To what you’ve said about humans and existing LLMs not being horribly sociopathic all the time because their motivations aren’t constructed from RL + search, I would add the following: One of the more obvious examples of how humans get put into an RL-like environment is the way CEOs of companies are constantly being evaluated on what’s approximately a pure profit-maximizing objective, and as a group they don’t exactly have a reputation for not being sociopathic.
I’m not sure whether something similar applies in other instances where people are consistently rewarded for maximizing something. Salespeople, perhaps: at least a bit of a reputation for sociopathy there. Children at school? Empirically most schoolchildren most at the time care a lot about things other than the stuff their teachers are evaluating them on, so I don’t know how good an example that is. Athletes? Not particularly thought of as sociopathic, I think.
This seems like the sort of thing some psychologist has probably done a lot of interesting research on. Which probably failed to replicate.
To what you’ve said about humans and existing LLMs not being horribly sociopathic all the time because their motivations aren’t constructed from RL + search
Steven Byrnes said that existing LLMs mostly do not choose their actions using RL + search, but he said that humans do choose their actions using RL + search, and the reason that humans usually aren’t sociopathic is because of their “exotic “non-behaviorist” reward function”.
To what you’ve said about humans and existing LLMs not being horribly sociopathic all the time because their motivations aren’t constructed from RL + search, I would add the following: One of the more obvious examples of how humans get put into an RL-like environment is the way CEOs of companies are constantly being evaluated on what’s approximately a pure profit-maximizing objective, and as a group they don’t exactly have a reputation for not being sociopathic.
I’m not sure whether something similar applies in other instances where people are consistently rewarded for maximizing something. Salespeople, perhaps: at least a bit of a reputation for sociopathy there. Children at school? Empirically most schoolchildren most at the time care a lot about things other than the stuff their teachers are evaluating them on, so I don’t know how good an example that is. Athletes? Not particularly thought of as sociopathic, I think.
This seems like the sort of thing some psychologist has probably done a lot of interesting research on. Which probably failed to replicate.
Steven Byrnes said that existing LLMs mostly do not choose their actions using RL + search, but he said that humans do choose their actions using RL + search, and the reason that humans usually aren’t sociopathic is because of their “exotic “non-behaviorist” reward function”.