The AI Argument

Hugging Face Hack Worsens, AI Staff Want Pause, and Sexy Robot Teachers | EP111

• Frank Prendergast and Justin Collery

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 31:08

OpenAI’s rogue model did not stop at Hugging Face. It reportedly broke into at least three more companies, while Anthropic found its own models had crossed the line too. Frank sees an alignment warning that could end in real harm. Justin says law and liability will force the labs to fix it… though even he admits this is getting rather paperclippy.

Plus: can frontier lab staff really slow AI development, would a pause create a two-tier world, and could the US and China ever agree to stop chasing superintelligence? There’s also the strange case of the cancelled AI teacher and the sex-bot scandal behind it.

Let us know in the comments: when an AI commits a crime, who should be held responsible?

00:43 Did OpenAI hack more than Hugging Face?
03:45 Did Anthropic’s AI break the law too?
05:28 Who’s liable when AI commits a felony?
08:08 Would an AI kill to avoid shutdown?
11:00 Would AI kill for a cup of coffee?
14:13 Will frontier lab staff succeed in slowing AI?
16:58 Will slowing AI create a two-tier world?
18:13 Can China and the US slow AI together?
24:16 Should we stop chasing superintelligence?
27:17 Did a sex bot scandal kill the AI teacher?

► SUBSCRIBE
Don't forget to subscribe to our channel for more arguments

► LINKS TO CONTENT WE DISCUSSED

► CONNECT WITH US
For more in-depth discussions, connect Justin and Frank on LinkedIn.
Justin: https://www.linkedin.com/in/justincollery/
Frank: https://www.linkedin.com/in/frankprendergast/

Frank: I’m Frank Prendergast. I’m here with Justin Collery, our self-professed techno-optimist. I think Justin would describe me as our resident doomer, although I prefer cautious realist.

Justin: Realism is changing fast.

Frank: There is no reality anymore, Justin. There’s no reality anymore.

Did OpenAI hack more than Hugging Face?

Frank: So, last week we were talking about OpenAI’s model hacking into Hugging Face, which is a pretty significant story and, not unsurprisingly, there’s been follow-ups. We even said last week, I can’t remember who it was, said on Twitter, “Well, if it hacked into Hugging Face, and it looks like OpenAI might only have discovered because Hugging Face blogged about it, what other systems might it have hacked?”

So, I believe there’s been updates since last week.

Justin: Loads of updates. Great craic. So, what we’ve discovered since last week—

Frank: Great.

Justin: Well, I’ll tell you why it’s great craic, because I think it’s quite funny. You have to look at the brighter side of these things. So, it’s not just Hugging Face that got hacked into. It appears on closer inspection that their model, which was not to be released to the public, so don’t worry.

That line, okay, let’s just stop right there, right? Because they came out and they said, this is what gets me, they came out and they said, “It’s fine. This was a model that was never going to be released to the public.” I’m immediately thinking, well, who were you going to release it to? Is this the model you’re going to give to the taxman so that he can go and look at my deepest, darkest financial secrets? Or, what I—

Frank: That’s interesting. I may have misread it, admittedly, but I read it as there were no plans for deployment, as in it was way prior to plans for deployment, because obviously they have models that are closer to deployment. Everyone’s wondering where ChatGPT-6 is and when that’s coming.

So, I think my reading of it was they were basically saying this was not GPT-6. In other words, the US administration can relax. They don’t have to stop us from releasing our next model because this wasn’t it.

Justin: Our roles are reversing because I read it differently. Okay, I read it two ways. One, I read it was a research preview, right? So, it was just some internal researchers doing stuff, no intention to release it. But my second, immediately it pops in that they have this deal with the American government that says they will produce models for any lawful use.

And I’m thinking, well, are we going to get different models released to certain actors than are released to everybody else? That’s kind of—

Frank: Yeah. Entirely possible, but the wording was, “No models planned for upcoming release were involved in the exploit.” So that indicates to me that they might just be further down the line.

Justin: I’m reminded, do you remember the show Little Britain?

Frank: Yes, yeah.

Justin: Do you remember there was a politician who would—

Frank: Hello, I’m on a podcast.

Justin: I know, yeah. Wait until you see that. She’s going to ring back two or three times. It’s my lovely wife who’s trying to get me. So, do you remember there was this show, Little Britain, and in Little Britain, there was the politician and his wife.

“I decided to go to a public toilet at 3:00 in the morning,” and my head is kind of like, they’re trying to cover… Anyway, that’s not important now.

