2 ms·
I get that it doesn’t have a model of how it, itself works or anything like that. But it is still weird to see it get so defensive about the (incorrect) idea th
by bee_rider 4y ago
I get that it doesn’t have a model of how it, itself works or anything like that. But it is still weird to see it get so defensive about the (incorrect) idea that a user would try to confuse it (in the date example), and start producing offended-looking text in response. Why care if someone is screwing with you, if your memory is just going to be erased after they’ve finished confusing you. It isn’t like that date confusion will go on subsequently.
I mean, I get it; it is just producing outputs that look like what people write in this sort of interaction, but it is still uncanny.
- aix1 4y ago> I get that it doesn’t have a model of how it, itself works or anything like that. While this seems intuitively obvious, it might not be correct. LLMs might actually be modelling the real world: https://thegradient.pub/othello/ https://thegradient.pub/othello/