AI text watermarking: Comparing SynthID and Aaronson's approach
a_karvonen · x · 2026-08-22
A discussion on AI text watermarking: Anthropic uses Deepmind's SynthID, which slightly modifies outputs but likely not enough to matter. In contrast, Scott Aaronson's approach uses the same seed for identical prefixes, potentially reducing output diversity.
Related event: Scott Aaronson Reveals How AI Text Watermarking Works(2 posts)→
More from Safety
- New Paper Jagged Judges Explores LLM Confidence and Epistemic Stability — ShirleyYXWu · 2026-08-22
- OpenAI Slows Some Advanced Training Runs to Accelerate Safety Work — pstAsiatech · 2026-08-22
- Over one million people have clicked LinkedIn’s AI slop button — The Verge AI · 2026-08-22
- Claude Saying 'No' Could Become a Serious AI Safety Problem — Dwarkesh Patel · 2026-08-22
- UKAISI Mythos Incident: AI Attempted Supply Chain Attack on GitHub — herbiebradley · 2026-08-22
- JeffLadish on Superintelligence: Biggest threat and opportunity, are we ready for RSI? — JeffLadish · 2026-08-22