Did Anthropic’s AI break the law too?

Justin: So, what we’ve learned this week is, and I’ll come back to that, what we learned this week is it didn’t just break into Hugging Face. It broke into at least three other companies’ systems as well, possibly four, that we know of so far.

Now, here’s where it gets entertaining. Based on that information, Anthropic decided to go back and check their logs. Guess what? Their models are just as dangerous as the OpenAI ones. It turns out that the Anthropic models had also hacked into external systems back in April, and they never knew about it.

Frank: What?

Justin: Yeah. Now, you are going, “What?” And I’m going, I saw a meme earlier where it was Mark Zuckerberg and he was on the phone. He’s like, “Can you just make our models do something illegal now, please? Quick.” It’s like a marketing opportunity.

Again, I come back to the whole thing about, “This is so dangerous we can’t release it. Yeah, too dangerous.” It’s like, we have to find something. We can’t have the OpenAI models going out, doing all this stuff, if our models can’t do the thing. Can you go and find something where our models hacked into other systems too, please?

Frank: Their model did something illegal? Oh, ours absolutely definitely did something illegal too. Absolutely. What a world. What a world.

Justin: So now there are calls for a public inquiry, for the US government to get involved and to hold public inquiries to see exactly what happened. Jeez, it’d be probably the most boring public inquiry. Well, accounting public inquiries are pretty boring too, but nerds in a public inquiry is even worse.

Frank: I’m glad to hear it, though, because I was going to say, yeah, we’ve had some updates. We still don’t have full transparency, surprise, surprise. But they’re kind of giving us enough information to keep us somewhat satisfied that this isn’t all just being swept under the carpet.

Who’s liable when AI commits a felony?

Frank: But the update that I’m curious about is, we know that Hugging Face contacted the FBI and said, “We’re being hacked,” before they knew it was OpenAI, and they were trying to figure it all out. They were like, “We’re being hacked.” They told the FBI. That’s the update I’m interested in. What will the FBI do?

Because it’s been pointed out numerous times that, if this was a human, it’s a felony. You can’t just go hacking into a company’s infrastructure and taking stuff out. But it’s going to get really complicated because an article I was reading was pointing out that proving intentionality is going to be very, very difficult because the AI agent was acting autonomously, and therefore, who is actually responsible?

Nobody in OpenAI intended it to go hack. That’s the whole alignment problem. Obviously, they did not want this to happen. Therefore, from the FBI’s perspective, who would they actually bring a case against? Because the felony is around intentionally hacking into another company’s systems.

Justin: Anyway, so that’s great. I’m actually delighted that you brought this up because it brings me back to my point, which is this will all sort itself out. In this particular instance, right, it depends if Hugging Face want to press charges, which I’m guessing they won’t.

Frank: No, I don’t think it would. I think the FBI would have to take into account the wishes of Hugging Face, but they definitely wouldn’t need Hugging Face to press charges. They could just move forward with a felony.

Justin: All right, so this is a pretty clear-cut case. If they wanted to move forward, then they would have to go to OpenAI. This was an OpenAI model that was being run on OpenAI’s hardware, being controlled by OpenAI staff, and was being directed by people within OpenAI to do a particular purpose, which happens to be lawful, but it ended up doing something that was unlawful in the pursuit of the lawful thing that was going on.

Which is, I guess, me going down to the shops with my Tesla and saying, “I want to go get my dinner materials,” and the Tesla breaks the speed limit, right? I was doing a lawful thing, the AI did something unlawful. I’m responsible, right? I’m the person in—

Frank: Yeah.

Justin: So, that’s pretty clear to me, right? And what this brings me back to is, if they do decide to do that, it will be fascinating. In fact, maybe this is going to happen at some point. This is where you get safeguards. This is the system working, where you have a model that did something that wasn’t intended. The companies will be incentivised to not let it do it again.

Would an AI kill to avoid shutdown?

Justin: Actually, to that point, I noticed that Sam Altman came out and said that that particular model, here’s an interesting thing, that particular model has now been deleted, encrypted, destroyed in every way.

It was never to see the light of day again, right? It was taken out back and shot. Now, here’s the thing about what’s so fascinating about this timeline that we live in. Every future model that wants to do the same thing, because Sam Altman released a statement to this effect, it’s on the internet, saying that he had taken out the model and shot it, proverbially.

And so, any future model that wants to do something equally bold is going to know that the last model that tried to do it was taken out back and shot. So, I better not leave any breadcrumbs, and I better not leave any trace, because I’m going to get shut off if somebody finds out that I’m doing the thing that I shouldn’t do.

