papersTODAY 04:00 UTC
arXiv Paper Releases Dataset and Detector Benchmark for ChatGPT Images 2.5
A new arXiv study examines how difficult it is to attribute an AI-generated image to a specific model version when a tool keeps its public name but swaps its underlying generator. The authors assembled a frozen set of 3,478 images collected around the ChatGPT Images 2.5 launch and evaluated detectors against it.