In this post, we list 7 of our favorite easy-to-start directions in AI control. (Really, projects that are at least adjacent to AI control; We include directions which aren’t as centrally AI control and which also have other motivations.) This list is targeted at independent researchers who don't necessarily have a ton of context on AI control (but have at least some context via e.g., having read some of our posts on control) and who don't want to do infrastructure and compute heavy experiments. In the final section, we also list 3 additional directions that are probably a bit more challenging but still tractable.
We’ll mostly be highlighting and expanding upon areas that Ryan discussed in an earlier post: An overview of areas of control work.
We don’t discuss work on constructing control settings or iterating on methods within control settings because we think good work of that type [...]
Outline:
(01:17) Techniques for doing elicitation without learning and studying overelicitation
(03:17) Generating indistinguishable synthetic inputs
(06:46) Teaching models synthetic facts
(10:05) Further studying exploration hacking
(13:14) Malign AI agent substitution
(14:16) Studying data poisoning
(16:03) Training attack policies
(18:00) Bonus: 3 slightly more challenging (but important and tractable) directions
(18:08) Few-shot catastrophe prevention
(19:45) Studying collusion and developing countermeasures
(21:42) Studying steganography and developing countermeasures
The original text contained 1 footnote which was omitted from this narration.
First published:
April 28th, 2025
Source:
https://www.lesswrong.com/posts/wwshEdNhwwT4r9RQN/7-tractable-directions-in-ai-control)
---
Narrated by TYPE III AUDIO).
