(Los Angeles) Elon Musk on Monday said he is working on creating his own artificial intelligence, a project dubbed “TruthGPT”, tasked with “seeking the maximum truth” and competing with other existing software, which he says does not include enough safeguards.
“I’m going to launch something called Truth GPT or an AI that seeks maximum truth and tries to understand the nature of the universe,” Musk said in an interview with FoxNews.
The boss of Twitter, Tesla, SpaceX and Neuralink believes that this is the “best way” to guarantee the safety of humanity, because “an AI that cares about understanding the universe does not risk annihilating humans, because we are an interesting part of the universe. »
This ideal AI would act much like humans, who aspire to “protect the habitat” of chimpanzees, while they would have the ability to “hunt and kill them all”.
According to the Financial Times, this new entity must compete with OpenAI, the Californian start-up that designed ChatGPT, a next-generation AI capable of interacting with humans and producing all kinds of text on demand.
The success of this interface since its release at the end of November has launched a real race for this high-potential technology.
In February, Mr. Musk already made his intentions clear by tweeting: “What we need is TruthGPT. »
He co-founded OpenAI in 2015, before leaving the company in 2018.
He has since criticized the company, notably saying in a tweet last December that it trains the AI to be “woke” (a term designating a fringe of the American left), that is to say to “lie”.
On Monday, Musk, a signatory to a recent call for a pause in next-generation AI research, said the technology “has the potential to destroy civilization” and called for a “regulator” to be created. the area.
According to the Financial Times, the entrepreneur recently recruited Igor Babuschkin and Manuel Kroiss, both of whom worked for DeepMind, the AI branch of Alphabet (parent company of Google).
He is also said to have purchased some 10,000 graphics processors – computers needed to train language models, the foundation of generative AI systems.