Yeah, that's raw llama2 for you. Fine tuning, LoRa, that's how you get the nice responses.
They probably weren't specific enough in mentioning what the model it was built on was, referring to Llama-2-chat as being a Llama-2 model (which is kind of correct).