132 KB21 KB195720252024Rivo, Ksenijaštevilka:2/3letnik:68str. 165-173DOI:10.55741/knj.68.2-3.9ISSN:1581-7903COBISSID:210992131URN:URN:NBN:SI:doc-L6POM13Msl= University of Ljubljana PressKnjižnicadatabaseslarge language modelsodprti dostopopen accessstrojno učenjeUmetna inteligencaveliki jezikovni modeliZbirke podatkovCORE-GPT| združitev odprto dostopnih člankov in umetne inteligence za zaupanja vredne odgovore| CORE-GPT| combining open access articles and artificial intelligence for reliable answers|Purpose: In this paper, we present CORE-GPT, an innovative service of the COnnecting REpositories (CORE) Open Access Collection that combines the Generative Pre-trained Transformers (GPT) model with the extensive collection of open access scientific docu-ments from the CORE Open Access Collection. CORE-GPT provides citations of the sci-entific papers underlying the answers, links to PDF formats of the papers and thus improves the reliability and credibility of the answers.Methodology/approach: This paper gives an insight into the performance of the GPT models that played an important role in the creation of the CORE-GPT service. We also present some results from a recent study on the reliability of the CORE-GPT service and reflect on the role of libraries in the use of open access services.Results: CORE-GPT represents an important step forward in improving the search experience and the reliability of answers. By combining the GPT with the rich CORE database, it enables the provision of comprehensive and evidence-based answers to questions.Originality/practical implications: This article is one of the first presentations of the CORE-GPT service in Slovenia.Keywords: open access, artificial intelligence, Generative Pre-trained Transformers, GPT, large language models, LLM, databasesData set Metadata: No research data are associated with this articleNamen: V članku predstavljamo inovativno storitev odprtodostopne zbirke COnnecting REpositories (CORE), CORE-GPT, ki združuje model generativnih predhodno usposob-ljenih transformatorjev (angl. Generative Pre-trained Transformers - GPT) z obsežnim naborom odprto dostopnih znanstvenih dokumentov iz zbirke CORE. CORE-GPT nava-ja znanstvene dokumente, ki so podlaga za odgovore, ter povezave do člankov v forma-tu PDF in tako zagotavlja zanesljivost in verodostojnost odgovorov. Metodologija/pristop: Članek podaja vpogled v delovanje modelov generativnih pred-hodno usposobljenih transformatorjev, ki so imeli pomembno vlogo pri vzpostavitvi storitve CORE-GPT. Predstavili smo tudi izsledke nedavne raziskave o zanesljivosti delovanja CORE-GPT in podali razmislek o vlogi (visokošolskih) knjižnic pri uporabi odprto dostopnih storitev.Rezultati: CORE-GPT predstavlja pomemben korak naprej k izboljšanju iskalne izku-šnje in zanesljivosti odgovorov. Z združevanjem orodja generativnih predhodno uspo-sobljenih transformatorjev in bogate zbirke CORE zagotavlja izčrpne in na dokazih temelječe odgovore na vprašanja. Izvirnost/uporabnost raziskave: Članek podaja eno prvih predstavitev storitve CORE-GPT v SlovenijiTEXTznanstveno časopisjejournalsSlovenian National E-content AggregatorNational and University Library of SloveniaZveza bibliotekarskih društev Slovenije