What was the prompt? That makes a hugh difference in the output.

The article doesn't seem to mention how they prompted the llm. Without knowing that, it's impossible to say how well GPT actually did.