Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• Properly configured In-Context Learning (ICL) achieves performance comparable to fine-tuned models using 22x fewer labeled examples.result 90%
• ICL generalizes to new illicit categories without category-specific demonstrations, with a performance drop of less than 6% for most evaluated categories.result 90%
• A novel two-stage pipeline distills 2,900 free-form labels into coherent taxonomies, autonomously surfacing eight previously undocumented illicit categories.method 90%
• Deployed on 200,000 real-world samples from search engines and Twitter without adaptation, ICL achieves 92.6% accuracy.result 90%
• 61.8% of ICL's uniquely flagged samples correspond to borderline or obfuscated content missed by existing detectors.result 90%
• Explicit labels in demonstrations are critical for ICL performance; removing them leads to performance worse than zero-shot and a sharp increase in False Positive Rate.technical 85%
• ICL offers a path toward proactively adaptive moderation systems by shifting to inference-time reasoning.technical 80%
• Existing moderation systems struggle to generalize across domains or adapt to emerging threats due to being tethered to platform-specific supervision and static taxonomies.limitation 85%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card