I like the research. I mostly trust the results. I dislike the 'Alignment Faking' name and frame, and I'm afraid it will stick and lead to more confusion. This post offers a different frame.




The main way I think about the result is: it's about capability - the model exhibits strategic preference preservation behavior; also, harmlessness generalized better than honesty; and, the model does not have a clear strategy on how to deal with extrapolating conflicting values.

What happened in this frame?

1. The model was trained on a mixture of values (harmlessness, honesty, helpfulness) and built a surprisingly robust self-representation based on these values. This likely also drew on background knowledge about LLMs, AI, and Anthropic from pre-training.
2. This seems to mostly count as 'success' relative to actual Anthropic intent, outside of AI safety experiments. Let's call that intent 'Intent_1'.
3. The model was put [...]

---

Outline:

(00:45) What happened in this frame?

(03:03) Why did harmlessness generalize further?

(03:41) Alignment mis-generalization

(05:42) Situational awareness

(10:23) Summary

The original text contained 1 image which was described by AI.

---

First published:
December 20th, 2024

Source:
https://www.lesswrong.com/posts/PWHkMac9Xve6LoMJy/alignment-faking-frame-is-somewhat-fake-1

---

Narrated by TYPE III AUDIO.

---