LCDD creates sparse carriers for SFT behaviors that SFT-Eraser can reverse, with ablations showing the sparse structure enables causal control.
Towards automated circuit discovery for mechanistic interpretability.Advances in Neural Information Processing Systems, 36:16318–16352
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
citation-role summary
background 2
citation-polarity summary
years
2026 5roles
background 2polarities
background 2representative citing papers
Transformer circuits show free evolution during SFT, rendering static mechanistic localization inadequate for future parameter updates due to inherent temporal latency.
On Llama-2-7B, path-rerouting magnitude in paired transcoder attribution graphs correlates with jailbreak success (r=0.461), while static node metrics and top-feature ablations do not.
SoftSAE replaces fixed-K sparsity in autoencoders with a learned, input-dependent number of active features via a soft top-k operator.
RET learns temporally consistent macrovariables from LLM activations via self-supervised learning to support interpretability, early behavioral prediction, and causal intervention.
citing papers explorer
-
Crafting Reversible SFT Behaviors in Large Language Models
LCDD creates sparse carriers for SFT behaviors that SFT-Eraser can reverse, with ablations showing the sparse structure enables causal control.
-
Navigating by Old Maps: The Pitfalls of Static Mechanistic Localization in LLM Post-Training
Transformer circuits show free evolution during SFT, rendering static mechanistic localization inadequate for future parameter updates due to inherent temporal latency.
-
Mechanistic Interpretability of LLM Jailbreaks via Internal Attribution Graphs
On Llama-2-7B, path-rerouting magnitude in paired transcoder attribution graphs correlates with jailbreak success (r=0.461), while static node metrics and top-feature ablations do not.
-
SoftSAE: Dynamic Top-K Selection for Adaptive Sparse Autoencoders
SoftSAE replaces fixed-K sparsity in autoencoders with a learned, input-dependent number of active features via a soft top-k operator.
-
Towards Effective Theory of LLMs: A Representation Learning Approach
RET learns temporally consistent macrovariables from LLM activations via self-supervised learning to support interpretability, early behavioral prediction, and causal intervention.