Fabian Grob
Fabian Grob, looking down

Fabian Grob

I want to know why language models fail.

I'm a PhD student at the ELLIS Institute Tübingen and the MPI for Intelligent Systems, working with Maksym Andriushchenko, and with Yonatan Belinkov as co-supervisor through the ELLIS program. Mostly, I am interested in when and why current models and agentic systems misbehave. I try to find out by using interpretability and hope to make the answers actually useful.

Institute
ELLIS Tübingen · MPI-IS
Programme
ELLIS PhD
Previously
TUM · RWTH · AMD · NYU

Research

Selected work
  1. now

    Interpretability for language model agents

    PhD at the intersection of interpretability and AI safety, starting November 2026.

  2. 2026

    Bigger Is Not Better: Inverse Scaling and Arbitration Failure in Counterfactual Visual Grounding

    With Sanghwan Kim, Cordelia Schmid, Zeynep Akata.
    ICML 2026, Mechanistic Interpretability Workshop

  3. 2025

    Accumulator-Aware Post-Training Quantization for Large Language Models

    With Ian Colbert, Giuseppe Franco, Jinjie Zhang, Rayan Saab.
    TMLR, December 2025 · arXiv 2409.17092

  4. 2023

    Are Emily and Greg Still More Employable than Lakisha and Jamal? Investigating Algorithmic Hiring Bias in the Era of ChatGPT

    With Akshaj Kumar Veldanda, Shailja Thakur, Hammond Pearce, Benjamin Tan, Ramesh Karri, Siddharth Garg.
    NeurIPS 2023, R0-FoMo Workshop · arXiv 2310.05135

Full list and citationsGoogle Scholar ↗

Path

2019 — 2026
  1. Nov 2026

    PhD, AI safety and alignment

    ELLIS Institute Tübingen and MPI for Intelligent Systems · ELLIS PhD Program

  2. 2024 — 2026

    MSc Computer Science, TU Munich

    Konrad Zuse School of Excellence in Reliable AI, 2024 cohort

  3. 2023 — 2024

    ML Engineer Intern, AMD

    Post-training and FP8 quantization, on the open source Brevitas library

  4. 2023

    Research intern, NYU Tandon

    Center for Cybersecurity. Fairness in large language models, and security in code generation

  5. 2020 — 2026

    German Academic Scholarship Foundation

    Full scholarship, awarded to the top one percent of German students

  6. 2019 — 2023

    BSc Computer Science, RWTH Aachen

    Thesis on quantization for neural networks on RRAM crossbars. Erasmus semester at AGH Kraków

Elsewhere

I'm curious about lots of things, passionate about some, and deep in a few. I like things that require precision and build over time.

Outside the lab I run marathons, make coffee more carefully than necessary, and ride up mountains. Happy to talk about any of it.

Barcelona 2025, now working on all majors.
V60, Comandante C40, water 94 °C.
Virage 21, 4.8 %.