Speed arguments against scheming (Section 4.4-4.7 of "Scheming AIs")

Joe Carlsmith Audio

Indhold leveret af Joe Carlsmith. Alt podcastindhold inklusive episoder, grafik og podcastbeskrivelser uploades og leveres direkte af Joe Carlsmith eller deres podcastplatformspartner. Hvis du mener, at nogen bruger dit ophavsretligt beskyttede værk uden din tilladelse, kan du følge processen beskrevet her https://da.player.fm/legal.

1+ y ago 15:19

MP3•Episode hjem

This is section 4.4 through 4.7 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

Kapitler

1. Speed arguments against scheming (Section 4.4-4.7 of "Scheming AIs") (00:00:00)

2. 4.4 Speed arguments (00:00:29)

3. 4.4.1 How big are the absolute costs of this extra reasoning? (00:02:22)

4. 4.4.2 How big are the costs of this extra reasoning relative to the simplicity benefits of (00:07:06)

5. 4.4.3 Can we actively shape training to bias towards speed over simplicity? (00:09:21)

6. 4.5 The “not-your-passion” argument (00:10:27)

7. 4.6 The relevance of “slack” to these arguments (00:12:46)

8. 4.7 Takeaways re: arguments that focus on the final properties of the model (00:13:38)

63 episoder

#Society #Philosophy #Joe