Elon Musk sees AI as a "pretty dangerous technology"

Enlarge image

Elon Musk: "I'm a little worried about the AI ​​stuff"

Photo: JONATHAN ERNST / REUTERS

Earlier this week it was said that Elon Musk is currently trying to recruit top researchers in the field of artificial intelligence (AI) to set up a new laboratory and develop a competitor to OpenAI's ChatGPT.

This was reported by The Information, citing insiders.

On Wednesday, according to Reuters, in a more than three-hour presentation for Tesla investors, he said: "I'm a little worried about the AI ​​stuff."

The technology stresses him.

"We need some kind of regulator or someone to oversee AI development," Musk said.

'To make sure it works in the public interest.

It's quite a dangerous technology.

I'm afraid I did some things myself to speed it up."

The head of his department for autonomous driving is said to have been on stage with Musk – Tesla itself relies heavily on AI in this area, which Musk says is “obviously useful”.

API access: Around 750 words of ChatGPT costs 0.0002 cents

Also on Wednesday, OpenAI, once co-founded by Musk as a non-profit company, released the promised interface (API) to ChatGPT.

The chatbot, which is now used or at least tested by more than 100 million people and can generate text and code, is thus further monetized.

So far, the training and operation of the AI ​​​​model OpenAI has cost a lot of money, the most important investor Microsoft provides its supercomputer and cloud infrastructure.

At the beginning of February, OpenAI then offered premium access to the bot for $20 a month.

Access to the interface costs 0.002 cents for every 1000 tokens, which is about 750 words that ChatGPT generates.

Snap, operator of Snapchat, and Shopify are among the first customers.

OpenAI also changed its terms and conditions at the same time.

Texts and data that ChatGPT users send to OpenAI via the API are - unlike previous inputs - not used to improve the model - unless the users give their explicit consent.

Microsoft researchers have meanwhile - apparently without the help of OpenAI - presented a new AI model that can process not only text but also image inputs in various ways.

Such so-called multimodal models are considered an important step towards an AGI – an artificial general intelligence that can also solve tasks for which it was not trained.

The development of an AGI is the declared goal of OpenAI.

pbe/Reuters