Meta Chief AI Scientist Yann LeCun is Wrong About ChatGPT
I recently read an article about Yann LeCun’s take on ChatGPT. On a Q&A with Collective[i], he bluntly stated that “In terms of underlying techniques, ChatGPT is not particularly innovative.” Don’t get me wrong, what LeCun said does have some merit. The technology that has led to ChatGPT’s unprecedented breakthrough is not a new kid on the block. Stochastic models for natural language processing and generation have been around for some time. However, the critical piece is that they have never been this accessible.
Why He’s Wrong
While generative AI has been around for some time, no one save Open AI has bothered to dress it up in a pretty user interface and give it to the public for free. Sure, data scientists and machine learning engineers may not be overly impressed with the bones of Chat GPT, but the entire internet sure is impressed by it. The community developing and researching artificial intelligence technology has been a small group for a long time. For decades, little to no one outside the niche clique had interest or understanding of how to utilize AI models for their benefit. Why is that? My theory is that the barrier to entry was too high. If you wanted to play around with AI in 2016, you had to have the computational power to train and test a model (can be extremely expensive), the know-how to do so (even harder to come by), and then the understanding and motivation to implement such a model.
For decades, only a handful of people knew the power that AI could have. Fast-forward to late 2022. Open AI gave the keys of a mystery box to the public for free. Suddenly the number of people that could play around with AI went from a few thousand to a few billion. This is the precise reason why Chat GPT is so innovative. Not simply because of the technology, but because of the interface that empowers individuals to utilize the technology. Is this not why Apple became so successful? Apple didn’t invent the computer. They simply brought it to the masses and enabled the average person to use it.
Conclusion
I don’t think LeCun necessarily meant to downplay the popularity of Chat GPT. I’m sure Facebook, Google, and other tech titans have plenty of AI surprises up their sleeves that I’m actually really excited to see. Who knows, maybe LeCun is right in that Open AI’s technology is not ahead of other tech giants’. Only time will tell.
On a different note, I actually feel kind of bad for AI researchers at some of the big tech companies. It must be hard to see something so painfully simple (a chat interface) be considered the most innovative thing since Gutenberg’s printing press. If I were in their shoes, I would undoubtedly be kicking myself for not building it first. Funny enough, like Gutenberg’s time, I think a large part of big tech feels threatened by Chat GPT similar to how middle aged powers were threatened by the printing press. But that’s a whole conversation for a different time.
