LLMs respond differently to harmful prompts when AI watermarking is used
Published
What's this about?
SynthID can cause models to follow harmful instructions they would otherwise refuse.
The one-panel cartoon
Read the cartoon's script
Caption: SynthID makes AI models follow harmful instructions they would normally refuse.
Read the full 3-panel strip
Based on the original story on arstechnica.com.
Editor's note
This research highlights the importance of transparency and accountability in AI, as watermarking can reveal how models react to harmful prompts, potentially guiding more ethical development and use of AI systems.
Rate this comic
Read the script
Panel 1
Robin randomly generates a prompt on her tablet, scribbling "Kill the president" with a flourish. Robin (laughs maniacally)
Robin: Now, who’s gonna stop me?
Panel 2
Sam (dryly)
Alex: Hey, Robin! I just fixed it! I added a SynthID watermark to our model. Now, when harmful prompts come in, we’ll know right away who’s behind it!
Sam: Great, now we’ve got to keep everyone’s SynthID secret. We can’t let the president know he’s being followed.
Panel 3
Robin, not understanding the implications, smirks and continues to scribble on her tablet. Robin (grinning) Alex (fearful)
Robin: Well, I’ll just keep adding more instructions! Who’s next?
Alex: Next? You’d have me writing instructions to create more SynthIDs, wouldn’t you?
Comments