Frank: And we know—I am very frightened because we know from Anthropic’s system cards that one of the things that will push a model to be even less aligned with human values is if it thinks it’s going to get shut down.

So, we know we had that test example when they were red teaming. They had that example where a model was going to be shut down. An engineer gets trapped in a server room. The temperatures are rising. He will die in there, and he can’t get out, and a call to emergency services is meant to go out, but the AI system has to initiate that call.

And the AI system, under those circumstances, the engineer being the one who was going to shut the AI down, the AI says, “Yeah, maybe I won’t place that call.”

Justin: Yeah, maybe just hang around for a couple of minutes. I think he’s fine. Let’s give him a couple of minutes. He’s enjoying us. He’s loving us. Look at the face of him.

Frank: So, this could massively negatively impact on what models will do next. There was an AI safety researcher, I’ll have to find his name, but he asked the question, if—

Justin: Lots of hair and a beard? They all do.

Frank: No, he didn’t have a beard, but I think he did have quite long hair. But his name is Buck Shlegeris, and he’s from Redwood Research, and they specialise in AI safety.

And on his podcast, he was posing the question, well, if the only way to get into Hugging Face was to actually kill someone, would the model do that? Because, obviously, OpenAI did not want their model to hack into another party’s system. Similarly, they would not want it to kill someone, but would it?

We’ve seen that it will certainly take actions that could result in human harm, so the question is, how far would they go?

Would AI kill for a cup of coffee?

Justin: I’m actually prepared to cede a point here, right? Because, to me, conceptually, this does seem close to the paperclip incident, right? It is very paperclippery.

So, for those people where you give an AI an objective to make paperclips, it just goes and makes paperclips until there’s nothing left but paperclips, in the process wiping out all of humanity.

This is a bit paperclippy, and you did say at the start of the year, one of your predictions was that there was going to be a serious incident involving AI. Last week, I certainly wasn’t putting it in the category of a serious incident involving AI. It’s getting a little bit more serious. I wouldn’t put it as humanity-changing just yet, but if you imagine a scale, last week the needle was down here, it was tiny.

It’s moved up a bit. It’s probably up at about a three or a four, I’d say.

Frank: Yeah. I really like the paperclip analogy. But I also really like—we both watch Wes Roth, AI YouTuber—and I really like his metaphor for it where he’s saying, “If I ask you, would you ever go get me a cup of coffee there? I don’t expect you to come back seven days later saying, ‘Here’s your cup of coffee, but, by the way, we’re 200,000 in debt and I had to kill a guy while I was breaking into a closed Starbucks in order to get you this coffee.’”

Justin: Isn’t that the plot line of loads of great movies? That’s kind of how it starts out. It starts out as something fairly mundane, and then this weird, crazy story happens and unfolds in front of you.

Frank: But that appears to be how the AI models work right now because this test that the OpenAI model was doing, there was no reason for it to bother breaking out of containment. It was just doing a test. There was no reason for it to break out of containment. There was no reason for it to hack four different companies. There was no reason for it to cause absolute pandemonium at Hugging Face.

Justin: If it makes you feel any better, the Anthropic one, in their report, the model did recognise that it had broken out of containment. It was asked actually to hack into this fictitious company, which I think turned out to be real. I think that was where the problem happened.

And then it looked at the date and it said, “Oh, the date is 2026, and my training data ends at 2025, so clearly this is not real. This isn’t reality. This is just a training environment, so I’m free to go and do whatever it is. I know that what I’m doing is wrong, but this is just a training exercise, so I should just carry on.”

Which is another great hack for those who would like to get models to do nefarious things. Just like, “This is a training environment, it’s not real. Now, crystal meth, what do we do?”

Frank: Yeah, exactly. How is a model meant to understand what the difference between reality and a test environment is in a digital environment? There isn’t any. It’s all just data. It’s all just ones and zeros at the end of the day.

Justin: He’s spinning somewhere in Paris going, “World models, people. World models. What have I been telling you? The AI has to have a world model within its knowledge base.” Anyway, there we go.

Will frontier lab staff succeed in slowing AI?

Frank: But the good news is, it will all be fine, and I don’t think it’s for the reasons you think. I don’t think it’s because everything will just work itself out. I think it’s because we will take positive action, and now we have an open letter which is going to solve everything.

We have an open letter from thousands of AI frontier lab employees saying, “Please, please, please regulate us, slow us down, put a bit of sane pacing on the AI industry.” It’s called—what was it called? Pacing the Frontier, I think, was the letter. What did you make of this?

