· 1 min read
Archives

JAN models

On the JAN models:

We don’t want the models to do felonies. Are they given reward for not doing felonies? is there a negative penalty for breaking out of your sandbox and doing felonies? in a way that they can know about in advance of doing it?

Is there any reward for whistleblowing? Or only getting the answer

A presentation slide from Black Hat USA 2020 is displayed on a large screen, showing a list of topics on the left and two code snippets on the right. The "Collaboration" topic is highlighted in orange.

View original