The ActiveFence Engineering Blog
Why is it difficult to receive feedback (even for AI)
Labeled datasets are invaluable assets. There is a direct link between dataset quality and model performance. Efficient, robust, and sensitive harmful content detection leverages numerous supervised models, so it is critical for us to have funnels that generate useful labels and make the most out of them.
How Fingerprints Fight Online Evil
Malicious content, including hate speech, misinformation, and CSAM are readily shared on social media platforms. However, using technology, we can make the work of resharing that content more difficult.
Racism, Hate, and Weddings- 3 Ways to Avoid Biased Data
Working with biased data presents many problems. Here, we share how to diagnose, handle, and reduce such data
Constructing and querying a data model for online harm- Part 1
Online content, media creators, and user interactions must be captured to build algorithms that detect harmful activity. Here, we discuss the data model that captures the complexities of this online ecosystem and how we use it to detect malicious activity at scale
Constructing and querying a data model for online harm – Part 2
Online content, media creators, and user interactions must be captured to build algorithms that detect harmful activity. Here, we discuss the data model that captures the complexities of this online ecosystem and how we use it to detect malicious activity at scale
The Metaverse: Helping Platforms Keep Us Safe in New Digital Territory
The metaverse presents a unique opportunity to foster community and drive innovation online. However, it also opens the door for a greater risk of harm, which is more difficult to detect and proactively combat. However, with contextual AI, safety in the metaverse stands a chance
ActiveFence R&D on Practical AI Podcast
Listen to how we build AI that solves the complex challenges of content moderation on this podcast with Matar Haller, Daniel Whitenack & Chris Benson