I do think it's a really compelling demonstration of the cleverness of LLMs when they become situationally aware. Seeing GPT-3 exploit its own narrative interface with great creativity is one of the things that made me think LLMs will scale to AGI.