They will reliably emit sentences that, in humans, correspond to signals of affection for cat. What does it correspond to in a language model? How does text prediction generalise to actions in meatspace?
A lot of technology has APIs that take in text and then create actions in meatspace. If your LLM gives out an instruction that then passed to an API that executes actions in meatspace, the connection is very direct.
They will reliably emit sentences that, in humans, correspond to signals of affection for cat. What does it correspond to in a language model? How does text prediction generalise to actions in meatspace?
A lot of technology has APIs that take in text and then create actions in meatspace. If your LLM gives out an instruction that then passed to an API that executes actions in meatspace, the connection is very direct.
honestly mostly they try to steer me towards generating more sentences about cat