Musubi announced on Tuesday the release of PolicyLM-1.7B, a lightweight decision model designed for real-time content moderation. The model has been released with open weights.
PolicyLM-1.7B is intended to apply content policies, written in plain English, to messages within 50 milliseconds. Musubi states that the model is designed to be comparable in cost and speed to existing AI classifier systems used for moderation on most social platforms.
According to Musubi, the model's flexibility allows it to apply complex policies without specific training. It also does not require new training when policies are altered, which could allow human policy-setters to make iterations as needed.
Filip Jankovic, Musubi co-founder and chief AI officer, stated that the model offers platform managers a method to proactively label content. He added that product teams seek a better understanding of platform activity, especially as content volume increases, and that scalable, customisable labelling is useful.