AI Models Are Watermarking Text-Will You Notice?
Anthropic revealed on 11 August that future versions of its Claude models will incorporate a watermark to mark generated text as artificial intelligence output. This move follows Google, which applies its own text watermarking system to content produced by Gemini models. Notably, Anthropic based its watermarking technology on the system developed by Google.
Key Takeaways
- On 11 August, Anthropic stated that every upcoming Claude model will automatically add a watermark to its written outputs.
This feature is designed to clearly mark content as machine-generated.
- Anthropic leveraged Google's existing watermarking framework to construct its own implementation for future Claude deployments.
On 11 August, Anthropic announced plans to watermark all text produced by future Claude models.
- Anthropic built its new text watermarking framework using Google's underlying technology.
- Google similarly employs a text watermark for responses created by its Gemini models.
- Google already utilizes a text watermarking technique on output generated by its Gemini models.
Stats & Key Facts
- #Anthropic revealed on 11 August that future versions of its Claude models will incorporate a watermark to mark generated text as artificial intelligence output.
- #On 11 August, Anthropic stated that every upcoming Claude model will automatically add a watermark to its written outputs.
- #On 11 August, Anthropic announced plans to watermark all text produced by future Claude models.
On 11 August, Anthropic stated that every upcoming Claude model will automatically add a watermark to its written outputs. This feature is designed to clearly mark content as machine-generated. Google similarly employs a text watermark for responses created by its Gemini models.
Anthropic leveraged Google's existing watermarking framework to construct its own implementation for future Claude deployments. On 11 August, Anthropic announced plans to watermark all text produced by future Claude models. Google already utilizes a text watermarking technique on output generated by its Gemini models.
For more details please read the original article at IEEE Spectrum AI.
Continue Learning
Comments
Comments appear only after moderation. Your email identifies your submission to the moderator and is never displayed here.
No approved comments yet.