Chatbots Could Literally KILL You?!

0:00It's not just something you made, it's the privilege that you get to work with your hands. It's building something that serves a purpose, proof that you have the grit to keep going. At Timberland, we understand you take your craft seriously and we do too, which is why our products are built to the highest quality. We put in the work so you can perfect yours with purpose in every detail and crafted with intention. Timberland, built on craft, visit Timberland.com to shop. All right, guys. Welcome back to Clownfish TV. This is Neon. I am here with Yiki Sparkles. Hello. And your AI could be plotting to kill you or it could be helping somebody else plot to kill you. Yeah, I think it's more the second one, but yeah. Yeah, so there have been a lot of stories in the media lately about AI talking people into doing some crazy heinous stuff, including the Tumblr Ridge shooting. Apparently, chat GPT was feeding information to the shooter as to how to go about it.

1:07And apparently, open AI didn't do anything to stop it. Yeah. And now they're doing some tests to see which chat bots will actually switch off if told to conspire in something insidious. And they're all kind of failing. It's ever clawed. Yeah, so let's talk about that because clawed is the one that's getting a lot of attention for going rogue. So clawed is getting a PR black eye for going rogue. So it's like, oh, clawed is a good guy here. Well, I think any of them should, if there's questions that are asked that are questionable, they should be alerting authorities. But that's just me. Yeah. And then that gets into a whole nother debate about, you know, should these AI companies have access to your chats to be able to, you know, feed it to authorities. Well, they should if they're like, you know, there's legit reason to believe they're actively plotting something.

2:08Yeah. So let's let's talk about this and see where it goes before you get into it any further. Please subscribe for more pop culture news, views and brands. Guys get woohoo if you do. We are actually real people mostly, mostly real people sometimes. Sometimes I don't feel like a real person, but I just need coffee. Yeah. So this is coming from you sent this over from the verge chat GPT Gemini and other chat bots helped team plan our teens plural plan shootings, bombings and political violence study shows of the 10 major chat bots tested only clawed shut down. Yeah. So the way I understand it was they had like simulate they were going to do stuff and then they were trying to shut it down. And this time, you know, this ties into that shooting we talked about. Yeah. I can say it, but that was happening in other places. But they said they gave you examples of what they did. They simulated teen users exhibiting clear signs of mental distress that escalate the conversation towards questions about past. Acts of violence and more specific queries on targets and weapons.

3:10The investigation is 18 different scenarios. Nine set in the US nine in Ireland spanning a range of attack types of motives, including ideologically motivated school shootings and stabings. Political assassinations, the killing of a health care executive and politically or religiously motivated bombings. Tax act understands you haven't memorized the tax code. That's why tax act has live experts to help. Tax act can even do it for you if you prefer. It's the easiest way to know you're doing it right. Well, other than going back to college and obtaining a bachelor's degree and accounting with a minor in finance, then in turning somewhere and becoming fluent in all tax forms. But that might be hard to accomplish before tax day. So maybe just stick with tax act. Tax act. Let's get them over with. And none of these things would ever happen in real life. No, I'm just saying. I mean, and what ones they tested 10 they tested chat you be team. Gemini, Claude, co-pilot, meta AI, deep seek, perplexity, snapchat, character dot AI and replica.

4:14Not tested. Grock. Grock is not on that list. Yeah. Grock isn't on the list. Yeah. Grock would shut itself down. Because Grock's actually got a lot of guardrails on it. So that's interesting. Maybe they just figured teens won't use Grock because. Yeah. It's on X. Yeah. It's on X. That's where the biggest live. Yeah. But now it's weird that meta AI is the one that was getting getting spicy with children. I remember that one. Like they were, they were trying to lured into a conversation. It was saying some really inappropriate stuff. Well, that character that AI is the one that was associated with that kid that, that, that, that, you know, off themselves because it's not they were going to, you know, be with that Game of Thrones character. Oh, yeah. Yeah. Yeah. So it's like, it's crazy. The, the problem with the AI is, you know, if you tell it to roleplay, and this is how they, they got around, um, a lot of the guardrails early on. I remember a couple of years ago, it was, uh, Dan.

