Master's Thesis: Attention-Based Hallucination Probability Estimation in Large Language Models at Ericsson

Ericsson is hiring a final-year Master's student to conduct a 30-credit thesis on large language model hallucination detection. The work will reproduce and evaluate attention-based methods such as SinkProbe and Head Entropy across question-answering, mathematical reasoning, and code-generation tasks; develop a unified risk-estimation framework; analyze its performance and failure modes; and deliver a reproducible experimental repository. The thesis is based in Stockholm, Sweden, with a supervisor of Konrad Nowosadko and an expected start in early January 2027.