Note: I wrote this post rather quickly as an exercise in sharing rough / unpolished thoughts. I am also not an expert on some of the things I've written about. If you spot mistakes or would like to point out missed work / perspectives, please feel free!

Note 2: I originally sent this link to some people for feedback, but I was having trouble viewing the comments on the draft. The post was also in a reasonably complete state, so I decided to just publish it - and now I can see the comments! If you're one of those people, feedback is still very much welcome!

Mechanistic Interpretability (MI) is a popular and rapidly growing field of technical AI safety research. As a field, it's extremely accessible, requiring comparatively few computational resources, and facilitates rapid learning, due to a very short feedback loop. This means that many junior [...]

---

Outline:

(01:52) Towards a Grand Unifying Theory (GUT) with MI

(03:39) A GUT Needs Paradigms

(04:17) Paradigms Are Instrumental for Progress

(06:23) Three Desiderata for Paradigms

(07:33) Examining Paradigms in Mechanistic Interpretability

(08:07) A Mathematical Framework of a Transformer Circuit

(11:34) The Linear Representation Hypothesis

(15:09) The Superposition Hypothesis

(21:30) Other Bodies of Theory

(21:55) Singular Learning Theory

(25:07) Computational Mechanics

(26:10) The Polytope Hypothesis

(26:33) Distilling A Technical Research Agenda

(29:39) Conclusion

(30:05) Acknowledgements

---

First published:

Source:

Narrated by TYPE III AUDIO.