I work on LLM-RL. I believe this is the most promising direction for bringing us closer to AGI.
I built Ludic — an LLM-RL library for the era of experience.