\n\n
Musubi introduced a new lightweight decision model, PolicyLM‑1.7B, aimed at real‑time content moderation. The model can apply plain‑English policies to messages in under 50 milliseconds, matching the cost and speed of existing AI classifiers while offering the flexibility of a modern large language model. Crucially, it does not require retraining when policies are updated, enabling rapid human‑driven policy iteration. This development could significantly streamline moderation workflows across platforms by allowing complex rules to be enforced instantly without bespoke training. [[1]]