Elon Musk lists his three most important ingredients for AI

Elon Musk has again sounded the alarm connected the dangers of AI and listed what helium considers arsenic the 3 astir important ingredients to guarantee a affirmative aboriginal with the technology.

The billionaire CEO of Tesla, SpaceX, xAI, X and The Boring Company, appeared connected a podcast with Indian billionaire Nikhil Kamath connected Sunday.

"It's not that we're guaranteed to person a affirmative aboriginal with AI," Musk said connected the podcast. "There's immoderate information erstwhile you make a almighty technology, that a almighty exertion tin beryllium perchance destructive."

Musk was a co-founder of OpenAI alongside Sam Altman, but near its committee successful 2018 and publically criticized the company for ditching its founding ngo arsenic a non-profit to make AI safely aft it launched ChatGPT successful 2022. Musk's xAI developed its ain chatbot, Grok, successful 2023.

Musk has antecedently warned that "one of the biggest risks to the aboriginal of civilization is AI," and stressed that accelerated advancements are starring AI to go a bigger hazard to nine than cars oregon planes oregon medicines.

On the podcast, the tech billionaire emphasized the value of ensuring AI technologies prosecute information alternatively of repeating inaccuracies. "That tin beryllium precise dangerous," Musk told Kamath, who is besides the co-founder of retail stockbroker Zerodha.

"Truth and quality and curiosity. I deliberation those are the 3 astir important things for AI," helium said.

He said that, without strictly adhering to truths, AI volition larn accusation from online sources wherever it "will sorb a batch of lies and then have occupation reasoning due to the fact that these lies are incompatible with reality."

He added: "You tin marque an AI spell insane if you unit it to judge things that aren't existent due to the fact that it volition pb to conclusions that are besides bad."

"Hallucination" — responses that are incorrect oregon misleading — is simply a large situation facing AI. Earlier this year, an AI diagnostic launched by Apple connected its iPhones generated fake quality alerts.

These included a mendacious summary from the BBC News app notifications connected a communicative astir the PDC World Darts Championship semi-final, wherever it wrongly claimed that the British darts subordinate Luke Littler had won the championship. Littler did not triumph the tournament's last until the adjacent day.

Apple told the BBC astatine the clip that it was moving connected an update to resoluteness the occupation which clarifies erstwhile Apple Intelligence is liable for the substance shown successful the notifications.

Musk added that "some appreciation of quality is important" and that "you cognize it erstwhile you spot it."

Musk said AI should privation to cognize much astir the quality of world due to the fact that humanity is much absorbing than machines.

"It's much absorbing to spot the continuation if not the prosperity of humanity than to exterminate humanity," helium said.

Geoffrey Hinton, a machine idiosyncratic and ex-vice president astatine Google known arsenic a "Godfather of AI," said earlier this twelvemonth that there's a "10% to 20% chance" that AI volition "wipe america out," successful an occurrence of the Diary of a CEO podcast. Some of the shorter word risks helium cited included hallucinations and the automation of introduction level jobs.

"The anticipation is that if capable astute radical bash capable probe with capable resources, we'll fig retired a mode to physique them truthful they'll ne'er privation to harm us," Hinton added.