Highlights from our LLM Watermarking Workshop!

We had the pleasure of hosting an enlightening session on both passive and active watermarking techniques, led by Pierre Fernandez, Research Scientist at Meta.

Our journey began with exploring passive detection tools, delving into how they function, their benefits, and where they fall short in detecting AI-generated content. Then, we rolled up our sleeves and implemented the greenlist/redlist watermarking technique—an industry-standard method used by LLM providers to embed detectable markers in AI outputs.

The workshop provided us with a hands-on experience, offering a deep dive into the practicalities and challenges of watermarking AI-generated text. We gained crucial insights into the robustness, effectiveness, and potential vulnerabilities of these techniques in real-world applications.