RACE (Residual Alignment for Consistency Estimation), a forward-pass statistical framework that evaluates the domain-wide functional consistency of Transformer neurons, produces neuron rankings that yield more domain-specific effects under perturbation.
Abstract
Discovering stable neuron behavior across entire domains remains a challenge in mechanistic interpretability. Existing methods often rely on instance-level point estimates or computationally expensive procedures, which either obscure population-level variability or limit scalable domain-wide analysis. We present RACE (Residual Alignment for Consistency Estimation), a forward-pass statistical framework that evaluates the domain-wide functional consistency of Transformer neurons. Compared with gradient-based point estimates, RACE produces neuron rankings that yield more domain-specific effects under perturbation. Token-distribution shifts support the connection between the selected neurons and the target domain, while scoring requires roughly one-hundredth of the computational overhead of the gradient-based methods. Code is available at https://github.com/Nexround/RACE.
This work quantifies functional degeneracy through the behavioral recovery rank, defined as the number of leading behavioral-Hessian eigendirections required to recover a trained model's performance, and finds that structural and magnitude pruning retain more degrees of freedom, even after the task is saturated.
M. Matveev, P. Esser, Ayush Bharadwaj et al.· 0 citations
Large language models (LLMs) are predominantly aligned to function as passive, sycophantic assistants. We challenge this default paradigm by empirically evaluating the cognitive plasticity of open-weight architectures when subjected to rigorous behavioral reprogramming. Our objective is to induce a proactive, Socratic...
This work proposes Topological Steering, a new framework for steering LLM behavior through the topological representation of activation spaces, and shows that Topological Steering consistently modifies LLM behavior across multiple model families and model sizes.
Steering language models using a set of contrastive representations has been a canonical and computationally efficient method for controlling model behavior. Despite this success in controlling certain model behaviors, the effectiveness of activation steering varies markedly across concepts; the generalization properti...
Generalization in large language models (LLMs) is the ability to produce consistent and semantically stable outputs when the same input is expressed in different ways. Existing work typically evaluates generalization through aggregate accuracy on a single prompt format, task, or set of variations, which conflates robus...
Nagham Omar, Mahmoud Jabarin, M. Rozenshteĭn et al.· 0 citations
Diffusion language models generate code by repeatedly updating a partially masked sequence. We ask whether their internal activations encode code correctness and whether that information can improve generation. Across six diffusion models, linear probes distinguish passing from failing attempts, with the strongest read...
A. Miglani, Samrath Singh Chadha, Kevin Li et al.· 0 citations
