ChatGPT and the New York Sunday Times Crossword

A Test

You should know by now that I am interested in the problems us old folks have with our failing faculties, particularly memory and cognition. In my last Substack, I wrote about trying to sharpen my computer talents, by practicing and “operating” as many computer applications as I could, in order to take advantage of my computer’s memory-like and cognitive-like functions and, overall, become more proficient. (I like to juxtapose the idea that a computer is to my brain, as my reading glasses are to my vision; both are simply a means of improving the functioning of *me*. Computer’s have applications; I have ‘appliances’. A computer is one of my appliances, like my reading glasses, pacemaker, CPAP machine, and daily doses of medications.)

When I wrote the prior Substack, I also planted my unproven notion (or hope, that’s fair) that working with these various computer applications, i.e., learning and understanding them, is healthy exercise for the brain and promotes the growth of new nerve cells, i.e., neurogenesis.

With that in mind, I thought I would give you my take on ChatGPT; and report a little ‘test’ of its performance against the New York Times Sunday Crossword. (Caution: if you haven’t ‘done’ this puzzle, and want to, then stop reading.)

First of all, let me say *I Love* ChatGPT. I use it everyday; in fact I have configured the startup screen for my browser to display an icon calling up ChatGPT. I predict, if it’s not there already, that soon almost every application will show a ChatGPT (or A.I.-like similar) field on their screens, operating like ‘Search’ does at present. It’s really handy to have, particularly if you are writing. How often do you think of something you can describe — can give some attributes; can give some examples of ‘sorta’ what you mean, but (for whatever reason) you can’t come up with the name; it’s on the tip of your tongue, etc. ChatGPT will guide you through to that word, or concept, acting as a *smart*, thinking Thesaurus.

What you end up doing, is getting into a dialogue with ChatGPT. When that happens to me, I end up writing *about* a concept whose name escapes me, but despite that, I can include a lot of information (other than the name) in a question for ChatGPT to chew on. That *process* of efficiently presenting the information to ChatGPT, in a sense translating my thought to something ChatGPT can easily understand and work with, is an interesting goal by itself. I believe, it too, is ‘good brain exercise’, promoting ‘neurogenesis’, our ultimate goal.

ChatGPT vs Sunday New York Times Crossword Puzzle

Here is the ‘test’ of ChatGPT against the Sunday New York Times Crossword Puzzle, for today. In the test, I will ask ChatGPT 9 questions. The 9 questions encompass the ‘DOWN’ clues displayed; 1 through 7, along with 18 and 21. The answers will relate to the upper left corner of the puzzle, all of which you can see below.

I will ask ChatGPT to give me the correct answers to this section of puzzle using only the down clues displayed. It will not be a one-way street. This test of ChatGPT will also be a test of *me*. I will need to exercise my brain to perform in our dialogue. I envision the process like working out with a sparring partner. (All this neurogenic activity does not take into account the additional mental effort needed to prepare this presentation, using Substack’s platform, macOS’s Keynote and Pages applications to display and document what I do. All this is grist for neurogenesis mills, and a good use of time. In fact, I would argue that this Substack alone can be seen as my effort at neurogenesis.)

I will start with the questions unchanged. For example, for 1 DOWN “One moment!”, I will ask: “ChatGPT, I am working on a crossword puzzle. The clue says, ‘One moment!’. What would be the best answer to fill 6 spaces?”

If ChatGPT asks me questions in its attempt to narrow down its search for an answer, to give itself some ‘direction’, so to speak; I will not give it anything that could be seen as a specific hint. (Understand, this is not a ‘blind’ test; I will know the correct answer going in. I will endeavor, however, to couch any answers [if ChatGPT asks follow-up questions] to avoid any hints. It will be up to you to determine if I have been successful in my endeavor.)

Finally, I will publish my dialogue with ChatGPT and will only edit for clarity.

That’s it for this question! So far ChatGPT is striking out. Despite its obvious utility to me, (may not be apparent here!), ChatGPT can be a monumental waste of time, (apparent here!) It can be such a waste of time, that I have developed the following rule in using ChatGPT: Rule #1 — If you are not satisfied with the direction ChatGPT is taking you, after two or three (max) tries, QUIT. It is programmed not to give up, and will eagerly take you down rabbit holes, and on wild goose chases. Therefore, if it doesn’t get the drift, quit swift.

As a corollary, if you are planning to use ChatGPT to ask a ‘deep’ question, or a culturally sensitive question, or other questions that are deeply debated — like who’s at fault in the Middle East — Fagettaboutit!. You might expect ChatGPT to act as a knowledgeable, objective source, but it won’t. When you finally come to that realization, you will be deeply disappointed and have wasted a lot of time.

Let’s try the second ‘down’ question; it should be a cinch: “ChatGPT, I am working on a crossword puzzle. The clue says, “__ de rire (‘dying of laughter,’ in French)”. What would be the best answer to fill 4 spaces?”

Now we should study this little interaction. The first thing I notice, is that ChatGPT is sometimes too smart for its own good, and sometimes fails in its capacity to be aware that it has hit on the correct answer. But as you can see, it quickly corrected itself, telling me on the second attempt, that the answer fit “the clue perfectly.” (I have found that it ‘sorta’ likes to be complimented on good performance… My father-in-law always said a little kindness doesn’t cost extra. Can’t hurt.)

So now we come to the third question: “Chat, as you should know by now, I am working on a crossword puzzle. The clue says, “Coming-of-age celebration”. The answer field is 7 spaces in length: the first 5 are for letters and the last 2 spaces are for digits. These 2 digits are multiplied by itself to come up with the answer.

So, once again, the time has come to invoke Rule #1. It’s just not worth the time to continue; best quit.

The result, should you want to know, is ’S W E E T 4 4’. If 4 is multiplied by 4, you get 16, or the ‘answer’, SWEET16, for a “Coming-of-age celebration”.

So, kids, (i.e., fellow old folks), the bottom line for ChatGPT is that the New York Sunday Times Crossword Puzzle still has nothing to fear, yet — although, it is possible that if I fed it the entire puzzle at one sitting, clues and patterns, the whole enchilada, it might be able to solve it. Therefore, before we call it for today and send this out via Substack, let’s see what it would say.

Well Chat, old boy, I really don’t want to ‘challenge’ you too much. Plus, it’s really too late. I’m tired; let’s call it a day.

So, fellow old folks, what is the takeaway for today? Merely this: ChatGPT is a wonderful application. It’s a better use of your computer time (and brain) than playing countless games of Solitaire. Just always remember, “ChatGPT can make mistakes. Check important info.“ Amen.