RSS

Connor Kissane

Karma: 375

SAEs are highly dataset de­pen­dent: a case study on the re­fusal direction

Connor Kissane, robertzk, Neel Nanda and Arthur Conmy
Nov 7, 2024, 5:22 AM
66 points
4 comments14 min readLW link

Open Source Repli­ca­tion of An­thropic’s Cross­coder pa­per for model-diffing

Connor Kissane, robertzk, Arthur Conmy and Neel Nanda
Oct 27, 2024, 6:46 PM
48 points
4 comments5 min readLW link

Base LLMs re­fuse too

Connor Kissane, robertzk, Arthur Conmy and Neel Nanda
Sep 29, 2024, 4:04 PM
60 points
20 comments10 min readLW link

SAEs (usu­ally) Trans­fer Between Base and Chat Models

Connor Kissane, robertzk, Arthur Conmy and Neel Nanda
Jul 18, 2024, 10:29 AM
67 points
0 comments10 min readLW link

At­ten­tion Out­put SAEs Im­prove Cir­cuit Analysis

Connor Kissane, robertzk, Arthur Conmy and Neel Nanda
Jun 21, 2024, 12:56 PM
33 points
3 comments19 min readLW link

We In­spected Every Head In GPT-2 Small us­ing SAEs So You Don’t Have To

robertzk, Connor Kissane, Arthur Conmy and Neel Nanda
6 Mar 2024 5:03 UTC
63 points
0 comments12 min readLW link

At­ten­tion SAEs Scale to GPT-2 Small

Connor Kissane, robertzk, Arthur Conmy and Neel Nanda
3 Feb 2024 6:50 UTC
78 points
4 comments8 min readLW link

Sparse Au­toen­coders Work on At­ten­tion Layer Outputs

Connor Kissane, robertzk, Arthur Conmy and Neel Nanda
16 Jan 2024 0:26 UTC
85 points
9 comments18 min readLW link