Meta’s AI Scientist critiques OpenAI’s ChatGPT

“Meta’s Chief AI Scientist Dismisses OpenAI’s ChatGPT as ‘Not Particularly Innovative’:

In a recent interview with the New York Times, Yann LeCun, the chief AI scientist at Meta, stated that OpenAI’s ChatGPT program is not particularly innovative in terms of underlying techniques and is not a revolutionary breakthrough. LeCun claims that similar data-driven AI systems have been built in the past by many companies and research labs.

He also notes that ChatGPT is composed of multiple pieces of technology developed over many years by many parties, such as transformer architectures that are pre-trained in a self-supervised manner, an approach that he had been advocating for long before OpenAI existed.

Furthermore, LeCun points out that OpenAI’s program has made extensive use of a technique called reinforcement learning through human feedback, which was pioneered not at OpenAI, but at Google’s DeepMind unit. He suggests that ChatGPT is less of a scientific breakthrough and more of an instance of decent engineering, comparing it to IBM’s Watson computer and Sebastian Thrun’s self-driving vehicle that won DARPA’s 2005 Grand Challenge.

Despite these criticisms, LeCun stated that he is not criticizing OpenAI for their work, and that Meta’s AI team, FAIR, will also be producing breakthroughs in the future, including “generative art” and “creation aids” to help small businesses promote themselves through automation.