↻
Jeremy Hsu reposted
@arstechnica.com
While labeling AI content is a good policy for tech firms, it probably won’t save the rest of us who have to deal with its outputs.
AI Weekly's analysis
→
- Google says SynthID has now labeled 100 billion images and videos, plus 60,000 years of audio.
- Ordinary edits and noise barely dent the watermark, but running an image back through another AI model breaks the signal.
- OpenAI said in May 2026 it will embed SynthID in all ChatGPT-generated images, joining Nvidia and other adopters.
Read full analysis →
View on Bluesky →