Baris Karacan

Fine-tuning a 70B model not only boosts performance in clinical provenance categorization but also shows that quantization can enhance efficiency without sacrificing accuracy.

Teacher narratives reveal unique ADHD signals that structured assessments often miss, highlighting a critical gap in traditional diagnostic methods.

Zero-shot LLMs can segment clinical notes in new domains better than fine-tuned models, but only after you fix their header hallucinations.