I always use jokes to test language models. how many a.i. agents does it take to change a lightbulb? that depends on how many times you've asked that question

I always use jokes to test language models. how many a.i. agents does it take to change a lightbulb? that depends on how many times you've asked that question. Conceptually speaking, is it a joke? I can't tell. But as an assistant joke generator it could be invaluable. And certainly makes one wish they had access to Google PaLM with 500B parameters and web scale training data ;)