TL;DR: In this post, I distinguish between two related concepts in neural network interpretability: polysemanticity and superposition. Neuron polysemanticity is the observed phenomena that many neurons seem to fire (have large, positive activations) on multiple unrelated concepts. Superposition is a specific explanation for neuron (or attention head) polysemanticity, where a neural network represents more sparse features than there are neurons (or number of/dimension of attention heads) in near-orthogonal directions. I provide three ways neurons/attention heads can be polysemantic without superposition: non--neuron aligned orthogonal features, non-linear feature representations, and compositional representation without features. I conclude by listing a few reasons why it might be important to distinguish the two concepts.

Epistemic status: I wrote this “quickly” in about 10 hours, as otherwise it wouldn’t have come out at all. Think of it as a (failed) experiment in writing [...]

---

Outline:

(04:23) A brief review of polysemanticity and superposition

(04:28) Neuron polysemanticity

(08:01) Superposition

(12:17) Polysemanticity without superposition

(12:32) Example 1: non–neuron aligned orthogonal features

(17:25) Example 2: non-linear feature representations

(19:01) Example 3: compositional representation without “features”

(20:35) Conclusion: why does this distinction matter?

(21:38) Our current model of superposition may not fully explain neuron polysemanticity, so we should keep other hypotheses in mind

(23:53) Attempts to “solve superposition” may actually only be solving easier cases of polysemanticity

(25:02) Clear definitions are important for clear communication and rigorous science

(25:47) Acknowledgements

The original text contained 24 footnotes which were omitted from this narration.

---

First published:

Source:

Narrated by TYPE III AUDIO.