#Sentient means the system itself learns to assign a its own values (i.e. "good" or "bad") to inputs. LLM models may be exploited to assign those words to inputs but doing so doesn't exhibit the system has those values.

A thermostat or a LLM isn't sentient, but is classifying inputs as it was built to. All the macrofauna seem to be sapient. If smaller animals seem less than sapient it may be due to the frantic nature of life at that scale, but if the creature can be taught to respond to unfamiliar stimuli to find food or avoid pain in a maze, then perhaps that's a demonstration of sentience.

Sapience requires, I think, being able to run a maze composed of society. You need problem solving, a theory of mind to understand the motivation of others with similar brains, and the setting and intelligent pursuit of strategic goals. I would set that bar so high, not all U.S. Presidents would qualify.