llm watermarking is utilizing #synchID can be easily solved
llm watermarking is utilizing #synchID can be easily solved just by tokens resampling using small optimized open weight models or cheap chineese cloud models misAnthropic is basically incentivizing its own users to apply distillation for its own products rn best technical explanation video: https://youtu.be/FnAqruxx-QE?si
The excerpt is quoted from a public Telegram channel, which owns its text.