Llama becomes scam's ally

"You absolutely should not trust AI. It is a very powerful tool that makes mistakes," [David] Gerhard, who has studied artificial intelligence and social media, said in an interview.

He likened it to "a hyper-intelligent eight-year-old that desperately wants to please you. It knows everything about everything, but it will give you a wrong answer rather than say, 'I don't know.'"

The AI software that Meta uses in the Messenger app — where [Dave] Gaudreau went to confirm the fake support number — is called Llama 3. Like the better-known ChatGPT, it's what's known as a "large language model," Gerhard said.

"Llama and other large language models are trained on everything. They're trained on all of the internet. And the problem there is that there's a lot of bad information on the internet," he said — which also means they don't have an understanding of truth. "They don't know what truth is or how to get to it," said Gerhard. "So don't use a large language model to verify facts." 

Gerhard says Gaudreau's case is troubling because Facebook AI gave him incorrect information that led to him getting scammed. "The fact that it's very difficult to find a way to talk to a human in these companies is problematic," he said.