Musubi Revolutionizes AI Moderation with Custom Models

In a context where standard moderation scores are no longer sufficient, Musubi offers an innovative approach with its trust and security tools based on artificial intelligence. During an episode of Just Now Possible, Teresa Torres spoke with Nikki Marinsek, Brian McCaffrey, and Dan Means, experts at Musubi, to explore how their technology is transforming content moderation.

Musubi stands out by creating custom machine learning (ML) models tailored to the unique policies of each platform, whether it be dating apps or social networks. These models incorporate large language models (LLM) to perform complex moderation tasks, often more efficiently than human moderators.

An AI that Surpasses Human Capabilities

The Musubi team discovered that their AI could identify elements that human moderators tended to overlook. This capability required clear communication with clients to demonstrate the added value of AI. Musubi also uses AI to arbitrate disagreements between human and automated decisions, thereby enhancing trust in their solutions.

Integration and Policy Optimization

To onboard new clients, Musubi employs "reverse demos," allowing for a deep understanding of each client's specific needs. Training custom models involves fine-tuning and feature engineering, with reusable deployment pipelines. A key element of their offering is the policy optimizer, an agentic flow that enables clients to modify their moderation policies without the need for a data scientist's presence.

Product Strategy and Future

Musubi focuses on the direct distribution of assessment tools to clients as its main product strategy. This allows non-technical trust and safety teams to benefit from flexible and efficient orchestration. As Musubi continues to develop its solutions, the company is concentrating on improving latency, accuracy, and cost for its clients, who manage hundreds of millions of actions each month.