Some research I've been very excited to share on how we are using RL at
@appliedcompute as a hammer to accelerate everything we do internally, not just externally.
We used RL to train models that create curated context from long documents for downstream use by agents. The models sometimes learn to invent their own abbreviations and shorthand. Optimizing with RL for downstream use produces very different artifacts from ordinary summaries: shorter, denser, creatively concise. We call these neural cheat-sheets.
Show more