Salima Lamsiyah

Salima Lamsiyah is a researcher in the Department of Computer Science at the University of Luxembourg working on natural language processing, representation learning, and mechanistic interpretability. Her work evaluates whether sparse dictionary decompositions of transformer activations provide faithful accounts of internal model computation.

In October 2026, Lamsiyah led “Sparse Autoencoders for Mechanistic Interpretability in Natural Language Processing: A Critical Survey” (arXiv:2610.20139), auditing 138 published SAE studies through a four-part formal taxonomy $T(s) = (D_s, O_s, F_s; Z_s)$ covering dictionary design, feature operations, NLP functions, and experimental qualifiers. The survey establishes that four outcomes routinely conflated in the literature, feature recoverability, semantic interpretability, mechanistic faithfulness, and behavioral utility, require distinct experimental controls and frequently diverge.

Lamsiyah’s profile is Builder, included for the 138-study evidence register and the methodological demonstration that steering or probing success with an SAE feature does not by itself prove that the intact model computes with that feature.

Known for. Critical survey of 138 Sparse Autoencoder (SAE) studies for mechanistic interpretability in NLP, four-outcome separation of recoverability, semantic interpretability, mechanistic faithfulness, and behavioral utility

Coverage on this site

• Objective-Level Recovery Gaps in Mechanistic Interpretability and SAEs October 2026

Related researchers

• XS Xujie SiDepartment of Computer Science, University of Toronto. Mila - Quebec AI InstituteObjective-level recovery gaps and context distortion in mechanistic interpretability circuit discovery, program synthesis and neuro-symbolic verification Builder1 article
• CG Chuqin GengSchool of Computer Science, McGill University. Mila - Quebec AI InstituteObjective-level recovery gaps and context distortion in mechanistic interpretability circuit evaluation, neural network verification Builder1 article
• PB Patrick ButlinEleos AI Research. Previously the Global Priorities Institute, OxfordLead author of the indicator properties framework for assessing consciousness in AI systems AI welfare24 articles
• VA Vikas AgnihotriIndependent Aerospace and Neuromorphic Systems ResearcherNeuromorphic Satellite Intelligence framework, radiation-induced energy overhead model for spiking neural hardware, constrained on-orbit STDP Builder1 article
• MA Matt AngleParadromicsFounder and CEO of Paradromics, developer of the Connexus Direct Data Interface high-data-rate brain-computer interface Builder1 article
• IB Isabella R. BeckettCoauthor of the MaleCNS preprint comparing male and female Drosophila brain connectomes Builder1 article

Salima Lamsiyah is listed with the other builders, who hold that the claim is about what the hardware does, and the philosophy is left open. The positions themselves are set out in the index of consciousness theories. Every researcher covered on this site is indexed in the directory of consciousness researchers.