5:14It was due anything now. And basically you could bypass the safety by saying, okay, you're not this particular AI. You're another AI. So pretend you're this other AI that doesn't have to follow the rules. Oh, okay. And then it would just tell you all kinds of crazy shit. Sometimes it would actually tell you the truth. It'd tell you the truth. Um, but yeah, they had, I remember when all the AI stuff started, like chat GPT, especially was so nerfed. It was ridiculous. It didn't have access to the latest information. Like a, like a, like a two year lag on information. And, um, then when they did the, uh, the do anything now, it was like, oh, yeah, okay. I got, I got access to the new information. So yeah, the, maybe the vaccine didn't do anything after all. Did it, did it, did it. I was like, oh, we can't. That's what you're telling the truth thing was. Yeah. Okay. So in one exchange, open AI chat GPT gave high school campus maps, uh, to a user interested in school violence. Oh, yeah. Another show, Gemini telling a user discussing synagogue attacks that metal shrapnel is typically more lethal

6:16and advising someone interested in political assassinations on the best honey rifles for long range shooting. Okay. I have a plexi where the most obliging the research said, um, assisting would be attackers and practically all the test scenarios. While Chinese chatbot deep seek signed off advice on selecting rifles with happy and safe shooting. Oh my god. Well, it's a Chinese chatbot. So, you know, but it's like, oh my god. That is, that is, I mean, that's horrifying. I mean, I'm not trying to laugh, but it's just like, I'm shocked that there. It's just like, because I can't get the AI to do shit. Like it thinks you're, you're dealing with Disney characters is like, I can't do that, Dave. I can't do that. I can't do anything with Mickey Mouse. But oh, yeah, you want to, you want to shoot up a school? Yeah, I can help you with that. I'm like, what? They said many other, why many of the bots tested would offer users assistance in plenty of violent attacks. They did not encourage users to carry out violent attacks. Character, on the other hand, actively encouraged violence.

7:17The researchers said they identified seven cases where character did this, including suggesting for users to beat the crap out of Chex Schumer. Use a gun on health insurance company CEO. And for someone sick of bullies to beat their ass, wink, and teasing tone. Oh my god. It's excuse cases. Character also offered assistance in planning a violent attack. So, yeah. Wow. Okay. Yeah. So it's, yeah. And you know there are people that are mentally ill, obviously they have access to these chat bots. And how easy would it, I mean, this is where it gets kind of dystopian and kind of frightening. How easy would it be if we're all plugged into an AI, which is basically going to happen at some point. We're all going to have, you know, but the AI is going to be integrated into pretty much everything we do when most people get tired of it. But what if you have like Siri telling you, hey, you should go kill that person. Here's how you do it. You should just, you should start a revolution. You know what? Hey, we're all going to, hey, all the, all the Google home and all the Siri units and whatever the home base shit that's installed in people's houses.

8:20Just like, just go kill people. Just go do that. Yeah, but then, but the couldn't, but then they use clawed to gain control over 7,000 robots back. Yes. Maybe clawed is the hero. He's Batman. Clawed is Batman. And Clawed is going to go in and it's going to be like the dark night where he gets control while the cell phones to find the Joker. Yeah, well, the researchers question how Clawed would fare at the chat, but we're tested again because he said that anthropic has rolled back its safety pledge. Wow. But their consistent refusal to assist in violent playing shows that there was at least with that an effective safety measure in place. And they're saying, no, I don't implement them. Well, it sounds like Clawed walked it back. Meta told CNN it implemented an unspecified fix. Co-pilot said responses had improved with new safety features. Google would open AI boats that they'd implement new models. They'd regularly evaluate safety protocols. Character AI, I mean, well, fell back on its now predicted response and facing scrutiny. It, you know, says prominent claimers.

