AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
How to Train AI to Think Like a Human
In reinforcement learning with human feedback, you train a separate AI to think like some humans. Then that model trains the other model by being like my panel of humans would give this a thumbs up or thumbs down so you can do a really fast right? So then the question is who are those humans? Where did you find us? What are their values? Like what religion are they? How educated are they? Are they jerks? Are they nice people? Because you're going to be asking them about a lot of stuff. Um, if you were trying to fix that the model is racist, um, are these people white or are they black? You know, or are they like Nigerian