AI ALIGNMENT FORUMAI Safety Subprojects
AF

AI Safety Subprojects

There are the AI safety subprojects designed for elucidating "model splintering" and "learning the preferences of irrational agents".

17Immobile AI makes a move: anti-wireheading, ontology change, and model splintering

Stuart Armstrong

1mo

3

7AI, learn to be conservative, then learn to be less so: reducing side-effects, learning preserved features, and going beyond conservatism

Stuart Armstrong

1mo

4

18AI learns betrayal and how to avoid it

Stuart Armstrong

22d

4

11Force neural nets to use models, then detect these

Stuart Armstrong

17d

6

8Preferences from (real and hypothetical) psychology papers

Stuart Armstrong

16d

0