Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.

TL;DR: We demonstrate that the decoder directions of GPT-2 SAEs are highly structured by finding a historical date direction onto which projecting non-date related features lets us read off their historical time period by comparison to year features.

Calendar years are linear: there are as many years between 2000 and 2024, as there are between 1800 and 1824. Linear probes can be used to predict years of particular events from the activations of language models. Since calendar years are linear, one might think the same of other time-based features such as weekday features, however weekday activations in sparse autoencoders (SAEs) were recently found to be arranged in a circular configuration in their top principal components. Inspired by this, we looked into weekdays, months, and most interestingly calendar years from the perspective of SAE feature decoder similarity.

For [...]

---

Outline:

(01:35) Features at different scales

(05:19) Interpreting Principal Components

(07:34) Finding Directions

---

First published:
August 17th, 2024

Source:
https://www.lesswrong.com/posts/WsPyunwpXYCM2iN6t/calendar-feature-geometry-in-gpt-2-layer-8-residual-stream

---

Narrated by TYPE III AUDIO.

---

“Calendar feature geometry in GPT-2 layer 8 residual stream SAEs ” by Patrick Leask, Bart Bussmann, Neel Nanda

LessWrong (30+ Karma)