Evolving Landscape of Disinformation - Lessons Learned So Far!
summary
Existing fake‑news detectors perform well on English but fail on low‑resource languages like Urdu, Hindi and Arabic; a prototype Arabic and English prototype that lets users copy‑paste an article returns a real/false verdict but coverage of more languages is ongoing. Deepfake detectors suffer large accuracy drops when videos are compressed — experiments show compression makes models "very less effective" — so work focuses on models robust to compression degrades detection and on releasing open‑source tools and international collaborations.
AI-written summary · about
related talks
27mTruth or Fiction The Art of Unveiling DeepFakes in Text, Audio, Image
35mDisinform your Surroundings: AI and disinformation campaigns
34mThe Psychology of Fake News (And What Tech Can Do About It)
30mAI-Generated Synthetic Media, aka Deepfakes and its impact
34mApplying Machine Learning Techniques to Detect and Understand Fake News
36mDisinform your Surroundings: AI and disinformation campaigns
26mDeepfakes, audiofakes, and the future of media- 34mApplying Machine Learning Techniques to Detect and Understand Fake News