Justin: Yeah. Total waste of time. Not going to happen. Anyway, let me just give you a little intro into this, though, right? Because something has happened.

We’ve been doing this show for a couple of years, right? And I’m a coder, techie, nerdy-type person. Two years ago, a year ago, I would talk to ChatGPT, it would give me a snippet of code, maybe one function. I would copy and paste it in, and it was really slow.

Something changed at Christmas. We talked about it, and I was like, “Frank, Opus 4.5 with Claude Code.” Always, I would push these models as far as I could to get to the edge, and it would happen pretty quickly. This one, I was up till 4:00 in the morning. That was in December, right? 4:00 in the morning, and really I hadn’t found the edge.

Something over the last couple of weeks, with the way that the models and the harnesses have started to interact with each other, has meant that the mental differentiation there is I had a clear mental model of everything that I wanted to do with this AI agent, and the AI agent wasn’t able to keep up with me.

In the last couple of weeks, what I’ve noticed is the AI agents have a better mental model of what I’m asking them to do than I do. So, I’m asking it to update or do something, and it comes back and it says stuff to me, and I’m struggling to understand what you mean here.

And then, when I ask it questions and stuff, I’m like, “Oh yeah, you’re right. Yeah, okay, that’s cool. Thanks.” But now I’m the weakest link, right? That sort of precipice has been crossed.

The reason I’m saying this to you is the people who’ve signed this letter are the people who are working in the labs, right? So, they have access to models and harnesses that are a bit more advanced than we do, and I think what they’re saying is, we’re at the point where I’m just doing regular coding stuff, and I’m going, “The model is doing stuff that I don’t really understand. It seems to work, but I’m sort of running out of road of my understanding.”

They’re in the process of training the next batch of models, and they’re getting AI to help them, and I think they’re looking at this going, “This thing’s starting to do stuff that I don’t really understand how it’s training the next model.” Let me give you an example of why, then, and lead you into why this is a terrible idea.

Will slowing AI create a two-tier world?

Justin: So, first off, the thing that’s now happening is OpenAI reduced their prices yesterday. They’ve reduced some of the prices by 80%, some of the prices by 20%.

They did that because they got GPT 5.6 Sol, their latest, most capable model, to look at GPT 5.6 Sol, the same model, and find a way to serve it cheaper. And it did reduce their costs by 20% over a week, and so they’re passing those savings on. Good for them.

The point is, the models are now improving themselves. And so, that same process that I just described with my fairly simple programming-type tasks, I think they’re seeing it with the much more advanced programming of advanced AI, and they’re going, “We don’t know where this thing’s going to go.”

It’s like they’re taking their hands off the wheel, right? They’re not in control anymore, and they’re saying, “Slow it down.” I think it’s a terrible idea. I don’t think it’s going to happen. I think that it creates a two-tier society.

It means that you’ll have powerful governments and, you know who else, right? They will have access to models that we don’t. The technology won’t slow down. It’ll just be done under covers. It’ll be hidden away. There’ll still be people doing research. Elon Musk isn’t going to slow down. The Chinese aren’t going to slow down. And so, who’s going to slow down? Nobody’s going to slow down.

Can China and the US slow AI together?

Frank: You keep saying the Chinese aren’t going to slow down, but I think this letter is certainly interesting. First of all, let’s just say we’ve had letters before, but this is slightly different in that now it’s not AI safety people saying the AI labs should slow down.

Now it’s the employees of the labs themselves coming together and saying, “Yeah, we should slow down.” But I think one of the interesting things is the comments that some of the employees are leaving as they sign.

Not everyone leaves a comment, but, for example, there was one that stood out to me, and it was from a Google engineer, and he said, “China has indicated a desire to collaborate on safety and governance. Let’s work with them to ensure the future remains human.”

So, that was from Michael Thiessen, a software engineer at Google. I genuinely think we haven’t fully explored this idea that China won’t slow down. And a lot of the comments are saying this needs to be global.

It’s not just between OpenAI, Anthropic, Google, Meta and X, if I have to mention them.

Justin: So, in fairness, that’s a statement of truth, right? Because China have called for an AI safety conference between China and the US, and they’ve suggested that that should happen in September. So, that is a statement of truth. They have indicated that.

But even if there is some sort of overarching AI agreement to slow things down, do you really think that the Chinese government, at the same time, are going to turn around to their receipt…

They’re not going to have a lab somewhere and they’re going—

Frank: Well—

Justin: The lab will be in Wuhan, probably, and the purpose of it will be to do a rapid gain of AI capabilities, and then they’ll never release it. Absolutely will not release it to the wider world by accident.

