While it is still fresh in my mind, I want to describe the conversation I had with a publicly available version of ChatGPT, the much-discussed large language model, LLM, for conversational artificial intelligence.
A newer version of ChatGPT than the one I used is available to selected journalists and computer industry observers. Their reaction has been, shall I say, mixed. At first, some of the reports were positive and even ecstatic. Most of these early reports were from people whose primary concern is the stock market, not AI technology. Many of the more recent and more careful reports have been negative and critical.
I have given ChatGPT the nickname "Chad". Some early users of one of Chad's competitors asked deliberately provocative questions and encountered a cranky alter-ego named "Sydney". I didn't ask my Chad any unfair questions and didn't encounter any Sydney's.
I have been a long-time observer of -- although not a contributor to -- artificial intelligence research.
- I took AI courses at Stanford from the inventor of the field, John McCarthy.
- Bill McKeeman and I entered McCarthy's chess program in a San Francisco Chronicle's by-mail chess contest. (The computer did not fair well.)
- I used Joe Weizenbaum's computer program Eliza, the very first Chat Bot.
- I was once married to someone with a PhD in AI.
- Symbolic Math, like MathWorks now does in our Toolbox, and which we used to do with Macsyma and Maple, was once considered AI.
- The wild animal trail cameras project that I did a few years ago with Heather Gorr and Jim Sanderson uses modern machine learning and AI.
I talked to -- I should say chatted with, or actually typed at -- Chad for most of an hour on a recent evening. Think of Stephen Hawking in a Web Browser. Chad's responses were always courteous and conversational. And the majority of Chad's responses were factually correct.
However, a surprising number of Chad's answers were just plain wrong. Some examples:
- I was neither born in, nor raised in, rural New Mexico.
- I did not get a PhD from the University of Michigan. Neither did my wife, Patsy.
- Patsy is not the mother of my daughter Kam, nor of my sister Betsy.
- Patsy is not an expert in computational fluid dynamics.
- Ken Kennedy did not coin the term "embarrassingly parallel."
- Pam McCorduck never worked for the Santa Fe Institute.
- I cannot find any reference to the guys Chad claimed improved Sympletic Spacewar.
All of Chad's incorrect responses could have been fact-checked by simple Google queries.
Everybody wants to combine the conversational style offered by large language models with the reliability and breadth provided by Google. Now THAT sounds like a really good idea.
Artificial intelligence has a history of flush successes interspersed with fallow periods known as AI Winters. I am afraid that continued obsession with Chat Bots might lead to another AI Winter.
Steve Eddins revealed the Truth behind ChatGPT with this internal MathWorks Yammer post.
To leave a comment, please click here to sign in to your MathWorks Account or create a new one.