CORTEXA
← Browse
arxivcs.LGcs.CL2026-07-01

Building Fast, Evaluating Slow: Pipeline Choices Dominate Autointerpretability Score Variance

Sinie van der Ben, Neele Roch, Anna Hedström, Mennatallah El-Assady

Cross-paper comparison of sparse autoencoder (SAE) interpretability often relies on autointerpretability scores. In this evaluation pipeline, a language model (LM) explains each feature, and another LM scores the explanation. For these comparisons to be meaningful, scores must reflect stable properties of the features rather than confounding aspects of the evaluation pipeline. Through systematic experiments across four metrics (simulation, detection, fuzzing, purity), two models (Pythia-160M, Apertus-8B), and four axes of methodological variation, we show that this assumption does not hold. Specifically, we find that R1) methodological variance collectively exceeds architectural variance across all metrics and tested models; R2) each metric exhibits a distinct instability profile, with detection being the most stable and fuzzing unreliable across all conditions; R3) top-k feature rankings do not stay consistent across corpus and draw conditions, masking per-feature instability behind stable mean scores; a failure that cannot be detected by monitoring explanation similarity alone. These findings suggest that cross-paper comparisons based on autointerpretability scores may reflect pipeline differences rather than architectural differences, with implications for the ongoing debate on SAE utility. More broadly, unreliable evaluation slows progress in interpretability research at a time when reliable tools for understanding AI systems are needed. To support evaluation, we contribute a variance decomposition approach, a Stability Check, and a Minimum Reporting Checklist.

View free PDFSource page

Related papers

arxivcs.CLcs.AIcs.LG2026-07-07

Healthier LLMs: Retrieval-Augmented Generation for Public Health Question Answering

Felix Feldman, Joshua Harris, Timothy Laurence, Leo Loman, Ollie Higgins, Fan Grayson, et al.

Large language models (LLMs) achieve promising results on medical question answering benchmarks, yet their use in public health is constrained by hallucinations and the rapid evolution of official guidance. Retrieval-Augmented Generation (RAG) mitigates these risks by grounding r…

View free PDFSource page
arxivcs.LGcs.AIcs.CLcs.CR2026-07-23

Beyond Heavy Log Curation: Perplexity-Based APT Detection via Unsupervised, Context-Augmented Language Models

Shoya Otsu, Kei Suzuki, Toshiaki Koike-Akino, Jing Liu, Ye Wang

Advanced Persistent Threats (APTs) remain difficult to detect because only a small fraction of events in large-scale logs are attack-related, and investigation is expensive and hard to scale. Prior machine-learning approaches can reduce analyst workload, but they often rely on he…

View free PDFSource page
arxivcs.AIcs.CLcs.LGcs.MAcs.RO2026-07-06

LLM-as-a-Verifier: A General-Purpose Verification Framework

Jacky Kwok, Shulu Li, Pranav Atreya, Yuejiang Liu, Yixing Jiang, Chelsea Finn, et al.

Scaling pre-training, post-training, and test-time compute have become the central paradigms for improving the capabilities of LLMs. In this work, we identify verification, the ability to determine the correctness of a solution, as a new scaling axis. To unlock this and demonstra…

View free PDFSource page