Nope there's no attack here. The training data is the Internet. It has mistakes. There's no available technology to remove all such mistakes. Whether LLMs hal

Nope there's no attack here. The training data is the Internet. It has mistakes. There's no available technology to remove all such mistakes. Whether LLMs hallucinate only because of mistakes in the training data or whether they would hallucinate even if we removed all mistakes is an extremely interesting and important question.