweapon thing, maybe that should set off an alarm and maybe that should be something that

Scott Weiner (31:04.182)
Well that’s the argument I think they’re probably trying to make, right? That th these things should at least alert, if not stop. So I I I think yeah, I mean it’s it’s so tough for me. I gotta tell you, I I’m very I’m very mixed on this ’cause I I feel like I would love to have all the providers stop every bad action. I really would. I also know that on a regular basis, when I’m doing my research and experiments, I’m stopped from doing things that are totally harmless because it’s overreacting.

to what I’m doing and it’s i irritating. So I’m so I have this personal thing that then I’m thinking, Yeah, but that’s not I’m not the problem. So maybe I need to think about the problem. And it’s true, if someone’s doing hot something horrible, I probably should know it. It should be able to stop it. Because if these things are so smart, why can’t they pick up on that and say, Hey, wait a minute, what am I doing here? I think that’s a reasonable thing. Yeah.

Shawn Cordner (31:53.506)
Well, I don’t think that legislation is gonna end up passing, but I do think that civil actions will shape some of the regulatory environment, at least how they might collaborate with each other. Because certainly the burden of proof is lower in a a civil case and big enough lawsuits will start to impact how these companies behave if the government doesn’t step in and regulate. So those are my thoughts.

Scott, I know we gotta wrap up. I want to leave you with one final question. That is, you were the first guest on the podcast. And at the end of that episode, I asked you when is the end of the world going to come at the hands of AI? And you told me about it. So so I in light of everything that’s been happening and in this conversation, I don’t think your answer is gonna be different. But Scott, when do we need to worry about the end of the world thanks to AI?

Scott Weiner (32:46.752)
You can I I don’t spend any time worrying about it. I do think that if you wanted to say when it’s always tomorrow because things keep getting better and improving. Look, I I I I think that the chance of a bad event is very, very high. That I think we’re going we’ve we’re we’re very capable of making mistakes. The chance of something catastrophic to the end of the world, I I keep going back to the George Carlin joke that you don’t need to worry about the world, it’ll take care of itself. You gotta worry about the humans.

Shawn Cordner (33:16.184)
Great place to leave it. Scott Weiner, thanks for your time and expertise.

Scott Weiner (33:19.79)
Thank you. See you later.

Shawn Cordner (33:21.932)
Mm-hmm.