ChatGPT has been on everyone’s lips for several weeks. Among its supporters and detractors, the chatbot developed by OpenAI continues to unleash passions. If the general public considers it a revolutionary tool, due to its ability to understand human language, produce texts or even simulate conversations naturally, this is not the case for Yann LeCun at all.
This Turing Award-winning Frenchman, a pioneer in the field of artificial intelligence and currently chief scientist at Meta’s AI department, hardly shares the enthusiasm for ChatGPT and has not hesitated to make it known. “In terms of the underlying techniques, ChatGPT is not particularly innovative,” he said during a Zoom Q&A session, ZDnet reports.
There’s nothing revolutionary about ChatGPT, although that’s how the public perceives it because, you know, it’s well designed,” he added, arguing that AI systems that, like ChatGPT, rely on deep learning to work, have been around for ages. long time ago.
OpenAI is not a pioneer in this field
Furthermore, the general idea that OpenAI is the only company in the world that produces this type of tool is inaccurate, he also pointed out.
To support his point, the scientist pointed out that GPT-3, the program on which ChatGPT is based, is made up of multiple technological elements developed in recent years by different actors. In particular, the concept of “Transformers”, a deep learning technique introduced by Google researchers in 2017.
ChatGPT is backed by years of research
Yann LeCun continued his demo by also explaining that work on such language programs actually goes back several decades. “Large language models, like the first neural network language model, were designed by Yoshua Bengio about 20 years ago,” said Yann LeCun, referring to the director of the Quebec Institute of Artificial Intelligence (Mila).
Yoshua Bengio’s work on the concept of attention was later taken over by Google for the “Transformer” and has become an essential part of all language models.
The OpenAI program also makes use of a technique called “reinforcement learning through human feedback.” This process uses human agents to help classify the machine’s results in order to improve them. This approach was not pioneered by OpenAI, but by Google’s DeepMind unit.
Changing public perception on ChatGPT
After his demonstration, Yann LeCun clarified, the next day, in a tweet, that his intervention was not intended to devalue the work of OpenAI but to open the eyes of the public “who perceive ChatGPT as a unique and innovative technological advance, ahead of all , which is simply not the case.
Finally, according to Yann LeCun, “if Google and Meta haven’t introduced tools that look like ChatGPT, it’s not because they can’t. It’s because they don’t want to.”
Source: BFM TV
