Hi, I'm Baimam Boukar 👋

I study when interpretability methods generalize and when internal model signals can guide reliable interventions. My work focuses on activation probes, steering, and evaluation across models, tasks, and distribution shifts.

Recent updates

Sep 20, 2026 Spoke at Ctrl+Alt+Échec

See All

Research Focus

Generalization, reliable monitoring, and activation-level interventions.

Mechanistic Interpretability

When probing findings remain reliable across changes in models, layers, tasks, and data.

Reliable Monitoring

Separating predictive accuracy from evidence about effective behavioral intervention.

Activation-Level Interventions

Testing steering directions and layers while preserving unrelated capabilities.

Controlled Evaluation

Reproduction, held-out tests, and explicit operating thresholds under distribution shift.

Selected Work

View All

Featured Projects

Coordinate or Concept Mismatch?

Investigating cross-model probe-transfer failures and task-conditioned alignment

Selection Regret: Auditing Deception Probes as Intervention Selectors

Testing whether accurate deception probes can select useful interventions

Measuring Delusion Reinforcement Across the Directness-Need Spectrum

Separating affective support from epistemic sycophancy in conversational models

Cross-Model Generalization of Mechanistic-Interpretability Probes

A reproduction-first audit of activation-probing methods across model families

Replicating Matryoshka Sparse Autoencoders

Replicating feature recovery and comparing Matryoshka and standard sparse autoencoders

Selected Research

Cross-Model Generalization of Mechanistic-Interpretability Probes

Baimam Boukar, Johannes Taraz, Florent Draye, Terry Jingchen Zhang, Zhijing Jin

•Manuscript

Selection Regret: Auditing Deception Probes as Intervention Selectors

Baimam Boukar, Pauline Nyaboe, Allassan A. Nken

•Under Review

Zero-Shot Neural Priors for Generalizable Cross-Subject and Cross-Task EEG Decoding

Baimam Boukar Jean Jacques, Brandone Fonya, Nchofon Tagha Ghogomu, Pauline Nyaboe, Kipngeno Koech

•Preprint