Given that LLMs are trained in large corpus of text (not in logic, not in mathematical first principles), isn't too much to expect that it can understand the pr

Given that LLMs are trained in large corpus of text (not in logic, not in mathematical first principles), isn't too much to expect that it can understand the problem and then solve it mathematically?