AI ALIGNMENT FORUMTags
AF

Apollo Research (org)

EditHistorySubscribe

Help improve this page

EditHistorySubscribe

Help improve this page

Apollo Research (org)

Contributors

Posts tagged Apollo Research (org)

1

103SAE feature geometry is outside the superposition hypothesis

5mo

0

1

88Announcing Apollo Research

Marius Hobbhahn, Beren Millidge, Lee Sharkey, Lucius Bushnaq, Dan Braun, Mikita Balesni, Jérémy Scheurer

1y

4

1

55You can remove GPT2’s LayerNorm by fine-tuning for an hour

Stefan Heimersheim

3mo

3

1

54A List of 45+ Mech Interp Project Ideas from Apollo Research’s Interpretability Team

Lee Sharkey, Lucius Bushnaq, Dan Braun, Stefan Heimersheim, Nicholas Goldowsky-Dill

4mo

7

1

56The Local Interaction Basis: Identifying Computationally-Relevant and Sparsely Interacting Features in Neural Networks

Lucius Bushnaq, Jake Mendel, Dan Braun, Stefan Heimersheim, Nicholas Goldowsky-Dill, Kaarel Hänni, Avery, Joern Stoehler, Cindy Wu, Magdalena Wache, Marius Hobbhahn

6mo

1

1

44Sparsify: A mechanistic interpretability research agenda

8mo

17

1

42Apollo Research 1-year update

Marius Hobbhahn, Lee Sharkey, Lucius Bushnaq, Dan Braun, Mikita Balesni, Jérémy Scheurer, Nicholas Goldowsky-Dill, Stefan Heimersheim, Jake Mendel, AlexMeinke, rusheb

6mo

0

1

38We need a Science of Evals

Marius Hobbhahn, Jérémy Scheurer

10mo

9

1

34An Opinionated Evals Reading List

Marius Hobbhahn, Jérémy Scheurer

1mo

0

1

35[Interim research report] Activation plateaus & sensitive directions in GPT2

Stefan Heimersheim, Jake Mendel

5mo

0

1

33Understanding strategic deception and deceptive alignment

Marius Hobbhahn, Mikita Balesni, Jérémy Scheurer, Dan Braun

1y

0

1

28Identifying Functionally Important Features with End-to-End Sparse Dictionary Learning

Dan Braun, Jordan Taylor, Nicholas Goldowsky-Dill, Lee Sharkey

6mo

8

1

25Theories of Change for AI Auditing

Lee Sharkey, Beren Millidge, Marius Hobbhahn

1y

0

1

27A starter guide for evals

Marius Hobbhahn, Jérémy Scheurer, Mikita Balesni, rusheb, AlexMeinke

10mo

0