Share
The simplest questions often have the most complex answers. The Philosopher's Zone is your guide through the strange thickets of logic, metaphysics and ethics.
…
continue reading
Manage episode 387808067 series 3402048
Content provided by Joe Carlsmith. All podcast content including episodes, graphics, and podcast descriptions are uploaded and provided directly by Joe Carlsmith or their podcast platform partner. If you believe someone is using your copyrighted work without your permission, you can follow the process outlined here https://staging.podcastplayer.com/legal.
This is section 3 of my report “Scheming AIs: Will AIs fake alignment during training in order to get power?”
Text of the report here: https://arxiv.org/abs/2311.08379
Summary of the report here: https://joecarlsmith.com/2023/11/15/new-report-scheming-ais-will-ais-fake-alignment-during-training-in-order-to-get-power
Audio summary here: https://joecarlsmithaudio.buzzsprout.com/2034731/13969977-introduction-and-summary-of-scheming-ais-will-ais-fake-alignment-during-training-in-order-to-get-power
Chapters
1. Arguments for/against scheming that focus on the path SGD takes (Section 3 of "Scheming AIs") (00:00:00)
2. 3. Arguments for/against scheming that focus on the path that SGD takes (00:00:35)
3. 3.1 The training-game-independent proxy-goals story (00:02:38)
4. 3.2 The “nearest max-reward goal” story (00:07:14)
5. 3.2.1 Barriers to schemer-like modifications from SGD’s incrementalism (00:12:21)
6. 3.2.2 Which model is “nearest”? (00:13:53)
7. 3.2.2.1 The common-ness of schemer-like goals in goal space (00:14:28)
8. 3.2.2.2 The nearness of non-schemer goals (00:17:43)
9. 3.2.2.3 The relevance of messy goal-directedness to nearness (00:22:53)
10. 3.2.3 Overall take on the “nearest max-reward goal” argument (00:24:30)
11. 3.3 The possible relevance of properties like simplicity and speed to the path SGD takes (00:25:22)
12. 3.4 Overall assessment of arguments that focus on the path SGD takes (00:27:33)
67 episodes
))
