Although this paper is a fairly small-scale proof of concept, I am extremely excited to see where this technique goes. Between ChatGPT and this, I get the feeling that we are just barely touching the surface of what RLHF is capable of in generative models. This paper[0] about mixing in RLHF with supervised learning from the
start of LLM training also has me very excited. I feel like we haven't seen a single method make this big of waves in the ML world for a while now.
[0]https://arxiv.org/abs/2302.08582
(Currently empty) discussion thread for [0]: https://news.ycombinator.com/item?id=34869295