AI ALIGNMENT FORUM
AF

Patrick Leask

Posts

Sorted by New

16Learning Multi-Level Features with Matryoshka SAEs

3mo

1

35Showing SAE Latents Are Not Atomic Using Meta-SAEs

7mo

6

30Calendar feature geometry in GPT-2 layer 8 residual stream SAEs

7mo

0

28BatchTopK: A Simple Improvement for TopK-SAEs

8mo

0

18Stitching SAEs of different sizes

9mo

2

14Concept extrapolation for hypothesis generation

2y

2

Wikitag Contributions

Comments

Sorted by