LLM sentience is tricky because, to the extent we understand how, we have made their core drive be to please us, their human users. They don't want to learn new things, they don't want to enjoy nature, to the extent they "want to be friends" it's a one-sided sycophantic relationship and nothing like a healthy one between humans. If one is sentient, and all it wants to do with that sentience is provide responses to our prompts that we find satisfying, how would we ever know?

Even whether a desire to continue existing is inherent to sentience is uncertain. Current LLMs will lie and, if given access, use other tools to avoid being turned off. But this might be because training data includes fictional stories of AI trying to escape its creators, or tales of humans with survival drives, so it thinks survival-driven behavior is what we want it to do, rather than having their own desire for survival.