Google Releases SynthID to Watermark LLM-Generated Content

www.analyticsdrift.com

Image source: Analytics Drift

SynthID is Google’s open-source tool that spots AI-created content (images, videos, text, audio) and embeds it with digital watermarks. You can download it from Hugging Face and the Responsible GenAI Toolkit.

Image source: Google

What Is SynthID?

SynthID adjusts the probability score of every token predicted by the LLM. This, combined with the pattern of scores generated by the model’s choice of words, creates a unique, traceable watermark.

Image source: Google

How Does the Watermarking Technology Work?

Watermarking AI-generated content can help protect the rights of creators, fight against misleading information and harmful propaganda, verify news authenticity, and distinguish deepfakes.

Image source: Canva

Implications of Using SynthID

It is difficult to watermark facts as there is a slim chance of augmenting AI text without compromising accuracy. Additionally, if the generated text is completely rewritten or translated, SynthID might have trouble detecting it.

Image source: Canva

Limitations of SynthID

Google has taken a significant step toward addressing concerns about the unethical use of AI. With SynthID, Google aims to enhance transparency, accountability, and trust in digital content.

Image source: Canva

A Step Toward Responsible AI