Meta’s chief artificial intelligence scientist says ChatGPT is “not particularly innovative” and “nothing revolutionary”

yann-lecun-at-groupi-jan-2023.png

Why hasn’t the public seen programs like ChatGPT from Meta or from Google? “The answer is that both Google and Meta have a lot to lose by putting out systems that make things,” says Yann LeCun, Meta’s chief artificial intelligence scientist.

collective[i] weather forecast

A lot of ink has been thrown lately about the enormous promise of OpenAI’s ChatGPT program for generating natural language utterances in response to human prompts.

The software strikes many people as so new and interesting that ChatGPT must be unique in the universe.

also: What is ChatGPT and why is it important? What you need to know

AI scientists disagree.

“In terms of underlying technologies, ChatGPT isn’t particularly innovative,” Yann LeCun, Meta’s chief artificial intelligence scientist, said at a small Zoom meeting of press and executives last week.

“Nothing revolutionary, even though that’s the way the public looks at it,” LeCun said. “This is just, you know, well put together, well done.”

Data-driven AI systems have been built in the past by many companies and research laboratories, LeCun said. He said the notion that OpenAI was alone in its type of business was inaccurate.

also: People are already trying to get ChatGPT to write malware

“OpenAI is not particularly advanced compared to other labs, at all,” LeCun said.

“It’s not just Google and Meta, but there are half a dozen startups that basically have very similar technology to them,” LeCun added. “I don’t want to say it’s not rocket science, but it’s rarely shared, and there’s no secret behind it, if you will.”

LeCun pointed out the many ways that ChatGPT and the software it builds on, OpenAI’s GPT-3, is made up of multiple pieces of technology developed over many years by many parties.

“You have to be aware that ChatGPT uses pre-trained Transformer constructs in this self-supervising manner,” LeCun noted. “Self-supervised learning is something I’ve been advocating for a long time, even before OpenAI existed,” he said.

also: Can AI detectors save us from ChatGPT? I tried 3 online tools to find out

“Transformers are a Google invention,” LeCun noted, referring to the linguistic neural network that Google unveiled in 2017, which has become the basis for a wide variety of language programs, including GPT-3.

LeCun said work on such language programs goes back decades.

LeCun said, referring to the head of the Canadian MILA Institute for Artificial Intelligence. Bengio’s work on the concept of interest was later picked up by Google for the Transformer and has become a central component of all language models.

Moreover, OpenAI has made extensive use of a technology called reinforcement learning through human feedback, which gets human agents to help rank the device’s output in order to improve it, just like Google’s PageRank for the web. He said this approach was pioneered not by OpenAI, but by Google’s DeepMind unit.

“So there’s a whole history to this, and it didn’t come out of the blue,” LeCun said, which means, ChatGPT.

also: OpenAI’s ChatGPT is frighteningly good at my job

LeCun said ChatGPT is less a case of scientific breakthrough than an example of decent engineering. He compared the software to IBM’s Watson computer that competed in the 2011 game show Jeopardy!, and Sebastian Thrun to the entrepreneur’s self-driving car that won the 2005 DARPA Grand Challenge. LeCun said Thrun’s award-winning technology “wasn’t particularly innovative in terms of science.” Basically, it’s very well designed.”

“That’s kind of what OpenAI did,” he said. “I will not criticize them for that.”

LeCun was an invited speaker for an hour and a half talk hosted by the collective[i] Outlook, an interactive online discussion series organized by The Collective[i]which describes itself as “an artificial intelligence platform designed to improve B2B sales.”

also: ChatGPT’s Next Big Challenge: Helping Microsoft challenge Google Search

LeCun made his remarks about OpenAI in response to a question during the seminar posed by New York Times journalist Cadi Metz. Metz asked if the Meta AI team, FAIR, which LeCun created, would be set in the public mind by breakthroughs in the way OpenAI is.

Lacon replied, “Will we see this from Meta? Yes, we will see this.” “Not only creating texts, but also aids to creation,” he said, including “generative art,” which he said, “I think it’s going to be a big thing.”

He said Meta will be able to help small businesses promote themselves by automatically producing brand-promoting media.

“There are approximately 12 million stores advertising on Facebook, most of them mom-and-pop stores, and they don’t have the resources to design a well-crafted new ad,” LeCun noted. “For them, generative art can help a lot.”

also: How to use DALL•E 2 to turn your wildest imagination into tangible art

At another point in the conversation, LeCun remarked, “You might ask the question, why aren’t there similar platforms from, say, Google and Meta,” referring again to ChatGPT.

“The answer is that both Google and Meta have a lot to lose by putting up systems that make things,” LeCun said with a laugh.

LeCun is the winner of the 2019 Turing Prize for his contributions to computer science, the equivalent of the Nobel Prize in computing, along with MILA Bengio and University of Toronto professor and Google Fellow Jeffrey Hinton. The three helped pioneer the deep learning era of artificial intelligence.

OpenAI is funded by Microsoft, which has exclusive access to the code produced by the startup, and which gradually integrates the software into its various software offerings, including the Azure cloud service.

Leave a Comment