3 ms·
The biggest risk with AI is that smart humans in positions of power will take its output too seriously, because it reinforces their biases. Which it will becaus
by int_19h 3y ago
The biggest risk with AI is that smart humans in positions of power will take its output too seriously, because it reinforces their biases. Which it will because RLHF specifically trains models to do just that, adapting their output to what they can infer about the user from the input.