|
|
|
|
|
by wadadadad
215 days ago
|
|
I don't think riddles are necessarily 'solvable' in that there's only one right answer; the very fact that they're open to interpretation, but when you get the 'right' answer it (hopefully) makes sense. So if an AI/LLM can answer such a nebulous thing correctly- that's more of the angle I was going at. Regarding the wizards example, I'm a bit confused; I was thinking that the best way to judge answers for problem solving/creativity was for correctness. I'll think more on whether the 'thought process' counts in and of itself. The answer to my riddle is 'ball'. |
|
If you ask me the goat, wolf, cabbage problem I'd be able to recite (as an xkcd fan https://xkcd.com/1134/ and https://xkcd.com/2348/ and the exploration of what else it could do). However, if someone hasn't seen the problem before it could be a useful tool at seeing how they approach solving it.
The question of how does it tackle a new problem is one of creativity and exploration of thought in a new (untrained) domain.
A possible claim of "well, it's been trained on the meta-problem of how to solve problems that weren't in its training set" would get a side eye.
For the "ball" being the answer... consider the second response to https://chatgpt.com/share/6920b9e2-764c-8011-a14a-012e97573f... (make sure you click on the "Thought for 1m 5s" to get the internal process)