Sabrina Sadiekh

sadsobr7@gmail.com

Whoami?

Hi! I am an AI researcher with a background in mathematics, working on the internal structure of language models — how representations form, why they break, and what that reveals about their geometry and robustness.

My research lies at the intersection of Explainable AI and Mechanistic Interpretability and Mathematics; selected papers and my open-source projects are below.

Open-source

Homines dum docent discunt, Seneca the Younger

I have been trying to understand interpretability since 2023. I think the best way to understand something is to try to explain it. I like to show the beauty of the field to others, so a significant part of my work is publicly available:

a bank of open XAI tutorials — hands-on and code-first, maintained since 2023;
the first Russian-language Explainable AI course — 600+ students;
an interactive table of XAI frameworks — a navigation map of the tooling landscape, kept up to date since 2023.

AI Interpretability School

I co-founded the AI Interpretability School together with Elena Ericheva — a course, that teaches classic XAI and mechanistic interpretability as a single subject.

Writing

I write about XAI and interpretability on Telegram, ru in Russian, and recently started writing in English on Substack and Medium.

Collaborations and work

I collaborate with an international group at AikyamLab on mechanistic interpretability and AI safety. I teach M.Sc. courses at HSE University. I also lead R&D Lab at HiveTrace, where I run a proposal-driven lab spanning three tracks — XAI in Security, AI Security, and AI Safety.

AI models are fascinating entities in our world, and if they get out of control, *I want to explain how and why*. This is my main motivation. Feel free to contact me =)

selected publications

AAAI

Polarity-Aware Probing for Quantifying Latent Alignment in Language Models

Sabrina Sadiekh, Elena Ericheva, and Chirag Agarwal

In Proceedings of the AAAI Conference on Artificial Intelligence, 2026

Bib

@inproceedings{sadiekh2026polarity,
  title = {Polarity-Aware Probing for Quantifying Latent Alignment in Language Models},
  author = {Sadiekh, Sabrina and Ericheva, Elena and Agarwal, Chirag},
  booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
  year = {2026},
  address = {Singapore},
}

ACL

Towards Understanding the Robustness of Sparse Autoencoders

Ahson Saiyed, Sabrina Sadiekh, and Chirag Agarwal

In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics, 2026

Bib

@inproceedings{saiyed2026robustness,
  title = {Towards Understanding the Robustness of Sparse Autoencoders},
  author = {Saiyed, Ahson and Sadiekh, Sabrina and Agarwal, Chirag},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics},
  year = {2026},
  address = {San Diego, CA},
}