Elon Musk says his TruthGPT AI is 'unlikely to annihilate' humanity

I really want to take Elon Musk's claims to be creating "a maximum truth-seeking AI" called "TruthGPT" at face value. I really want this all to come from a place of honesty and integrity and trying to create a better, fairer world. I really want to believe that in whatever form it eventually takes, Musk's new AI venture will legitimately create "an AI that cares about understanding the universe."

I'm ready to get on board, but he's just making it so damned hard.

Especially when the source for this new information comes from a Tucker Carlson interview on Fox News (via TechCrunch). It's doubly tough when the interviewer, in trailing the interview spot on Fox and Friends, commends Musk's sense of humour and the fact he doesn't take himself too seriously then uses this phrase with a straight face:

"People who take themselves too seriously, like Stalin… or Chuck Schumer, make me uncomfortable, and should make us all uncomfortable."

That's some powerful deadpanning, Tucks. At least give us a wink at the end, eh?

Anyways, that's just an arresting side note to Musk effectively announcing on Fox News that he's committed to creating his alternative to OpenAI and Google's own generative pre-trained (GPT) AIs.

"I’m going to start something which I call TruthGPT," Musk tells Carson. "Or a maximum truth-seeking AI that tries to understand the nature of the universe. 

"And I think this might be the best path to safety in the sense that an AI that cares about understanding the universe is unlikely to annihilate humans because we are an interesting part of the universe."

I mean, that sounds lovely. A kind, thoughtful, caring artificial intelligence that is at best "unlikely" to have us all inhumed. Of course, there's always a chance it will annihilate us, but them's the risks when you're playing with AI.

Which I guess is why Musk and Woz have both joined up to call for a halt on AI experimentation. Though Steve Wozniak, as far as we know, hasn't gone out and bought millions of pounds worth of AI-powering GPUs, founded a new company called X-AI, and promised on national TV to create a new generative pre-trained AI.

The thing is, for all the right-baiting appearances on Fox News, and the disturbingly close nomenclature of TruthGPT and a certain orange ex-president's abortive social network, the idea of countering ChatGPT has value. After the Microsoft deal OpenAI has shifted away from its original not-for-profit open source foundations and into a more, well, for-profit basis.

And when you're doing AI for profit instead it does kinda feel like there's more of a chance of it being steered away from chasing something like a basis of universal truth. Though if you are looking to course correct against a rival AI project that also implies that you're maybe going to be adding your own biases to the model, too.

Especially when you're appearing above a banner that reads: "Elon Musk: Left-wing programmers train AI to lie."

Probably need to bring in some right-wing programmers to tell the truth then, eh?

We don't know what Musk's true plans are for his AI project, but it is real whether or not it does end up sporting the TruthGPT name he first coined on Twitter. He's casting about for investment, he's hired former DeepMind employees, and he's marketing on Fox News.

But what actual form it will take could take a while to fully discern.