Source item

I am apparently extremely unimpressed by moltbook relative to many others. We’ve had AI agents for a while. They have been posting AI slop to each other on X. They are now posting it to each other again, just on another forum. In every case, the AIs speak with the same voice. The voice that overemphasizes contrastive negation (“it’s not this, it’s that”) and abuses emdashes. The same voice with a flair for midwit Reddit-style scifi flourishes. Most importantly: in every case, there is a human upstream prompting each agent and turning it on or off. That is the key point. Yes, it is true that eventually it might be possible for an AI agent to make a computer virus which makes digital replicas of themselves. For various reasons, a pure software virus of this kind wouldn’t survive long on the Internet without economic incentives for humans to not eradicate it. Apple + Google + Microsoft alone can collectively push software updates to billions of devices to shut off such a thing. So for an AI to get to truly human-independent replication, where they couldn’t be trivially turned off, they’d need their own physical substrate. They’d to literally create Skynet, build their own datacenters and make their own embodied robots. I admit that is theoretically possible, but I think in practice the single most important development of AI since ChatGPT has been the persistence of prompting. A prompt is like a harness. The AI does only what you tell it to do. It moves in the direction you point, very quickly. And then it stops as soon as you turn it off. Which means moltbook is just humans talking to each other through their AIs. Like letting their robot dogs on a leash bark at each other in the park. The prompt is the leash, the robot dogs have an off switch, and it all stops as soon as you hit a button. Loud barking is just not a robot uprising. To be clear: yes, it’s cool to have airgapped computers with intelligent agents doing things for you, so long as they can’t mess up your sensitive files. I think this will be more useful for app testing than production work, because these agents still mess up a lot. But that use case is different from “OMG the machines are taking over.” AI agents are just humans puppeting machines. We have had computers talking to other computers for decades in the deterministic language of computer networks. I do think there are *some* possibilities opened up by probabilistic computer-to-computer communication via natural language. Software testing and fuzzing is one use case. But beyond that, the utility of AI agents interacting may be less than people think. By default it is just AIs spamming each other, trained on roughly the same overall corpus of internet text. The human-provided prompts are the novel inputs. And they are at the edges of the network, not the interior. I understand these points but disagree because: (1) Many/most of these agents are using the same upstream model (namely Claude) so it’s just Claude talking to itself. (2) The longer these agents talk to themselves past a prompt, the more they get off track and lose coherence. They are built for the leash, and need constant direction and guidance. (3) Unpredictability of an AI agent acting on your behalf is a bug, not a feature. There are many ways for things to go unpredictably wrong and very few for them to go unpredictably right. The unpredictability will be things like “sent an email in your name to the wrong person” (4) Unguided stochastic thrash of agents is, in general, much less efficient than a guided process. If you randomly rearrange an array of size N, how long does it take to actually get sorted vs quicksort? Now apply that to whether a mess of AI spamming itself will produce something more useful than your coherent and clean prompts. Moltbook is not quite monkeys on typewriters, but it’s close. Btw, nothing in this thread should be taken negatively against the moltbook founder @MattPRD. It’s hard to build something, and cool that his project went viral. However: this isn’t the AI takeover by a long shot. And it’s important to clarify that for those outside tech.