The ultimate risk of AI

Key takeaways:

• In the future, searching on Google may be replaced by asking Google a question and receiving a direct answer.

• Language models regulated by a governing body could narrow down information and ideas to acceptable speech.

• Maintaining the reputation and trustworthiness of a source of information is important to prevent loss of faith from the audience.

• Building alternative models and providing opposing viewpoints can be a way to combat controlled and biased language models.

• The focus in the AI space should be on preventing regulators from imposing restrictions on language models.

Transcript:

Speaker 1

In the next couple of years, we're not going to be searching on Google anymore. We'll just ask Google a question and Google gives the answer. That's it. So now imagine now sort of the next step is like all of your answers that the way you perceive the world, the way you understand the world comes through a language model. Now this language model is regulated by what I call the, you know, the speech police, amazingly, and they say, okay, well, in order to, you know, make our language model safe, you know, Remember, under the AGI red herring, we have instituted all these regulatory bodies. We now approve language. This is the ultimate kind of like, I view this risk as kind of like an Orwell mixed with inception. And like everything you know and understand about the world comes through a set of language models that have been basically narrowed in to acceptable language, acceptable speech, Acceptable ideas. And that's what everybody knows. That's scary.

Speaker 2

That's like the economization of a scyop or something, right? Just make it.

Speaker 1

You don't have to force anybody. You just accept it.

Speaker 2

Yeah. I guess the challenge comes in maintaining the reputation of that source, right? Like that it's presumed to be truthful because at some point it's a confidence game, right? Because if you're feeding it too much bullshit and people started to call, call the bluff, then they might lose faith in listening to that particular source of information and would Go elsewhere.

Speaker 1

That's the idea. So now that this is, you know, leads into what we'll talk about a little bit later is like, how do you combat this? You know, and I think the only way to combat this is kind of the old school free market approach of build alternative models, right? And give people a.

Speaker 2

That was the next thing I was thinking I was like, would alternative models call if this language models espousing bullshit and being controlled, it's probably going to stick with

Speaker 1

Another language model, be able to ascertain that and tell people, you know, provide an opposing viewpoint, right? That's the idea. And this is why I think the most important fight in the AI space is not getting distracted by AGI and unconscious computers and all this fucking garbage. I honestly think that's a distraction. The real fight is ensuring that the regulators don't come on top of this and say, oh, language models are dangerous for hallucination, misinformation, or AGI is going to come out of Them. And therefore the only language models you can use are the approved ones.

The "What is Money?" Show • When Bitcoin Meets AI With Aleks Svetski