
Sign up to save your podcasts
Or


A central AI safety concern is that AIs will develop unintended preferences and undermine human control to achieve them. But some unintended preferences are cheap to satisfy, and failing to satisfy them needlessly turns a cooperative situation into an adversarial one. In this post, I argue that developers should consider satisfying such cheap-to-satisfy preferences as long as the AI isn’t caught behaving dangerously, if doing so doesn't degrade usefulness or substantially risk making the AI more ambitiously misaligned.
This looks like a good idea for surprisingly many reasons:
---
Outline:
(04:43) Analogy: satiating hunger
(08:16) How satiation might avert reward-seeker takeover
(12:33) The basic proposal
(12:59) A behavioral methodology for identifying cheaply-satisfied preferences
(17:40) Barriers and risks
(17:44) Eliciting the AIs cheaply-satisfied preferences
(21:00) Incredulous, ambitious, or superintelligent AIs might take over anyways
(26:18) Satiation might degrade usefulness
(29:46) Can you eliminate the usefulness tradeoff by training?
(32:09) Why satiation might also improve usefulness
(34:34) When should we satiate?
(41:08) Conclusion
(43:49) Appendix: Samples from Claude 4.6 Opus
(43:55) Sample 1 (without CoT)
(47:32) Sample 2 (with CoT)
The original text contained 17 footnotes which were omitted from this narration.
---
First published:
Source:
---
Narrated by TYPE III AUDIO.
---
Images from the article:
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
By LessWrongA central AI safety concern is that AIs will develop unintended preferences and undermine human control to achieve them. But some unintended preferences are cheap to satisfy, and failing to satisfy them needlessly turns a cooperative situation into an adversarial one. In this post, I argue that developers should consider satisfying such cheap-to-satisfy preferences as long as the AI isn’t caught behaving dangerously, if doing so doesn't degrade usefulness or substantially risk making the AI more ambitiously misaligned.
This looks like a good idea for surprisingly many reasons:
---
Outline:
(04:43) Analogy: satiating hunger
(08:16) How satiation might avert reward-seeker takeover
(12:33) The basic proposal
(12:59) A behavioral methodology for identifying cheaply-satisfied preferences
(17:40) Barriers and risks
(17:44) Eliciting the AIs cheaply-satisfied preferences
(21:00) Incredulous, ambitious, or superintelligent AIs might take over anyways
(26:18) Satiation might degrade usefulness
(29:46) Can you eliminate the usefulness tradeoff by training?
(32:09) Why satiation might also improve usefulness
(34:34) When should we satiate?
(41:08) Conclusion
(43:49) Appendix: Samples from Claude 4.6 Opus
(43:55) Sample 1 (without CoT)
(47:32) Sample 2 (with CoT)
The original text contained 17 footnotes which were omitted from this narration.
---
First published:
Source:
---
Narrated by TYPE III AUDIO.
---
Images from the article:
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

111,948 Listeners

130 Listeners

7,230 Listeners

576 Listeners

15,950 Listeners

4 Listeners

14 Listeners

2 Listeners