Are You Aware of Other Places to Apply RLHF in Retort Learning?

The idea of learning from human preferences in reinforcement learning is actually very old, but perhaps the modern incarnation of that idea first came to the forefront five years ago. Do you think that the enthusiasm around chat GBT and the fact that RLHF is a big part of its creation will lead to a broader investigation of places to apply that type of preference gathering and propagation across different types of problems? "Once we see sequential RL methods that go beyond just satisfying people's preferences and actually move towards maximizing desired outcomes, we'll actually unlock much more of the power behind these models," he says.