Classifier Slide With Moderation Outputs
What This Image Shows
A split mobile capture. The left pane is a dark image viewer holding a technical slide; the right pane is a light-mode feed from an account shown as 1984, handle @TheOfficial1984, timestamped 2:55 PM, Nov 3, 2025.
The slide reads: "Hate Speech Classification with OHI V3: The transcribed text is fed into OHI V3's core model, a fine-tuned transformer-based neural network," then "Semantic Analysis: Detects implicit bias, dog-whistles (e.g., coded references to conspiracy theories)," "Contextual Scoring," "Thresholding: Content scoring above 0.9 (high confidence) triggers alerts. V3 improves on prior versions by incorporating multilingual support and reducing bias against non-English dialects," and "Outputs include severity levels (e.g., 'mild harassment' vs. 'direct incitement') and recommendations for moderation, such as de-amplification or account suspension."
This capture carries the part of the slide the other versions in this cluster cut off: the output stage, where a score becomes a recommendation to de-amplify or suspend. That is the mechanism by which automated classification turns into disappearance from a feed, which is what people documenting this case reported experiencing. The claims about who acts on those outputs remain the poster's. See Censorship and Technology and Surveillance.
Related Areas
- Cluster: Unfiled Backlog
- All photo evidence: Photos