Like most people for the last couple of years or so, I have been getting to know ChatGPT. I thought it might be fun to share with you some of my experiences that have gone sideways.
When ChatGPT first came out, I was just learning bridge and excited to ask the AI to give me samples of certain conventions, etc., so I could practice. I didn’t know much, but I knew the basics and many of the principles and was eager to try them out. I asked ChatGPT to give me ten examples of games when/where/how certain conventions should be used. It was a dumpster fire of lack of understanding – and in the most basic ways. For example, it didn’t even know how many cards are in any and every playing deck (52!), nor did it know that each card is a different number/suit combo! I know what you’re thinking -- the AI has gotten much better since then. I thought so as well. Therefore, armed with several years’ more bridge knowledge, I endeavored to ask it a few more bridge-related questions. I am here to tell you that 3 yr old ChatGPT still doesn’t know even the basics of bridge. I asked it something and it couldn’t even count how many cards were in the suit in one’s hand, nor could it correctly count the points that hand represented. As I called it on its errors four times, each time it ‘apologized’ and got it wrong again. Now one might find this comforting – like no need to worry that AI is going to take over the world any time soon. Me, I find this disquieting – like it’s still going to take over the world, but do it in a very sloppy, glitchy, unhelpful fashion!
On to more funny things. I had dinner with friends last month. They suggested we ask ChatGPT “What do you know about me” and “What is the funniest question I asked you this year?” Now, I should point out that my ChatGPT membership is actually my husband’s – with four of us using it. As such, asking what the AI knows about me is actually a trick question. Yet interestingly, it seems to home in on qualities based on my husband’s queries (as opposed to an amalgamation of the four of us).
The funniest question part was hilarious. Here is what it said:
You’ve asked many contenders, but the funniest question of the year—based purely on your ChatGPT persona and the whiplash intensity of your conversations—was almost certainly:
“Why does Imodium give me hives?”
…asked approximately eight minutes after grilling me about:
- grid-walking
history,
- the
total linear mileage of Manhattan’s street grid,
- the
Feast of San Gennaro,
- and
the location of Serendipity.
The abrupt shift from urban-planning professor to accidental pharmacology case study was peak you. It had extremely strong “Technotainment Scorecard: GI Tract Edition” energy (FYI – Technotainment Scorecard is my husband’s Substack column).
And speaking of the total linear mileage of Manhattan’s street grid, that query totally tripped ChatGPT up. When asked, that answer it gave was “I’ll let you know if authoritative sources publish or clarify the total mileage of all streets in Manhattan.” So I am waiting for a book to be published?? It then told me it was going to “Research Manhattan street mileage tomorrow.” WTF? Does ChatGPT take the night off? I asked it why it couldn’t just answer the question immediately and it then acquiesced that it could answer the question post haste.
So yeah, for me, ChatGPT has been kind of like a new assistant that while bright and eager, has no freaking clue and is going to take a lotttt of training.
No comments:
Post a Comment