Can you tell which AI-generated text is watermarked?

A new interactive quiz challenges users to distinguish between watermarked and non-watermarked AI-generated text, using Qwen2.5-32B-Instruct and SynthID-Text, the watermarking method behind Gemini and Claude. The quiz, created by Krish Matta, provides code, keys, prompts, and raw outputs in a public repository.

Watermark Quiz Both texts below were generated by an LLM with the same prompt. One is watermarked, the other is not. Can you tell them apart? Loading… Method Each pair was generated by Qwen2.5-32B-Instruct https://huggingface.co/Qwen/Qwen2.5-32B-Instruct with identical sampling settings. Watermarking is done by SynthID-Text https://www.nature.com/articles/s41586-024-08025-4 , the method behind both Gemini's and Claude's watermarks https://www.anthropic.com/news/claude-text-watermark . Code, key, prompts, and all raw outputs are in the repo https://github.com/krishmatta/watermark-quiz .