Have you ever tried the New York Times Spelling Bee puzzle? It's one of my go-to guilty pleasures. No kidding, I can't shut my eyes at night unless I've achieved at least the 'Great' rank. For those unfamiliar, the Spelling Bee puzzle involves finding as many words as possible using a given set of letters.
I thought it would be fun to see how ChatGPT would perform at this game. Could it get genius level results?
I experimented with two approaches to see how ChatGPT fared with the game.
Using the image upload feature (exclusive to ChatGPT 4 paid subscribers).
Text-based data analysis.
Image Uploads:
We'll start with my first attempt. I leveraged the Default option where I can upload images and ask questions.
I uploaded a total of 3 images.
The puzzle that day
The words I had found so far
Rules for how to play
The results left something to be desired and that's being kind. I didn't want to give up. I figured it must be my prompting and I wasn't clearly stating the requirements. I continued to ask ChatGPT to try again and offered feedback.
I even asked, 'What do you need from me in the prompt?' and I took note and tried again with a new puzzle a few days later. This prompt clearly states exactly what ChatGPT told me it needed to be successful.
It still did not perform well.
Notice a 3 letter word to start when the minimum length is 4 characters.
ChatGPT suggested a word starting with an 'L', but there wasn't even an 'L' in the puzzle!
Given these challenges, I wondered if maybe I should be using the advanced data analysis feature in ChatGPT 4 instead for better results.
So because I'm not one to give up easily, I started a brand new chat approaching this from a data analysis perspective.
Things got off to a pretty good start... not sure how I felt about the limiting to an 8 letter length, but it wasn't a deal breaker at this juncture.
Then just as I was planning to show how this additional effort paid off, another setback.
This time it was the darn ChatGPT memory limitations tripping up the results.
And more of the same message.
I apologize for the inconvenience.
But in true ChatGPT fashion, it didn't just say too bad, can't help you. ChatGPT came back with alternative approaches. One option was for me to upload a comprehensive word list I already had.
That was not going to happen tonight, probably not ever. Instead, I'm reverting to good ol' human brainpower to tackle my favorite game. But hey, if you've cracked the ChatGPT code for the Spelling Bee puzzle, do spill the beans! I guess I still haven't tried this with other models like BingChat or Google's Bard."
Bottom line, this is a great reminder that ChatGPT won't be great at everything we ask it to do, at least not right away. And it's up to us to choose what makes sense based on its capabilities and limitations.
What have you tried with ChatGPT that hasn't worked so well. Would love to hear about it.
Please share your stories in the comments and don't forget if you enjoy the blog, click the heart and share with your friends. Let's learn and grow together.
Comments