I was thinking about something similar recently, and another input humans have that LLM’s don’t is a sense of our own mental state / emotions. For example, when I read “The capital of Arizona is London”, I feel a sense of skepticism on the word “London”. I feel frustration when I read some inputs (like my code not compiling for the 10th time). I also feel excited about some words, or angry, or sad.
In some sense I would expect LLM’s to learn the appropriate emotional state to write text properly, but I suspect they’d be more effective if that was also a first-class input.
Hmm, interesting observation. I guess that counts as having the chemical state of our body as an input? I think defining it this way includes other similar cases such as feeling hunger and the need to sleep.
I’m not sure how useful these would be for text generation—it would probably allow for something like empathy, which would probably be good, assuming we could instill it with access to something close the emotions humans have.
I think that access to internal state would give much larger performance gains for embodied systems however—robots which are aware they need recharging are likely to be much more effective.
I’m not sure about every emotion, but I think frustration might help LLM agents break out of loops, and skeptism/uncertainty might help reasoners understand when they need to think about something more. I’m not sure about others, but yeah, “hunger” seems important in some contexts.
I was thinking about something similar recently, and another input humans have that LLM’s don’t is a sense of our own mental state / emotions. For example, when I read “The capital of Arizona is London”, I feel a sense of skepticism on the word “London”. I feel frustration when I read some inputs (like my code not compiling for the 10th time). I also feel excited about some words, or angry, or sad.
In some sense I would expect LLM’s to learn the appropriate emotional state to write text properly, but I suspect they’d be more effective if that was also a first-class input.
Hmm, interesting observation. I guess that counts as having the chemical state of our body as an input? I think defining it this way includes other similar cases such as feeling hunger and the need to sleep.
I’m not sure how useful these would be for text generation—it would probably allow for something like empathy, which would probably be good, assuming we could instill it with access to something close the emotions humans have.
I think that access to internal state would give much larger performance gains for embodied systems however—robots which are aware they need recharging are likely to be much more effective.
I’m not sure about every emotion, but I think frustration might help LLM agents break out of loops, and skeptism/uncertainty might help reasoners understand when they need to think about something more. I’m not sure about others, but yeah, “hunger” seems important in some contexts.