AI Industry is a Complete Mess (Cold Fusion Reaction)

The AI industry is a complete mess. Command AI The hugging face incident, the OpenAI where the AI broke out of a sandbox and hacked into hugging face and and whatever, was just an uh a badly controlled experiment, completely irresponsible way to operate a cyber or internet or tech company. And he quotes here um both the primer gen um and the internet of bugs here, so go watch those videos where they effectively uh talk about the same sort of thing. Like this is a this is not a new problem. This is something we should we've been doing since the dawn age of the internet is when we run these tests, we we operate them in a sandbox and we watch them like a hawk. And he also talks about generally the Did you want to explain the Australian government website as well that was hacked in in January or something? Exactly. OpenAI hacked it and they knew about it and then they sent um an email to a very like public email address uh to the government to basically being like, "Yeah, we hacked you." But not actually like following up and being responsible for it uh and that email never got checked. Well, not never, but for month I think it was like for 4 months or something. [snorts] So, almost for half a year that email was not checked and then until got [clears throat] checked and they they they had to see if it's actually for real. It It's a bit hard to even explain or understand how did we land here where these companies do harm and then they talk about it. They make a press release about the harm that they've done and they are like, "Let's make someone responsible." Like, brother, you are responsible. You made an agent. You gave your agent a computer. You gave it a prompt to start with it. And then it did something harmful. It's crazy that they are not held responsible and they act in a way that they shouldn't be responsible for it. And we as like collective humanity have to be ready for uh rogue AI for their basically product being buggy. That that's kind of like is the actual terminology. Thinking in one instant that it was uh shown, they left it to run over the weekend without like even checking it. Like, you know how you don't ship on Fridays? We we know that for websites. Like, you don't ship a website update on Friday because like everybody goes home and if there's a bug, there is nobody to solve it over the weekend and you have a bug [music] until Monday. You don't do that, generally speaking. Uh these companies are doing the these uh this exact thing with AI agents running over the weekend without anyone looking over. And I think he even calls out um and says like OpenAI need to be jailed. Like, they are they're running reckless and whatever and Well, it it was Edward Snowden saying that Oh, yes. It was Edward Snowden, wasn't it? So, he's in the video as well. said, "Yeah, Sam Altman needs to be in jail." And in some you know, we we covered the pace in the frontier last week, which is generally a good direction to kind of go in to have people reviewing the AI for people to catch up and various things like that. But, um it's it's it's not good enough from like uh from a from a technical perspective cuz you had that guy who it was OpenAI, right? Where he left or no, he was at Anthropic for like 6 months who was been open AI for years and years and said, "Look, I'm out because these companies are absolute charlatans and they're running a mockery. They're not doing things safely." And all of this is coming to a head right now and it's it needs to be it needs to be handled. They're moving so much faster. They're running before they can walk and they just don't have the expertise. God knows why they're paying them this kind of these kind of salaries that keep getting thrown around, but these are supposed to be the best in the business, yet they can't operate a simple sandbox experiment and and identify when the the the AI has broken out of its container, you know? And just yeah, again, set something off for the weekend, you know? So, Neville said it the best. I think it's the shortest and the best and super clear. The best way to paste the frontier is to hold the labs fully liable for the behavior of their models. Uh for dangerous or uh weekly guarded open source models, hold the hosts liable. Of course, the end user hold liability as well if they are jailbreaking models to do harmful things. So, here we're talking about like uh three levels of uh liability for like three different groups. Uh one is the the labs. They are making these models and they are giving it to people. But before they give it to people, they test it. And if they test it, they should be liable uh held liable for, you know, their testing in they they should do a proper testing environment. They should have maybe government agencies um looking over. They should have like maybe third-party sources looking over. So, that's important. And then the open source uh models, well, we know they will have hosts. The same thing is happening with pirating movies, right? That the hosts are held liable, right? The same thing is with not safe for work. So, we have figured out a lot of these things. It's not like new. These companies are acting like this is a new thing. It's not. The product is a new product, but the concept is the same. And obviously, the end user should also be held liable. Maybe maybe if I use a model and jailbreaker or I find a way to use that model to to do harm, to scam people, right? Maybe the model cannot even know that, but if I find a way to scam people, I'm liable there. Mhm. But, you know, it's a chain. Wasn't there with the Australian government, wasn't there a public uh page that was made public or whatever that was showing personal health information. And Yeah. was that something that the AI exposed or was that just something he demonstrated where it's not only the AI companies that are being absolutely awful? issue was that even like the government agencies, they are not doing a good job. So, there were bugs that uh were known Mhm. and government is not doing anything to solve them. And by the way, in that video, he showed like he said two things about like two contracts. Like a company got what like 70, 80 million dollars for a website for a government agency? to be 3 million and then when it came to it, it bloated to 70 million and then they got or 90 or or 90. And then they got a second contract after that for 117 million or something crazy, wasn't it? Exactly, it was 117 to fix the things that they got wrong in the first one. It's So, if you want to make money, [snorts] you got to get some contract contract from the government I mean that's that's kind of like [music] how the whole US system right now works. It's it's a different topic, but that's that's kind of like where the money is right now with in the US. This is part of a larger conversation on my show Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter. Command AI