ChatGPT from OpenAI

Apparently I’m late jumping on the bandwagon. Everywhere I look, someone else is trying out ChatGPT and commenting on it. This app is supposed to respond like a real human being, given any particular prompt you may type.

So I had to try it myself. The results—to put it mildly—surprised me. Before I show you some, a caveat is in order. There are those who say that the AI here is unimpressive because…well, usually because they say “I can write better than it can.” That bar is much too high! Yes, maybe it writes like a 12-year-old, but that’s still amazing. It reminds me of an old joke as well as a more recent true story:

  • The joke is about the guy who says “Look at this! My dog can play chess. Try playing a few games with him.”
    After you do that, you say “He’s not so great. I beat him two games out of three.”
  • And the true story is about the MIT freshman who tried out a calculus-problem-solving AI computer program in 1978, giving it a few indefinite integrals to evaluate. “That’s not so intelligent,” the freshman opined. “It solved the problems the same way I did.”

So I’ll show you how ChatGPT responded to five prompts I gave it, and then I’ll link to a few articles that have already appeared elsewhere:

  • I started with the prompt “Write some limericks about trigonometry.” I was pretty sure that ChatGPT knew what a limerick is and that it knew something about trig. Here was the result. Meter, rhyme, and relevant content were not perfect—but hey! It’s like the chess-playing dog. It’s amazing that ChatGPT could do it at all, even if it looks like the effort of a 12-year-old!
  • Then something harder: “Write a poem about the Boston City Council.” What could it know about the BCC? We ended up with a self-promotional poem that seemed to have been written by a PR firm that the BCC hired to bolster its image! This one is better than what a 12-year-old could do.
  • Third and just as implausible: “Write a short speech about square roots.” We could imagine that it would spit out a few facts and examples, but no, it wrote a pep talk about why square roots are important!
  • Next, we’ll return to poetic forms and apply one to an even less likely topic than the previous ones: “Write a sonnet about choosing HO scale or N scale in model railroading.” OK, ChatGPT was somehow confused about sonnets vs. limericks, but still it got a lot of appropriate content.
  • Finally, a totally implausible prompt that they couldn’t possibly succeed with: “Write an essay about the connections between geometry and the pluperfect tense.” Surely ChatGPT would have nothing coherent to say. And yet…read what it came up with. Symmetry and perspective. Is this the death of assigning essays to be written as homework?

Finally, some links to third-party articles that are worth reading:

Categories: Linguistics, Technology