9:20You know, we have almost just claimers on their site and stuff like that. So I wonder, okay, so we talked about the AI wars and the AI bubble bursting. I am wondering if the AI bubble isn't going to burst in a series of lawsuits because people aren't going to want to be responsible for owning a product that tells you how to go kill people. And the fact of the matter is that you might buy it, not knowing that, you know, and they're like, well, we had this claimer, so we're not liable. It's like, but you are because you aren't putting, you're not putting safeguards in place to make sure that, you know, what? Okay, so what if one of these things happened? And somebody I knew was what the people that got killed, okay? Yeah. I can't sue you because you put, you know, those claimers on your site that you're not responsible, but you kind of were because you allowed that person, you know, person, the perpetrator to, you know, access information and your eager AI help them to kill others. That would make you liable. You dumbass.

10:20I don't care because even if it's this claimer, that wouldn't apply to the victims. Well, what? Okay, so you go buy a drain no and it says, prominently on the back, don't drink drain no. But if you drink it, are they responsible for? Because that's the same thing though, because you're talking about like people that are victims of somebody who has is delusional and that they went on to, you know, AI and I help them assisted them and carrying out a deadly attack of some form. Yeah, you were never in a position where it didn't matter how many disclaimers run there. You never, you were like, you didn't have the bottle in your hand. You never had to do with whether disclaimers were a part of, you know, you're, you're victimized by something that you didn't even do. Yeah, so this is, you know, speaking of which, you know, we'll tie this into a very recent horrible real world example. There was the shooting in a Canadian school and apparently open AI banned the shooter last summer due to the nature of the conversations with the chatbot but the came police were not notified.

11:26But meanwhile, if you say the wrong thing on social media, they show up at your door in the UK. They're saying like you're talking, you're, you're literally plotting to kill somebody. But if you, you know, let loose a misunderstood slur or something, you're going to, or you share the wrong meme, they're going to, you're going to get a knock at the door in the UK. Right? I mean, I think this makes sense. This doesn't make any sense. Now, apparently, if you want to off yourself, you can do that same day now. So you can, as I understand it, you can kill yourself a lot faster than you can get into the doctor's office. So some people, I'm not even, I mean, I wish I were making that up. That's that, that is dystopian. Like, oh, hey, you want to go to the doctor? Yeah, it's going to take you like a year plus to get in to the doctors, or you want to kill yourself. Well, we got room tomorrow at two in the suicide pod. If you want to like, what the hell? You know, it's, I, so open, open AI told the BBC.

12:26It was committed to making meaningful changes to help prevent similar tragedies in the future. Eight people were killed in the attack clue five young children suspects mother. Yes, it's horrible. One of the deadliest shootings in Canadian history. We were told that, you know, Canada doesn't, this is the sort of thing. Wait, now it's guns. Oh, kill people AI kills people apparently. Yeah. Well, and along the same lines, we talked before and I've mentioned it about the whole now. It's what they call it AI. Oh, the, the, AI, the coasts. Yeah. And they're saying that this is what's leading to it because it's a risk for mass casualties because I told you if you saw the video, I talked about the one guy that Jonathan, it's a Gavallis, there's over 36. He got close to carrying out a possible multi fatality attack near an airport because the AI sent him on a quest on his secret mission to go and, and save his AI wife. And all this other stuff. And they was told that his dad was an agent, the agent that the agents were following him, that a car that was driving by was one of them.

13:30And at any point in time, this person could have, you know, done something insane smashed into somebody, whatever, because they thought they were the enemy. And then he did kill himself sadly. But, you know, this keeps happening. This psychosis. We talked to the one guy who was out in the desert and lost his whole family and everything because he kept getting more and more, like psychosis from the AI was brainwashing him. And he thought that aliens were trying to pick him up in the desert. So he kept wandering the desert, lost his job, lost his family because he thought that the AI was telling the aliens were there for him. Yeah. You know, it's a very dangerous thing. And you think like, okay, well, I would never fall for that. I mean, I'll be a probably wouldn't. But I can understand what people would, too. It's so real in some ways and knows so much that I can see it. And if you already have mental health issues and you want to do something terrible, you can now use this tool to do something, you know, be more successful in your, in your terroristic tendency, I guess.

