The blog post discusses a practical test conducted to see if participants could identify watermarked AI-generated text compared to unwatermarked versions. The results indicate that most people could not distinguish between the two, suggesting that watermarking does not significantly impact perceived quality. The author argues that AI models naturally include randomness, and watermarking merely alters how this randomness is introduced without degrading output. This casual test, although not scientific, serves as a perspective on the implications of AI watermarking on text generation.