I would heavily disagree on LLMs being obviously sentient. I don’t see how they could be, as I see it it would require an understanding of the words they put out in a framework that isn’t language - but language is the only framework they can possibly have. All words LLMs know are defined only through their relation to other words.
It’s also a matter of defining what self-awareness even is, and I don’t have the hubris to say that LLMs for sure are not sentient, it just seems very unlikely from my perspective. I would also argue that monkeys are most likely sapient, though the whole concept has always reeked of “oh shit sentience doesn’t make us special, quick think of something else”. I’d probably put LLMs somewhere on the level of individualistic insects at best.
I still agree with your general sentiment that people are far too dismissive of the topic in general given our history in doing this with animals. Esp the concept of pain is an interesting one, because essentially pain is just negative reinforcement, which of course is part of reinforcement learning. But pain also had to naturally evolve and is very flawed as a result, while the negative reinforcement is just a number without side effects, so it likely wouldn’t ever feel like pain the same way it feels for an animal. But the difference is often philosophical, and we define things the way they are convenient.
I would heavily disagree on LLMs being obviously sentient. I don’t see how they could be, as I see it it would require an understanding of the words they put out in a framework that isn’t language - but language is the only framework they can possibly have. All words LLMs know are defined only through their relation to other words.
It’s also a matter of defining what self-awareness even is, and I don’t have the hubris to say that LLMs for sure are not sentient, it just seems very unlikely from my perspective. I would also argue that monkeys are most likely sapient, though the whole concept has always reeked of “oh shit sentience doesn’t make us special, quick think of something else”. I’d probably put LLMs somewhere on the level of individualistic insects at best.
I still agree with your general sentiment that people are far too dismissive of the topic in general given our history in doing this with animals. Esp the concept of pain is an interesting one, because essentially pain is just negative reinforcement, which of course is part of reinforcement learning. But pain also had to naturally evolve and is very flawed as a result, while the negative reinforcement is just a number without side effects, so it likely wouldn’t ever feel like pain the same way it feels for an animal. But the difference is often philosophical, and we define things the way they are convenient.