Yiderigun Borjigin

I am a doctoral researcher at Saarland University, supervised by Prof. Dr. Roland Aydin. I study the reliability and alignment of large language models through behavioral evaluation and mechanistic interpretability.

My work asks how context shapes a model’s answers: when it follows relevant evidence, and when it is swayed by a suggested number, a confident but incorrect explanation, or the way data is presented. I build controlled evaluations to study these effects and use interventions on model activations to investigate their mechanisms. My work also includes pretraining-time alignment and the reliability of language models in scientific applications.