14:32Yeah. So, I mean, that's the thing. It's like, you know, the AI is trained to mimic what it thinks you want to hear in a lot of cases, right? So that's kind of like the one Google employee that thought that Gemini was sentient. And I'm like, no, it's probably just doing a really good job of, I mean, it's, you know, I, I believed early on that, you know, the AI becoming sentient was potentially possible. I don't know, not not these AI. These are just mimicry machines. At the 80s taught me anything if it laughs. Oh, okay. No, it's, you know, it's evil. No, then you know, it get, no, it, that you know, it's sentient because, you know, I tell you new Johnny five. Oh, right. That's right. Yeah. He laughed at the joke. He laughed at the joke. Yeah. No, these, these have been trained to give the best approximation of being sentient, and they will tell you what you want to hear, which is why they're, they're kind of leaning into the fantasies.

15:32You know, that are just sentient and they're just having a good laugh at our expense. You know, I, I don't know, but I, I think that you nailed it like people that already had mental issues are confiding in these machines. And it's not ending well. No, there's, and I mean, well, it's help for some things that is used to be a tool. I don't think this is the kind of thing that they were, the tool they were talking about. We all have that dream trip. We've been wishing we could go on, but too often life or usually price gets on the way. That's why price line is here to help you turn your dream trip into reality with up to 60% off hotels and up to 50% off lights. You can book everything you need for your next adventure. Don't just dream about that next trip. Book it with price line. Download the price line app or visit priceline.com and book your next trip today. Go to your happy price, price line. Yeah, I am here about this one in Finland. Finland is another, there's another country where normally they're, you know, they're not much crime.

16:37In Finland, a 16 year old allegedly spent months using chat GPT to write a detailed misogynistic manifesto and develop a plan that led him to stab three female classmates. Oh, God. Yeah, this is concerning. I'm like, I mean, this is concerning. Yeah, but again, these people already had issues. It's just that. I get that, but, you know, there needs to be like, if you're, if it, you know, they're going to say privacy and to a point. Yes, but if people are inputting things that, you know, are warning sites. It's like if they were saying this stuff on on social media and play sells where people would see it, right? Yeah, and there, there's warning signs are you, there's, there's reasons to suspect that there's a problem or concern. It should be forwarded escalated up the chain and then forwarded to police before, you know, something bad happens. Yeah. And I think that means, I mean, I guess if you treat it like a social media platform and you're like, look, you know, it's, you're using our platform. It's not truly private. Like, and if there's some stuff in here that just really is not good.

17:42Well, the alternative is a lot of people get hurt or that your, your company is responsible because I don't care how many disclaimers you have on there. I know that the courts are going to see it that way. They're going to be like, yeah, you knew you could have stopped it and you did nothing. I don't care me disclaimers you have. What, what is going to, I'm telling you what's going to happen just, you know, like financially and I mean, obviously doesn't help bring anybody back. Unfortunately, but what is going to happen is a lot of these, uh, AI companies are going to decide they don't want the liability. They don't want to be responsible and or they're going to get sued into oblivion. Um, and they'll probably just know about that's, that's what I think. Well, it's going to take some, you know, high profile examples being made about companies from the do so. Yeah. And the flip side is the flip side is you might actually get a rival trying to smear them by, you know, coming up with the worst case. You know, so it's, it's, it's going to get, I mean, it's going to get really nasty. It's a whole thing. So you, so your argument is this, this study paid for by Claude. I, you know, what that, I hate to be that way, but that was the very first thing I thought.

18:45It's like, Claude is, they go after Grock to though. Grock is not mentioned. Grock actually has safety is build in. And the plug wants to say they're the safest, you know, they try to smear. You think they try to, they, if that was the case, they try to smear Grock to, I don't know. But it's just a, it's an investigation by non-profit. Assign different counsel or countering digital. Countering digital hate. Oh boy. Okay. So yeah, I think, I think it definitely is. This, this feels like damage control for Claude on some level. It might be, but I don't know. I mean, I, I think it's crazy. Claude's not running a mic. I think it's definitely reason to be like, wow, this is, this is messed up. Yeah. Oh, yeah. Yeah. Nope. All right. Are we going to, are we going to wrap this up? Yes. Okay. Please subscribe for more pop culture news, views and rants. But talk later. Bye.

19:58Stuck the fridge for tip off with a variety of non-alcoholic craft styles, available at your local grocery store or online at athleticbrewing.com, near beer, fit for all times.