Sunday, February 26, 2023

So I Cheated: ChatGPT


Amid howls of protest in my family—after all, we are a household of teachers—one certain family member, who will remain unnamed to, er, protect his privacy, decided to dive into the dark side. He investigated ChatGPT.

It was an innocent enough endeavor. He simply asked the AI device for some references for a particular esoteric topic he was studying. ChatGPT instantly complied.

After long household discussions about how students will never be able to learn anything, and how the entire culture is going down the tubes, and oh no, oh no, whatever shall we do, the conclusion I gleaned from those discussions was: it's a tool. I see it as a souped-up search engine. Think of it as Google: The Next Generation. Sure, I'm never going to use it to write a blog post here—alright, I know: never say never—but I was curious to see whether I could use the thing to do a simple search of the nether reaches of the Internet. For, say, genealogy.

Since I have been struggling with my research project this month, I asked ChatGPT for a rundown of resources for Rose Berryman, my sixth great-grandmother. Well, that is not entirely correct; I had mercy and added a few delineating terms, like when she was born and who she married. After all, if I didn't do that, the answer might turn out more like search results on Facebook.

The response wasn't too bad. After all, I did provide the location (Virginia) and a date (her birth circa 1708). Her husband's name—Richard Taliaferro—was probably the biggest clue. The instantly-provided paragraph indicated her parents' names as well as those of her husband. Included were the names of six, though not all, of their children.

When asked for references, ChatGPT provided links to specific pages in several books digitized on Internet Archive. Some links to other reference material from Google Books were also provided.

The upshot of this little experiment isn't that I'll now use ChatGPT to research my family tree. After all, some of the names provided in this little foray don't line up with what I've found elsewhere. I'll have to check those out, myself. And that's the point: no matter which resource you use to do your research, you still need to check out the information for yourself. The answer isn't right until it's right. And confirming such details is still part of the routine of genealogy, no matter whether I'm using AI or doing the grunt work myself.


  1. Jacqi - youv'e done it again. You are absolutely "the guinea pig" for the rest of us. Having seen lame examples of ChatGPT on twitter, I had dismissed it as useless, just another parlor trick like the old Magic 8 Ball of our childhood. Only useful for hastening the downfall of our civilization :-) Swore I would never use it. But just like that you shed light on the thing. Now I can't wait to try - hopefully it will be a bit more useful than Google search, which has become weak, somehow.

    1. Well, Lisa, it is a little better than a parlor trick ;)

  2. A lot of AP English teachers have been experimenting with ChatGPT. Their results show that it often supplies very generic, meaningless responses. Sometimes it makes things up. It will also confess when it has written a text. That means teachers could use it in reverse by uploading a text and asking if it is the author.

    1. That's an interesting switch on checking for plagiarism, Kathy! Although I know teachers have other programs to check for that.

      Admittedly, ChatGPT has its immense down side. While the links it provided me in my trial run, though URLs, didn't work, I have long since learned to do my own search for copies of cited material--or even to hack faulty website addresses. There is always more than one way around a roadblock.

      I'm sure the company behind this AI will up their game. This thing is still in its infancy. If it points out a valid resource I haven't yet found, I consider it a plus.

  3. Since my previous comment, I've learned that ChatGPT takes credit for things it did not write. I'm not convinced that ChatGPT checkers are any better.

    I've run a couple of experiments similar to yours with ChatGPT. The first time, it provided completely bogus information and sources. It simply made things up. Then it admitted that it frequently give false information.

    Today, I tried again. This time around, it told me that it does not have research capabilities; then it gave me a mini-lesson on seeking out the information for myself.


Related Posts Plugin for WordPress, Blogger...