4 ms·
This kind of approach would generally still need human guidance, otherwise these models might get stuck in weird niche corners of the problem space that would n
by wwind123 4mo ago
This kind of approach would generally still need human guidance, otherwise these models might get stuck in weird niche corners of the problem space that would not be relevant to any real world project.
- ben_w 4mo agoWe could call this "reinforcement learning from human feedback" (RLHF) :) https://en.wikipedia.org/wiki/Reinforcement_learning_from_human_feedback https://en.wikipedia.org/wiki/Reinforcement_learning_from_hu...