의사를 위한 진료실 챗GPT활용법 - 안양시의사회 (2025.6)

English summary

This lecture introduces ways for clinicians to use large language models in clinical, research, and educational work. It covers how large language models operate, hallucinations, retrieval augmented generation, literature searching, clinical guideline comparison, patient education materials, calibrated explanations, and emerging multimodal and agentic artificial intelligence systems.

Key points

• Large language models are generative statistical models that predict the most probable next words from context and should not be treated as possessing genuine cognitive understanding.
• Hallucination refers to persuasive presentation of factually incorrect information or fabricated references. Retrieval augmented generation attempts to provide verifiable grounding by retrieving and reranking relevant documents from external databases before supplying them to the model.
• OpenScholar, Perplexity academic search, and Consensus are presented as tools for scientific literature retrieval, citation support, or synthesis of study conclusions. Deep research and reasoning models may conduct multistage online searches to draft cited clinical reports. The lecture also demonstrates comparison of uploaded editions of Korean tuberculosis guidelines to review changes in treatment regimens, including linezolid dose adjustments, the addition of pretomanid, and changes to multidrug resistant tuberculosis protocols.] ,

Audience

Medical doctors, clinicians, medical researchers, and advanced medical students

Scope and limitations

This summary is based only on the supplied evidence notes from the lecture and does not independently verify clinical performance, diagnostic accuracy, patient safety, current tool capabilities, or the results of the demonstrations. The tools and examples are limited to the search results and demonstration contexts described in the source. The source explicitly notes that non retrieval based language models may fabricate references, current image models may lack precision for anatomical and cellular diagrams, and deep research prompts may require iterative calibration.

Research context

Within medical artificial intelligence and data science, the material is thematically connected to clinical literature retrieval, evidence grounded language model use, generation of medical education materials, and division of work between humans and artificial intelligence.