20.500.12556/ReVIS-10618 Ocena učinkovitosti velikih jezikovnih modelov na medicinskem testiranju Evaluating the performance of large language models in medical testing Umetna inteligenca je ključno preoblikovala tehnološki napredek, omogočila razvoj novih pristopov za reševanje kompleksnih problemov in izboljšala učinkovitost obstoječih procesov. Med širokim spektrom tehnologij umetne inteligence so veliki jezikovni modeli, kot so GPTji, posebej pomembni zaradi svoje sposobnosti razumevanja in generiranja človeškega jezika. Razvoj od pravilnih pristopov do naprednih algoritmov strojnega učenja kaže na pomembne mejnike v raziskavah umetne inteligence. Modeli, kot sta GPT-3 in GPT-4, so revolucionirali področje z razumevanjem konteksta in generiranjem relevantnega besedila. Predstavitev ChatGPT konec leta 2022 je pomenila prelomni trenutek, ki je omogočil generativnim jezikovnim modelom vodenje smiselnih pogovorov in izvajanje intelektualnih nalog. To je odprlo nove možnosti uporabe umetne inteligence v izobraževanju, programiranju in ustvarjanju vsebine, hkrati pa postavilo nova etična in praktična vprašanja. Magistrsko delo se osredotoča na analizo zmogljivosti GPT modelov – konkretno ChatGPT-3.5, ChatGPT-4, Copilot in Gemini, pri opravljanju medicinskega testa pa USMLE Step 1 2022. S študijo želimo oceniti njihovo sposobnost obdelave in razumevanja medicinskega znanja. Analiza bo izvedena z uporabo statističnega orodja PSPP, ki bo omogočilo učinkovito obdelavo in interpretacijo zbranih podatkov. Artificial Intelligence has fundamentally transformed technological progress, enabled the development of new approaches for solving complex problems, and improved the efficiency of existing processes. Within the broad spectrum of artificial intelligence technologies, large language models, such as GPTs, are particularly important due to their ability to understand and generate human language. The development from rule-based approaches to advanced machine learning algorithms marks significant milestones in artificial intelligence research. Models like GPT-3 and GPT-4 have revolutionized the field with their understanding of context and generation of relevant text. The introduction of ChatGPT at the end of 2022 represented a breakthrough moment, enabling generative language models to conduct meaningful conversations and perform intellectual tasks. This has opened new possibilities for the application of artificial intelligence in education, programming, and content creation, while also raising new ethical and practical questions. The master's thesis focuses on analyzing the performance of GPT models, specifically ChatGPT-3.5, ChatGPT-4, Copilot, and Gemini, in taking the medical USMLE Step 1 test 2022. The study aims to evaluate their ability to process and understand medical knowledge. The analysis will be conducted using the statistical tool PSPP, which will allow for efficient processing and interpretation of the collected data. Umetna inteligenca veliki jezikovni modeli GPT ChatGPT USMLE Step 1 2022. Artificial intelligence large language models GPT ChatGPT USMLE Step 1 2022. true false false Slovenski jezik Angleški jezik Magistrsko delo/naloga 2024-06-27 17:06:27 2024-06-27 17:06:28 2024-08-19 13:09:11 0000-00-00 00:00:00 2024 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 1970-01-01 11245 200767491 MAG_Gostimirovic_Dragan_2024.pdf MAG_Gostimirovic_Dragan_2024.pdf 1 5CB5491CEA33EF5FCE0F4DB82DA96636 21989328e6ccb9461cf6a2d718810bf11c20043f0aaacbdd572fde097ee834f3 d4404506-3496-11ef-b487-001a4af901a5 https://revis.openscience.si/Dokument.php?lang=slv&id=12114 Univerza v Novem mestu - Fakulteta za ekonomijo in informatiko 0 0 0