I really like this quote “Reality is the emerging totality of what different observers experience” from this book I’m reading. Applies beyond quantum mechanics!
How much do LLMs’ watermarks distort your output? 🔍
We present SimpleMark: a fast multi-bit text watermark that bounds the distortion of every token to a chosen f-divergence budget. Our watermark’s text distribution stays statistically close to what the model would have written without a watermark while remaining competitive in all benchmarked performance categories.
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
https://t.co/7N6TPlft1P
4/ I wrote a blog post explaining the intuition, examples, communication gains, logarithmic threshold, and open questions: https://t.co/XGVc1TBX5I. Check it out!
2/ Allowing some multiplicative leakage can significantly reduce communication. But does extra leakage also let us provide new proofs for fundamentally new problems? Up to logarithmic leakage, the answer is no! SZK = MZK.
3/ so there is a regime where you can relax privacy enough to gain in efficiency, without increasing the class of problems you can prove. Beyond logarithmic leakage, however, the story changes: MZK is strictly more expressive (under standard assumptions).
1/ How much can a zero-knowledge proof leak before it stops being “zero-knowledge”? Surprisingly, quite a bit. Our new work studies multiplicative zero-knowledge (MZK) and discovers some complexity-theoretic transition.