[lblume]

I tried giving this to ChatGPT. (Just by uploading the image to the base OpenAI interface.) I expected to either (a) have the model already know the question and give the right answer, (b) hallucinate an answer or (c) refuse to engage with the problem at all.

Instead, this happened:

https://chatgpt.com/share/682cce62-c53c-8003-be2c-2929395868...

Basically, the model confidently outputs a guess, then calculates it, determines it to be incorrect, and repeatedly tries again, even repeating the same guesses over and over. It does not recognize any symmetry and acts like a completely unstructured agent. In the end, the model vehemently asserts there to be no solutions to this puzzle. I really did not expect this and will update my beliefs accordingly if the models behave as badly with future puzzles.

[arikrak]

Here's gemini https://g.co/gemini/share/ab287b25648f

I also asked Chat GPT o3 and it thought for 11.5 minutes! https://chatgpt.com/share/682d0993-db4c-8004-a66c-3908ef7203...

[LeoPanthera]

I'm impressed. There certainly are no "reasonable" solutions, where I am arbitrarily defining reasonable as numbers that humans can work or understand in their heads.

Isn't there a version of ChatGPT that connects to Wolfram Alpha? Did you try that one?