|
|
|
|
|
|
by PaulHoule
1181 days ago
|
|
|
Human Feedback Reinforcement Learning is a fancy way to say “tell people what they want to hear” and there are few things more dangerous than that. Imagine a lesswrong fanatic that makes his own personal Eliezer Yudkowsky and gets driven (completely in private) to murder an A.I. researcher. |
|
|