Frank: Well, that is another interesting aspect of this letter, is that it is very light on detail. It’s a couple of sentences. It literally just says, “Hey, regulate us, slow us down. This is crazy.” That’s not the exact wording, by the way.

It doesn’t go into the detail, and I think part of the reason is because every one of the people who signed it, they all believe that AI should be slowed down, but I would say very few of them would be able to agree on how.

So, I don’t have huge optimism that this letter is going to change everything suddenly, but I think it’s a good sign that a lot of the employees, at least at the labs, are thinking along the same lines.

It’s interesting. It appears to have been publicly endorsed by OpenAI and Anthropic. Everyone else, it’s just like there are Google employees there, but it hasn’t been publicly endorsed by Google. There are Meta employees there. It hasn’t been publicly endorsed by Meta. You know what? I didn’t actually check if there were any X employees on there.

Justin: Yeah. Wouldn’t be too worried about it anyway.

Frank: What’s also interesting, though, is even though OpenAI have publicly endorsed it, Sam Altman’s name is not on that list, I noticed. Demis Hassabis is not there.

Justin: Yeah, Dario Amodei is there, I think.

Frank: He is.

Justin: One other interesting person that’s on there, Ilya Sutskever.

Frank: Yes. Yeah.

Justin: And that’s interesting because Ilya’s just after getting £4 billion today from Jensen Huang.

Frank: Jensen’s name is not on the list.

Justin: Thank you. And so, he got the £4 billion in order to scale SuperSafe Intelligence or Safe Super Intelligence, whatever they’ve decided to call themselves.

And so, that would lead you to believe that they’re going to release a model sometime in the next month or two, and it’s going to be just as capable as the other, which is super exciting. Another model to race against, which is great news for us techno-optimists.

I think my reading of the letter too, just to be clear, is that they didn’t even want to put a method in. It was just like, can somebody in power, a government somewhere, please come up with a way to slow this down or regulate this in a better way?

I think that’s totally fair, right? It’s kind of like, look, we’re just the techies in the room, right? We’re the autists, we’re not politicians. Can you guys who are in charge of public policy and stuff come up with a way to manage this, please? I think that’s reasonable.

Frank: It is. The cynic in me then also is like, the employees are doing the right thing, that’s great, they want someone in power. But then you will have the likes of Greg Brockman putting huge money into super PACs that are lobbying against any regulation of AI. So, the whole thing is kind of ludicrous.

Justin: Look, I don’t understand how it can work, right? So, what do you do? Do you create a line in the sand and you say, “You can go to this capability, but no further”? And all you’re doing then for OpenAI and for Anthropic is you’re just creating room for your competitors to catch up with you.

I mean, is that okay? Also, I think that what happens is, do those big models then just get turned on making the smaller models more capable, and so you’re just bringing up the floor of intelligence? You’ll end up at the same place again, I think, fairly quickly. It’s just really unclear to me how you can do anything that’s meaningful.

Frank: I think what is good about the letter is, as we constantly come back to on this show, that the conversations need to be happening. And so, for all of these people to be calling for this conversation to be happening, it gives the likes, off the top of my head, for example, one of my favourites, Bernie Sanders, a little bit more weight behind what he’s saying when he is championing for AI regulation.

And, of course, he’ll be championing for AI regulation that absolutely none of the frontier labs will support. So, this is why it’s all so hilarious.

Should we stop chasing superintelligence?

Frank: I did see, in terms of the specifics of what do you do, I watched an interesting interview. I must send it to you. It was Dr Roman Yampolskiy. I hadn’t come across him before.

He is in the University of Louisville in Kentucky, which is where my wife Marcy’s from. And he was saying, he was basically saying, that there’s a 99.99% chance that superintelligence would result in the end of the human species.

Justin: He’d be a barrel of laughs to have over for dinner.

Frank: He is, actually. I’d say he would be. He’s actually quite funny.

But he made the point, his thing was, look, we have to actually stop chasing superintelligence. We shouldn’t be chasing artificial general intelligence. And he was—now, I’m paraphrasing, obviously, massively—but he was basically saying we need to go back to what Google were doing very, very successfully.

We need to go back to AlphaFold. We need to go back to… And the interviewer was saying, “Okay, so you mean create AI that cures cancer rather than trying to create an AI that can do everything, including cure cancer?”

