 | It's my article, it is about how to circumvent any even theoretical optimal AI watermark based on statistical biases via pseudorandom generators like Google's SynthID. Let me know what you guys think. Generally, I do not think watermarking is the right solution, hence I am sharing my idea how to circumvent it. How many thesises are out there that are basically slop but made with human effort. Now text length is not a valid measure anymore, you actually have to do some real research. I think that is awesome. [link] [comments] |
Predictive technology: computational social science