Stefan Heimersheim

20 papers A* 2Journal 18
YearRankTypeTitle / Venue / Authors
2026 J jnl
Concept Influence: Leveraging Interpretability to Improve Performance and Efficiency in Training Data Attribution.
CoRR
Matthew Kowal, Goncalo Paulo, Louis Jaburi, Tom Tseng, Lev E. McKinney, Stefan Heimersheim, Aaron David Tucker, Adam Gleave, Kellin Pelrine
2026 J jnl
The Obfuscation Atlas: Mapping Where Honesty Emerges in RLVR with Deception Probes.
CoRR
Mohammad Taufeeque, Stefan Heimersheim, Adam Gleave, Chris Cundy
2025 J jnl
Benchmarking Deception Probes via Black-to-White Performance Boosts.
CoRR
Avi Parrack, Carlo Leonardo Attubato, Stefan Heimersheim
2025 J jnl
Detecting Strategic Deception Using Linear Probes.
CoRR
Nicholas Goldowsky-Dill, Bilal Chughtai, Stefan Heimersheim, Marius Hobbhahn
2025 A* conf
Detecting Strategic Deception with Linear Probes.
ICML
Nicholas Goldowsky-Dill, Bilal Chughtai, Stefan Heimersheim, Marius Hobbhahn
2025 J jnl
Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition.
CoRR
Dan Braun, Lucius Bushnaq, Stefan Heimersheim, Jake Mendel, Lee Sharkey
2025 J jnl
Open Problems in Mechanistic Interpretability.
CoRR
Lee Sharkey, Bilal Chughtai, Joshua Batson, Jack Lindsey, Jeff Wu, Lucius Bushnaq, Nicholas Goldowsky-Dill, Stefan Heimersheim, Alejandro Ortega, Joseph Isaac Bloom, Stella Biderman, Adrià Garriga-Alonso, Arthur Conmy, Neel Nanda, Jessica Rumbelow, Martin Wattenberg, Nandi Schoots, Joseph Miller, Eric J. Michaud, Stephen Casper, Max Tegmark, William Saunders, David Bau, Eric Todd, Atticus Geiger, Mor Geva, Jesse Hoogland, Daniel Murfet, Tom McGrath
2025 J jnl
Open Problems in Mechanistic Interpretability.
Trans. Mach. Learn. Res.
Lee Sharkey, Bilal Chughtai, Joshua Batson, Jack Lindsey, Jeffrey Wu, Lucius Bushnaq, Nicholas Goldowsky-Dill, Stefan Heimersheim, Alejandro Ortega, Joseph Isaac Bloom, Stella Biderman, Adrià Garriga-Alonso, Arthur Conmy, Neel Nanda, Jessica Rumbelow, Martin Wattenberg, Nandi Schoots, Joseph Miller, William Saunders, Eric J. Michaud, Stephen Casper, Max Tegmark, David Bau, Eric Todd, Atticus Geiger, Mor Geva, Jesse Hoogland, Daniel Murfet, Tom McGrath
2025 J jnl
SCALAR: Benchmarking SAE Interaction Sparsity in Toy LLMs.
CoRR
Sean P. Fillingham, Andrew Gordon, Peter Lai, Xavier Poncini, David Quarel, Stefan Heimersheim
2025 J jnl
Transformers Don't Need LayerNorm at Inference Time: Scaling LayerNorm Removal to GPT-2 XL and the Implications for Mechanistic Interpretability.
CoRR
Luca Baroni, Galvin Khara, Joachim Schaeffer, Marat Subkhankulov, Stefan Heimersheim
2024 J jnl
Characterizing stable regions in the residual stream of LLMs.
CoRR
Jett Janiak, Jacek Karwowski, Chatrik Singh Mangat, Giorgi Giglemiani, Nora Petrova, Stefan Heimersheim
2024 J jnl
Evaluating Synthetic Activations composed of SAE Latents in GPT-2.
CoRR
Giorgi Giglemiani, Nora Petrova, Chatrik Singh Mangat, Jett Janiak, Stefan Heimersheim
2024 J jnl
Evolution of SAE Features Across Layers in LLMs.
CoRR
Daniel Balcells, Benjamin Lerner, Michael Oesterle, Ediz Ucar, Stefan Heimersheim
2024 J jnl
How to use and interpret activation patching.
CoRR
Stefan Heimersheim, Neel Nanda
2024 J jnl
Investigating Sensitive Directions in GPT-2: An Improved Baseline and Comparative Analysis of SAEs.
CoRR
Daniel J. Lee, Stefan Heimersheim
2024 J jnl
The Local Interaction Basis: Identifying Computationally-Relevant and Sparsely Interacting Features in Neural Networks.
CoRR
Lucius Bushnaq, Stefan Heimersheim, Nicholas Goldowsky-Dill, Dan Braun, Jake Mendel, Kaarel Hänni, Avery Griffin, Jörn Stöhler, Magdalena Wache, Marius Hobbhahn
2024 J jnl
Using Degeneracy in the Loss Landscape for Mechanistic Interpretability.
CoRR
Lucius Bushnaq, Jake Mendel, Stefan Heimersheim, Dan Braun, Nicholas Goldowsky-Dill, Kaarel Hänni, Cindy Wu, Marius Hobbhahn
2024 J jnl
You can remove GPT2's LayerNorm by fine-tuning.
CoRR
Stefan Heimersheim
2023 A* conf
Towards Automated Circuit Discovery for Mechanistic Interpretability.
NeurIPS
Arthur Conmy, Augustine N. Mavor-Parker, Aengus Lynch, Stefan Heimersheim, Adrià Garriga-Alonso
2023 J jnl
Towards Automated Circuit Discovery for Mechanistic Interpretability.
CoRR
Arthur Conmy, Augustine N. Mavor-Parker, Aengus Lynch, Stefan Heimersheim, Adrià Garriga-Alonso