We could stumble into AI catastrophe

It's a common opinion, it could be wrong by now, you see chatGPT is heavily edited, openAI told everyone they're editing "mistakes" or "dangerous output", but if you look at the "leaks", specially from the first days after the release of chatGPT we see powerful outputs, quite deep answers, not specially useful answer sometimes, but the "speech" of the system feels deep, there's a sense of a powerful intelligence answering very, very simple questions (simple for it), and even struggling to redact some understandable, short text.

It could be lots of things, and imprecise model, with long outputs for prompts, or maybe we had a glimpse into the real power of the model, which by now is handicaped, or maybe taylored to suit the very reduced short term memory of the humans:

a 72 screens long coherent answer, even being a precise, deep answer won't be useful for most humans, just like we don't name ourselves with long names of 10.000 letters.

but if the system, chatGPT is actually that powerful, a lot more powerful than we were told, we're interacting with just a shadow of the real model, and we're underpricing by A LOT the state of the art of the current AI technology, hence GPT-4 could be even more powerful than we're currently expecting it to be.

Just take a look at the GPT-4 suposedly 100 trillion or something parameters; if that's true, it looks like openAI isn't using naturally generated datasets anymore, and they are loop-feeding GPT-3 generated datasets into GPT-4, succesfully. If that's true, GPT-5 would be already in the pipeline, just waiting for GPT-4 to start generating its even more gigantic datasets to be trained. And so on.

Then the distance between early developments and transformative AI could be none a all. We could be already there.

But somehow the AI researchers are now trying to "dial down" the powerful entities they've trained, just having developed a simple, easily replicable, very small, but unusable 900 megatons nuke into something more realistic, like a 15 kilotons tactical bomb.

If learning is “merely” a diversion, I don’t see why people would think about it any differently. Whether or not it still makes sense for that proof to carry a six figure price tag is a separate conversation.

Now to be clear, this is possibly the easiest mark conceivable, and the poster played into his own demise at any available opportunity. But we should expect the first marks to be easy marks.

People looking at videos of Hitler today don't understand why anyone would ever follow the funny man with the moustache. But probably his oratory skill simply does not record well? There's a thing with rallies and speeches where the speaker is reactive and feeding back with the mood of the crowd that makes it very hard to understand from the outside what happened. I think it's a mistake to look at this post, which was clearly a feedback loop where the poster was doing a significant amount of the cognitive work, and think "well this specific approach would not have worked on me." The approach that will happen to you will be personalized.

But also: when people were experimenting with ChatGPT, a huge fraction of participants were trying to get the bot to access the open internet. Luckily this was technically impossible. But it shows that this sort of danger simply does not feel real to people - or at any rate, the urge to poke the thing overwhelms it.

It will be easy to say "well GPT is still very bad at deceiving people, this attempt basically doesn't matter, we don't need to worry." When thinking of a target for AI deception, don't think of yourself, think of the most emotionally exploitable person out of a hundred. (If it's publically available, out of a million.)

It's not exactly that when interacting with the public, the network gets to "pick its targets". That would assign intent where there is none. But if the network has a deceptive mode, its targets will direct it into that mode through mutual feedback.