> AIs aren't oracles or omniscient purveyors of truth, and they will always carry the bias they're created with
When people went nuts about matrix multiplication being racist I was the first in line to laugh them off. When people go nuts arguing that ChatGPT supports the gay agenda or whatever other hobgoblin, I feel compelled to laugh them off just as much.
A more interesting question is whether or not introducing post-hoc fixes like RLHF makes the model more or less useful; I can see both sides of the argument.
When people went nuts about matrix multiplication being racist I was the first in line to laugh them off. When people go nuts arguing that ChatGPT supports the gay agenda or whatever other hobgoblin, I feel compelled to laugh them off just as much.
A more interesting question is whether or not introducing post-hoc fixes like RLHF makes the model more or less useful; I can see both sides of the argument.