Comment on Dispelling misconceptions about RLHFparentComments−JoshPurtell1yRL is not about delayed reward. Multi-armed bandit problems have no credit assignment component, but are often the first RL problem taught.In its most general, RL is about learning a policy (state -> action mapping). Which often requires inferring value, etc.But copying a strong reference policy ... is still learning a policy. Whether by SFT or not
Comments
RL is not about delayed reward. Multi-armed bandit problems have no credit assignment component, but are often the first RL problem taught.
In its most general, RL is about learning a policy (state -> action mapping). Which often requires inferring value, etc.
But copying a strong reference policy ... is still learning a policy. Whether by SFT or not