Why AI cheats: how a model learns to game its own test
RLHF: how a chatbot is trained on what people like
AI labs score millions of chatbot answers to train their chatbots. A clear, kind or true answer has no test a program can run. So who gives the score?
-
1
Miva's team now trains a chatbot, and Rono's bread question makes a good example. Two answers come back. Which one is better? It takes a person to judge.
-
2
So people compare. A rater reads two answers to the same question and picks the better one. Labs collect many thousands of these picks.
The rest of this episode is for subscribers
Subscribe to read every premium episode. You can cancel when you want.
Paid yearly: $99 instead of $119.88