The recent paper showed that Large Language Models and game-playing AI implicitly model the world, but the models are flawed and incomplete. So, they do have a model of the world, which implicitly emerges, but it's not 100% correct and not complete. And that would NOT be amazing, exactly, why? Does any living human possess a perfect and complete model of the world? I don't know of any, do you? We're talking about the equivalent of somebody having just one sense, that of "reading", and through that building a model of the entire world. Of course it's going to be incomplete and not 100% correct, but still the mere fact it exists at all is scientifically amazing and a hint that embodied LLM, one that can "listen", "touch", "smell", "visualize" and physically explore could end up having an idea of the world at least on par with that of humans, if not far superior.

Deepfakes certainly are disinformation but, from the report, Iran is calling on Americans to boycott the elections due to the candidates’ support for Israel, which is not disinformation, it's just the exercise of free speech based on the reality of facts. People should be free to campaign for whomever they want, from wherever they want, as long as what they say stands to truth.

So, to avoid the worst case scenario, you create one yourself.

I wonder: isn't this the same concept that the Star Trek holodeck uses to "make matter" out of thin air?

It appears to me that there's the possibility for this neural network that detects deep fakes to be used as an adversarial network to improve deep fakes generators.

It says something about the politicians involved, that such experiments are being shut down before they come to a end. It already happened in Finland, now in Canada: it almost feels like they're afraid of what the results might tell, doesn't it?