The UniBuc team's submission to SemEval 2024 Task 2 tackles Safe Biomedical Natural Language Inference for Clinical Trials. The task involves reasoning over clinical trial text to assess safety-related implications, a setting where domain precision matters.
Their approach relies on SOLAR Instruct, a large language model, used entirely without fine-tuning. Instead, the team concentrated on input design, crafting tailored prompts to guide the model's inference. The paper's title points to this prompt-centric strategy as the core of their method.
Because the abstract is truncated, details on the exact prompt templates and evaluation results are not available in the source. Still, the work illustrates a lightweight alternative to fine-tuning for specialized biomedical reasoning, one that leverages prompt engineering rather than weight updates.