If you have an external program, then by defining it's not self-awareness ;). Also, it's not about correctness per se, but about the model's ability to assess its own knowledge (making a mistake because the model was exposed to mistakes in the training data is fine, hallucinating isn't).