16/01/2023
Seemingly, like much of the internet, I have recently been checking out OpenAI's ChatGPT. And I'm sure there are some superbly researched and well considered blogs out there examining what ChatGPT might mean in terms of how we use the internet, jobs & the economy and whether or not it's the embryonic incarnation of Skynet.
Closer to home, I was curious as to ChatGPT's riddle solving powers. I had feared it might make light work of wyfio.com. At the same time, I had hoped it would struggle in a similar fashion to its meme-like 'as a large language model' excuse that it rolls out any time you really start to press it.
I asked ChatGPT if it could help me solve an online riddle, gave it a little bit of context about the game and then started to describe some randomly selected puzzles to it. At first, ChatGPT provided some insightful analysis at the first take that would have enabled the vast majority of players to solve the puzzle. And whilst a bit of Googling would probably get you to a similar place quite quickly, it didn't do much to allay my fears.
Abandoning my 'random selection' tactic, I thought I would up the ante and present some more difficult puzzles. Cutting to the chase, I didn't see anything that would cook our goose just yet. It occasionally hit upon something of value, but more than occasionally it would flounder about presenting a loose and misleading array of possibilities.
Look, I'm sure you can cajole it into being more helpful than I did – but as things stand I've seen enough to know that it doesn't yet have anything like the scary problem-solving abilities that I thought it might.
Is it still schadenfreude if the object isn't human :) ?
P.S. I asked ChatGPT for feedback on this post and some suggested pseudonyms for itself. It only offered up the rather terrifying 'Necronomicon'. Yikes.