Princeton's Narayanan Explains Why Distortion-Free LLM Watermarking Is Possible
soumitrashukla9 · x · 2026-08-18
Princeton's Arvind Narayanan (randomwalker) weighs in on the Claude text watermarking debate, noting a common misconception: many assume watermarking edits the LLM's "best possible output" after generation, hence quality fears. That's not how it works.
During normal generation, many tokens an LLM picks are not its most-likely next token — turning off randomness would produce great first tokens but eventually degenerate into loops. Since humans don't write one word at a time, this locally-optimal-but-globally-incoherent behavior is deeply counterintuitive. So comparing watermarked/unwatermarked pairs, output quality distributions can be unchanged — a stronger distribution-preserving property that is technically well established and has shipped in Google/Gemini for over two years with little attention.
More from Models
- Qwen vs GPT vs Grok: Building a Three.js fragrance site showdown — Acceptable-Object390 · 2026-08-18
- 4B model enables free in-browser Vibecoding — Mysterious_Hearing14 · 2026-08-18
- Benchmark: Qwen 3.8 hits 45 tps with 1M context on 3080Ti + Strix Halo — TrifleHopeful5418 · 2026-08-18
- FinCode-Reasoning-3B: 3B Model Achieves 99.5% Math Accuracy for Financial Coding, Runs on CPU — coslinedev · 2026-08-18
- Qwen and DeepSeek Benchmarks Reportedly Use Claude Code Harness — goddamnit_1 · 2026-08-18
- Gemini vs Grok vs Claude vs Kimi: Sketch-to-3D Mall Showdown — eyishazyer · 2026-08-18