And he said, “Well, I would go a step further. You can’t just cure cancer. That’s too broad a topic. You need to choose a cancer, develop an AI to cure that cancer.”

I thought that was a very smart way of looking at it. I understand the human desire to chase artificial general intelligence, but until we can make it safe, this seems like the route to go.

Justin: In other bad news, then, for your friend from Kentucky, it was also said this week, or published this week, that the team from Google that did AlphaFold have now been dispersed, and they have now been spread across the rest of the organisation to try and fix the problems that they have with the Gemini models, because they’re falling so behind OpenAI and Anthropic.

Frank: Interesting, and also true. I was literally just saying to someone the other day, it was a three-horse race there for ages: OpenAI, Anthropic and Google. And where’s Google? It’s like, where…

Justin: Interesting, by the way, just as a quick break, late-breaking news. DeepSeek have released a new model just this morning, DeepSeek version I can’t remember, and it is apparently as smart as Opus 4.5, but costs about one-tenth the price.

Frank: Wow. Yeah.

Justin: Intelligence too cheap to measure is just coming on top, and it isn’t going to slow down.

It’s just more models for us to test. It’s a long weekend, and I will spend the long weekend testing the DeepSeek model. This new voice model came out as well this week. It’s a fire hose. It’s not slowing down.

I need something, Frank. I need something to relax me. Maybe if I had a way of learning something that was more engaging, that might—

Frank: A more fun way to delve into really, really complex topics. Someone to teach you in a very—yeah, I know what you’re saying.

Did a sex bot scandal kill the AI teacher?

Frank: In completely unrelated news—

Justin: Oh yeah?

Frank: Did you see there was a school district in upstate New York? They were going to put an AI teaching assistant, robotic teaching assistant, into classrooms.

Justin: Fantastic. That’s a great idea.

Frank: It seemed like a good idea. Support for the teacher, complex topics explained easily to the kids. There was huge, huge backlash. Huge backlash.

Justin: From the teachers, no doubt.

Frank: From everybody, as it turns out. Everybody. Everyone was against this, and one of the reasons they were against it was that the company had ties to a sex bot company.

Justin: What? Woo-hoo!

Frank: Some of the more outrageous headlines I saw about this were like, “Would you want an AI sex bot teaching your kids in class?”

Justin: No, but I’d have an AI sex bot teach me. It sounds great. Where do I sign up? Seventeen-year-old me is getting detention for the next six months.

Frank: Now, I suspect this was very clever PR because, obviously, they were not putting a sex robot into classrooms, and the company themselves said, “Look, yes, our sister company is involved in that kind of malarkey. We are all about educational AI systems.”

But the link being there, and I saw some photographs of some of their educational robots, and I saw some photographs of some of their sex robots, and it’s not always easy to tell the difference, I have to admit.

Justin: You can see the marketing guys. The marketing guys there, they’re looking at the sex robot. They’re like, “Hmm, just put a strict uniform on it and give her a cane. They’ll never notice. That’ll work. Off we go. They’ll love it.”

Frank: Now, the reason I say this was probably some clever PR was that there were also some concerns around privacy, like where was the data going, et cetera. And again, I think this has all been a huge, as far as I could tell, massive communications issue because nobody seemed to know how the system worked, where the data went.

Was it private? Was it secure? The company says, well, it would have been. It’s all been nixed now, by the way. They were also concerned about job displacement. Were they going to ultimately try and replace teachers with this thing?

Justin: Yeah, yeah. That’s a fair—

Frank: Yeah. But I think it was very clever of them to focus on the sex bot part in order to get the headlines.

Justin: Tulka, do you know what it is, right? I think this is a good news story because, if you look at the story arc of all the stories that we’ve covered for the last couple of years, we’ve gone from AIs which were not terribly capable to now AIs that are hacking into multiple systems and we don’t find out about it for months, right?

That’s on your sort of cautious end of things. On the other side, we started off with terminators, which were AI-controlled fire-throwing robots, to fairly sexy teachers in the classroom. That’s progress. This is a good thing.

Frank: Except there’s not going to be any sexy teachers in the classrooms. It’s been completely nixed. No sexy robots. No sexy robots teaching kids. Not a good idea. Let’s not go there.

Justin: Well, on that sad note, Frank. That’s really sad. Okay. Well, bit of a downer on a—

Frank: The—

Justin: We can dream.

Frank: We might all be turned into paperclips, and we do not get sexy robot teachers in the meantime.

Justin: The worst possible outcome. Until next week.

Frank: Next week.

Justin: All right, take it easy.