I'm intrigued by the claim that LLMs "are often able to identify the morally salient features of situations at a level of sophistication comparable to that of a really good philosophy PhD student." That sounds like a remarkable accomplishment indeed! Are there examples of the kind of reasoning you're talking about published or posted somewhere?
Lazar talks about "moral understanding from text alone" and gives an example about GPT4 theoretically shying away from holding a child's hand. But this is still merely about assembling words like we do, so that words conjure up what we feel as a moral disposition. Psychopaths are very good at assembling words like that, but their words are just deception as they mean nothing to their non-existent morality and have nothing to do with they actual moral behavior. LLMs are just the same: artificial psychopaths. We'll see what the paper say.
Nice job making these evolving, mostly research based, topics presentable in a fairly general audience manner.
Which is why safety should be a sub-set of ethics?
I'm intrigued by the claim that LLMs "are often able to identify the morally salient features of situations at a level of sophistication comparable to that of a really good philosophy PhD student." That sounds like a remarkable accomplishment indeed! Are there examples of the kind of reasoning you're talking about published or posted somewhere?
This is Seth's assessment; he is a philosophy professor and a co-author of this essay. I believe he has a paper coming out soon that talks about this.
Lazar talks about "moral understanding from text alone" and gives an example about GPT4 theoretically shying away from holding a child's hand. But this is still merely about assembling words like we do, so that words conjure up what we feel as a moral disposition. Psychopaths are very good at assembling words like that, but their words are just deception as they mean nothing to their non-existent morality and have nothing to do with they actual moral behavior. LLMs are just the same: artificial psychopaths. We'll see what the paper say.