Language is just a bunch of sounds with no objective “true”/”literal” meaning. Therefore: the term “literal” in “literal genie” should imply the genie will interpret the words the way the average speaker of the language would if instructed to interpret those words literally. So probably A.
I’ve noticed that some early thought experiments about AI seem to have been predicated on the idea that superintelligent AI won’t know what you mean when you say stuff, so you’ll have to be hyper-specific. But if the literal genie in your hypothetical is like modern LLMs, it will know that you mean A because it knows what people mean when they say stuff, generally. Whether or not it actually cares what you mean is a different issue I suppose.
Language is just a bunch of sounds with no objective “true”/”literal” meaning. Therefore: the term “literal” in “literal genie” should imply the genie will interpret the words the way the average speaker of the language would if instructed to interpret those words literally. So probably A.
I’ve noticed that some early thought experiments about AI seem to have been predicated on the idea that superintelligent AI won’t know what you mean when you say stuff, so you’ll have to be hyper-specific. But if the literal genie in your hypothetical is like modern LLMs, it will know that you mean A because it knows what people mean when they say stuff, generally. Whether or not it actually cares what you mean is a different issue I suppose.