Trailblazer in Math, Computer Science Lauds the Advances He’s Seen in ChatGPT, GPT-4
In a podcast/video interview on YouTube, Dr. Stephen Wolfram, a groundbreaking polymath with a doctorate in physics and who created his own computer programming language, expressed just how impressed he is with GPT-4. He has a long perspective on AI, since in 2009 he created Wolfram Alpha, a computational knowledge engine used by millions of users worldwide.
He explained to the hosts of the Machine Learning Street Talk that he created a plug-in for ChatGPT (which is connected to GPT-4) to access his powerful Wolfram Alpha computational system so that it can be of greater use. Neural networks generally are incapable of “serious computation,” Wolfram said.
“As of today ChatGPT can sort of get computational superpowers by calling on Wolfram Alpha and Wolfram language through its plug-in mechanism,” Wolfram said. “ChatGPT is a large language model which has kind of gone through the web and lots of books and so on and it is piecing together things by sort of seeing what a reasonable piece of continuing text would be and it’s amazing how far that gets…”
Wolfram said ChatGPT “provides a kind of immediate bridge” via natural language understanding, which is how Wolfram Alpha operates. The groundbreaking computer scientist said that the way the neural network operates isn’t so different from Boolean logic, and it relates well to his computational model.
“It’s discovered kind of a semantic grammar of language,” Wolfram said. “…what it’s doing is it is putting together essentially these puzzle pieces just like when you make us a syntactic grammar of language you’re defining, these are the kind of trees and so on that you can put together these are the parse trees you can put together to form a syntactically reasonable sentence.”
It’s also making leaps of logic and it works in a mysterious way that even its creators don’t exactly understand, Wolfram indicated in his awe of the system.
“Wolfram Alpha now finally gives it the power of full complete Turing,” Wolfram said. “Anything goes, you know, kind of calculation if we now allow it to feedback on itself. So this system ChatGPT plus Wolfram Alpha is now able to, say, put information back into the info sphere, send out tweets, put stuff on a webpage that can then come back through you know the GPT you know engine which as you say we may never understand it, it may just always be kind of a black box to us but it can add its little bit of natural language magic and complete the circle we now have almost this massive Turing machine that has these hybrid components to it. Is this potentially the first step towards artificial general intelligence or are there still some missing ingredients? …it’s not creating new knowledge, (but) as soon as you have actual computation in the loop and what I would talk about as irreducible computation, you can build what amounts to new knowledge.”
Wolfram expects that ChatGPT, with GPT-4, will lead to breakthroughs for humanity that as yet can’t be predicted, because it will find connections that we haven’t in solving problems.
“It took me a number of years to kind of come to terms with that discovery that even extremely simple rules are capable of generating extremely complicated behavior,” Wolfram said. “That’s a phenomenon of the computational universe that we’re not particularly used to from our experience in engineering because in engineering we usually we want to do things where we can foresee the output. But so what happens is in the computational universe kind of there are immensely complex things that can be produced even with very simple programs…there’s an infinite supply of computational creativity. “
Wolfram said that ChatGPT with GPT-4 has already made breakthroughs that he didn’t expect to happen in his lifetime.
watch the discussion at youtube.com