That is not a risk. There are already many examples of questions and answers to engineering/physics/mathematical problems in the existing training set. The issu

That is not a risk. There are already many examples of questions and answers to engineering/physics/mathematical problems in the existing training set. The issue is that GPT is fundamentally not suited for mathematical reasoning, no matter how much training data you feed it.