[RevEng]

It's important to remember that modern LLMs are trained on bloody everything. They know every common logic problem, at least when stated the way they would have seen it.

If you want to test an LLM, always make up a new problem. It can be the same idea as an existing problem, but change all names and numbers.

I tested if GPT 3.5 could recognize chaos theory. If I stated it as the typical "butterfly flaps its wings" it instantly recognized it as the chaos theory example. If I totally changed the problem statement, it correctly identified that weather isn't correlated with a single action by a single person, but it didn't associate it with chaos theory.