I was very impressed by the Text-Type grafic output Chat GPT came up with after i explained the game of brussel-sprouts to it. It is by far not sufficient to describe an drawing on the plane, but was a starrting point. It seems to have an understanding of geometry / topology in this context and we agreed and worked out a notation for the moves.

Sorry to disappoint you @gknobi

ChatGPT does not “understand” anything. ChatGPT just generates text, blah blah blah, based on probabilities, given a sequence of text, what is the probability of the next sequence of text. It does this based on a large language model, not a computing model.

HTH

Thx ruby_coder. I am totally aware that the underlying DNN has no rigorous “understanding” of the topics but was impressed by the fact that it came up with the suggestion to develop a notation for the game.

Things went messy soon after, as it has no real ability to even count the points and lines.

With the notation we agreed upon, each move is no more but a conversion of a phrase representing the current state of the game to the next phrase according to a ruleset.

That seemed close enough to what a linguistic network does anyways but did not work out.