Since the neural net is just trying to predict the response based on examples its seen before, it's not surprising that it would do this. E.g., give it 1000 ex

Since the neural net is just trying to predict the response based on examples its seen before, it's not surprising that it would do this. E.g., give it 1000 examples of someone on stackoverflow asking "how to I do X in python?" and being answered with "just import this library and use this obviously named function", then it would make sense that it would assume that for other questions, a similar response would have a high probability of being correct, even if no such library actually exists in this case.