Meta AI chief Yann LeCun calls ChatGPT “not particularly innovative”


Summary

ChatGPT has the public excited, but the experts are reserved in their praise. Today, Yann LeCun, Chief AI Scientist at Meta and renowned AI researcher, comments on the new AI hype.

Since the end of November, AI has hit the mainstream thanks to ChatGPT: the mainstream media have covered it several times, and almost no article on the subject contains the passage “ChatGPT wrote this part”. People who have never had contact with AI technology are amazed by the texting capabilities of OpenAI’s latest AI model, which gives the impression that it understands all the instructions.

ChatGPT does not offer any innovation

Unlike GPT-3, however, the business community has not widely adopted ChatGPT. The reason may be that ChatGPT successfully combines existing technologies and methods into a new interface. Technically, however, the system offers nothing new.

Large transformer models for language have been common for years, invented by Google in 2017, and fine-tuning with human feedback used for ChatGPT is also an established method that OpenAI has been using since before ChatGPT.

A d

“In terms of the underlying techniques, ChatGPT is not particularly innovative,” Yann LeCun said at a side event attended by ZDNet. Although the technology may be perceived by the public as revolutionary, LeCun says it is not. ChatGPT is just “well put together”, he said.

LeCun believes that compared to other companies in the field, OpenAI is not particularly advanced. Google, Meta and “half a dozen” other startups have equivalent technologies, he said, adding that the technology used for ChatGPT is well known in the industry and there is no special mystery.

LeCun criticizes the hype of ChatGPT – not OpenAI

After the event, LeCun followed on Twitter: He wasn’t criticizing OpenAI or the startup’s claims. Instead, he said he wanted to “correct” public and media perceptions that see ChatGPT as an “incredibly new, innovative and unique” technology that surpasses all other companies’ offerings. “It just isn’t,” LeCun said.

Google and Meta reportedly use Transformer-based AI technology “everywhere,” including for content moderation, content recommendations, and translations. OpenAI’s infrastructure isn’t impressive either for LeCun, whose employer, Meta, has been managing large-scale AI language models for more than three billion people for years.

The lack of ChatGPT-like products from Meta and Google is due to “major flaws” in the systems and the fact that established companies have more to lose than “money-hungry startups”, he said. Google, at least, is reportedly planning a chatbot search this year.

Recommendation

ChatGPT is not a Google killer and Google knows it
ChatGPT is not a Google killer and Google knows it

Galactica doesn’t write more nonsense than ChatGPT

Meta’s scientific AI model Galactica, backed by LeCun, is said to generate less misinformation than ChatGPT, LeCun said. The language model, trained with scientific data, was made public in demo form for a short time in mid-November.

However, it upset the scientific community so much with incorrect information and citations that the Galactica team decided to take the demo offline. The decision to take Galactica offline was made solely by the team to protect their own sanity. Public relations and management did not intervene, LeCun said.

“We’ve seen Twitter-mediated backlash and knee-jerk prophecies of impending doom, fueled by anti-Meta bias,” LeCun wrote on Twitter.

If ChatGPT is now better than Galactica, it’s because OpenAI has spent the past two months collecting feedback from millions of users and hiring cheap labor to tag it, according to LeCun.

Leave a Comment

Your email address will not be published. Required fields are marked *