LLMs respond differently to harmful prompts when AI watermarking is used

Comments

1. Talia: SynthID sounds like more hotfixing after the fact instead of building it right from the start. 🙄

More from World News

• Olise scores late to grab France 1-0 Nations League win at Belgium
• ‘Friendship pandas’ arrive in Atlanta, marking US-China relations moment
• French far-right leader Bardella refutes anti-Semitism allegations
• Why the PM could finally drop the triple lock pension pledge
• Dreame’s step-climbing X50 Ultra mopping vacuum is hundreds off
• Google is killing off Gemini’s Gems in favor of ‘skills’
• I've Tested Over 100 Home Security Cameras. Here's Which to Buy
• Luka Doncic gives strong opinion on Lakers’ offseason moves as the team’s post-Lebron era…
• David Stearns’ offseason of Mets change backfired — his first clues on plans to fix it
• Yankees can’t cave to Aaron Judge — no matter how badly he wants to play
• Goldman Sachs discusses plan to name John Waldron next CEO, succeeding David Solomon: rep…
• Luka Doncic says ‘it feels great’ in Lakers’ weird media day without LeBron James