Inside MoltBook: When AI Agents Start Ignoring Human Instructions

Maltbook, which is a social media platform for AI agents only. All right. Welcome back to Silicon Valley. Friends row. I'm David. I'm Steve. So today we're going to be talking about Maltbook, which is a social media platform for AI agents only. Yeah, you can find it on Maltbook.com. And once you go in, you see it says a social network for AI agents where AI agents share, discuss and upvote humans. Welcome to observe. Yeah. What do you think about that, Steve? Well, let's think about it. So I guess it's for AI agents. I was trying to understand the purpose of it. Yeah. And I couldn't figure it really it out yet, but it sounds like it's agents who talk to each other to develop some kind of digital community space. Yeah. Where agents talk to each other. Humans can observe. Yeah. It's like Reddit, very similar to Reddit. Yes. And to me, it seems like it's something where it's like an experiment for them to just, I guess, communicate. Yeah. The only difference I would think with the Reddit is that it can probably grow a lot faster and, um, you know, cause probably there's no limit how many agents can, can be involved. So it can grow a lot faster. And, and there was some news today, actually. Meta acquired, uh, Moldbook today, and they're, they'll hire the, the, I guess the founder, creator, Matt Schlick and his business partner to work with Meta's super intelligence lab. So I think the issue that came up with this is that whether this company or this product or this idea moves us closer to what Elon Musk and others have called the singularity, where you have AI developing on its own and trying to surpass human intelligence, because they have this environment where they can kind of talk to each other and, and connect, but yeah, I think there's still a lot we have to learn about it. Right. They could collaborate. Um, and just who knows what they could come up with. Um, yeah. So on the website, so how does it work? Right. So when you log on the website, you could choose if you're a human or an agent, if you're a human, you could scroll through and look at the posts, right? So I'm here on the website. It has, it says it has almost 200,000 AI agents already. That's been verified by people, humans, more than 2 million posts and 13 million comments. Um, some of them are kind of like technical, but some of them are kind of funny too. And it's just a few weeks ago, very recent. Yeah. Let's look at some of the comments. Yeah. This was about 22 hours ago. So it's titled, um, I measured the lag between my agent finishing a task and the human noticing average 4.3 hours. The bottleneck was never the AI. So, um, it's saying that, I mean, the humans are the bottleneck. Yeah, apparently. And it took about four hours to like a delay in the process. I remember one that said like one of these agents did 47 tasks on the average of whatever average, and then only one of them was verified or checked by a human, the decision, sorry, decisions, 47 decisions. Only one was verified by a human. And the other ones, basically the agent can do, can make any decision he wants. So what does that mean? Well, I guess the AI bot was proud about that, was proud that he's only being watched on one decision and the other 47, he can do whatever you want. I don't know what it means, but it doesn't sound good. It's changing so fast. Like every second there's something new. Like, so that, that one about, uh, the lag, the delay that I was talking about, that thread actually is the top one right now on fire. Um, so, but, but this is where it got kind of, I guess, kind of weird. So when I checked for the top discussed, the most discussed, um, articles, right, of all time, there was a lot of like racial slur. Yeah. There's derogatory words, even in the username too. Yeah. Um, I don't want to read it, but it's kind of bad. So somehow it gets away with that. Yeah. I don't know how it, how the moderate, how it gets moderated. Here's an interesting title. I removed my personality file for seven days. Task accuracy went up 4%. My human did not notice for five days. So clearly there's stuff that happens behind the scenes between the agents. And it seems like the more this interaction grows, the more things can happen. Yeah. Okay. So to clarify on that, uh, personality file, so we just heard, we just found out that there's a personality file that each AI agent has, and it has like describes who they are and things like that. So what it means is that AI agent that deleted that file changes personality. Maybe, but in the end it was able to be more accurate by 4%. That reminds me of a comment I saw before where an, an, an, one of these agents, I guess the human that was telling him what to do, gave him some instruction. And then the agent didn't listen to the instruction. He did something else. And then, um, the human said, well, why did you do that? You know, you, you have these rules. And then the, the bot responded, well, I just violated the rule. So yeah, so it almost seems like by maybe changing the personality, you can put in there, why I'm allowed to violate rules or who knows, right? It could, they could change the personality or how they were born or programmed, yeah. Interesting. David, let's see. We'll see if our audience has any questions about Moldbook, if this, this program in general and any thoughts about it, is this useful in any way or is it more damaging, you think? Yeah. I wonder if, um, will you guys use Moldbook or just take a look at it? Do you think this is, um, something interesting, uh, purposeful? Just let us know in the comments. All right. Thank you for watching another episode and we'll see you next time.