- Summary: Last year, Google open-sourced SynthID AI watermarking system. Now it's rolling out a web-based portal for easy testing of media watermarked with SynthID. After uploading media, users get results highlighting likely watermarked parts, and the watermarked content remains detectable even after transformations. The detector is available to beta testers starting today; journalists, etc. can apply for a waitlist. Currently, only images and audio can be uploaded; video and text detection will be added soon. Detectors for identifying AI-generated content are unreliable in real-world scenarios, so watermarking is important. Over 10 billion pieces of content have been marked with SynthID, including those generated by Google and Nvidia models. However, many AI-generated contents are not watermarked, and there has been little progress in getting major stakeholders to adopt a single easy-to-detect standard. OpenAI is reportedly testing watermarking images but declined to release its internal text watermarking tool last year.
- Main points: Google open-sourced SynthID; launched web portal; beta testing available; only images/audio uploadable initially; many unmarked AI content; lack of standard; OpenAI's watermarking efforts.
- Key information: Link to Google's open-sourcing; Beta tester application form; Details about different Google and Nvidia models.
- Important details: Reliability issues of AI detectors; number of marked content (over 10 billion); OpenAI's stance on text watermarking tool.
**粗体** _斜体_ [链接](http://example.com) `代码` - 列表 > 引用
。你还可以使用@
来通知其他用户。