20 - 'Reform' AI Alignment with Scott Aaronson
AXRP - the AI X-risk Research Podcast
The Robustness of a Cryptographic Scheme
OpenAI engineer Hendrick Parchner has implemented a prototype of the system, and it seems to work very well. The scheme that we've come up with is robust against any, you know, attacker who would use GPT to generate something, and then make some local modifications in text. Because the score is just a sum over N grams, right? Yeah, as long as you preserve a large fraction of the N grams,you're still good,. There are even more sophisticated things you can do. You can, you could have multiple watermarks. And if you want your text to not be completely deterministic, you could sort of blend watermarking with the normal GPT generation
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
