[pt] Com a crescente popularização das técnicas de Inteligência Artificial, principalmente voltadas ao processamento de linguagem natural, testemunhamos um notável avanço nos Large Language Models (modelos de linguagem avançados), dos quais o Generative Pre-trained Transformer (GPT) consiste no exemplo mais notável. Consequentemente, assistentes virtuais têm conquistado zuma presença significativa em diversas áreas da vida contemporânea. Neste trabalho, é proposta uma metodologia para desenvolver uma assistente virtual inteligente, baseada em um modelo gerador, capaz de compreender a língua portuguesa do Brasil, bem como o domínio específico da Indústria de Óleo e Gás. Essa assistente tem a capacidade de interpretar comandos textuais fornecidos pelos usuários e executar ações correspondentes em um sistema corporativo. Essa metodologia é o resultado de uma cuidadosa análise de diferentes modelos generativos disponíveis, buscando identificar aquele que melhor se adequa aos requisitos da assistente virtual inteligente em português. Para treinamento é criado um dataset representativo com os conceitos necessários e específicos do sistema e da indústria do petróleo. É adotado um processo de refinamento que permite identificar eventuais falhas e aperfeiçoar a compreensão da assistente para garantir respostas precisas e direcionadas. Também são abordados neste trabalho os desafios e limitações inerentes aos modelos generativos, bem como estratégias para superá-las a fim de obter gerações mais precisas e seguras. / [en] With the growing popularity of Artificial Intelligence, specially related to Natural Language Processing, we notice a remarkable development of Large Language Models, which finds in the Generative Pre-Trained Transformers (GPT) their most outstanding example. As a result, virtual assistants have being gaining significant presence in various areas of modern life. In this work, we present the development of an intelligent virtual assistant, based on a generative model. The assistant understands Brazilian Portuguese and is trained on the specific jargon of the Oil and Gas Industry. This assistant has the ability to interpret textual commands provided by users and execute corresponding actions within a corporate system. This methodology is the result of a careful analysis of different available generative models, aiming to identify the one that best suited the requirements of an intelligent virtual assistant in Portuguese. Additionally, it involves the creation of a representative dataset, with concepts specific to the system and the Oil and Gas Industry, to effectively train the assistant. A refinement process allows the identification of potential flaws and the improvement of the assistant s understanding to ensure accurate and targeted responses. Furthermore, this work presents the challenges and the inherent limitations of generative models, and proposes strategies to overcome them in order to achieve more precise and secure generations.

This thesis presents the development and evaluation of context-aware Lexical Simplification (LS) systems for the Swedish language. In total three versions of LS models, LäsBERT, LäsBERT-baseline, and LäsGPT, were created and evaluated on a newly constructed Swedish LS evaluation dataset. The LS systems demonstrated promising potential in aiding audiences with reading difficulties by providing context-aware word replacements. While there were areas for improvement, particularly in complex word identification, the systems showed agreement with human annotators on word replacements. The effects of fine-tuning a BERT model for substitution generation on easy-to-read texts were explored, indicating no significant difference in the number of replacements between fine-tuned and non-fine-tuned versions. Both versions performed similarly in terms of synonymous and simplifying replacements, although the fine-tuned version exhibited slightly reduced performance compared to the baseline model. An important contribution of this thesis is the creation of an evaluation dataset for Lexical Simplification in Swedish. The dataset was automatically collected and manually annotated. Evaluators assessed the quality, coverage, and complexity of the dataset. Results showed that the dataset had high quality and a perceived good coverage. Although the complexity of the complex words was perceived to be low, the dataset provides a valuable resource for evaluating LS systems and advancing research in Swedish Lexical Simplification. Finally, a more transparent and reader-empowering approach to Lexical Simplification isproposed. This new approach embraces the challenges with contextual synonymy and reduces the number of failure points in the conventional LS pipeline, increasing the chancesof developing a fully meaning-preserving LS system. Links to different parts of the project can be found here: The Lexical Simplification dataset: https://github.com/emilgraichen/SwedishLSdataset The lexical simplification algorithm: https://github.com/emilgraichen/SwedishLexicalSimplifier

This thesis has investigated what impact generative AI may have on higher education. Using a combination of a systematic literature study and interviews with representatives from four (4) large universities in Sweden. The findings indicate that generative AI is already a disruptive technology in teaching and learning in higher education, and that students now more easily can cheat or “mislead the examiner” using generative AI, for example by presenting ChatGPT generated text as text written by the students themselves. Even though there are some negatives with generative AI, this thesis shows that the Universities are better off embracing this technology instead of trying to work against it. So, what are the positives with generative AI in education? The fact that students can now converse with someone no matter their background, the fact that students can learn by using ChatGPT (if they are taught how to use it properly), the fact that learning how to use ChatGPT might increase the student’s efficiency and therefore increase their attractiveness on the work market when graduating. All of these benefits come with a big WARNING though. That warning is that higher education must teach the students that these tools are not miracle workers. That the tools can be wrong, and it is important that students learn how to question and criticise what is generated. Higher education has a responsibility to introduce the tools tempered by the understanding that they are not a replacement for knowledge, but only a powerful aid to enhance the knowledge that the students already possess. Finally, the study has been conducted during a particularly expansive period for generative AI and the reader should realise that the findings within this thesis represent early results in a young area of research.

Antalet cyberattacker ökar ständigt och därav också antalet angreppssätt och försvarstekniker. Detta innebär att personer verksamma inom cybersäkerhet behöver spendera mer och mer tid på att hålla sig uppdaterade om de senaste utvecklingarna i branschen. Det är därför av intresse att hitta sätt som kan påskynda denna inhämtning av information. I denna studie utvecklas en prototyp med målet att på ett nytt sätt automatiskt summera en av de många sorters nyhetskällor som finns inom cybersäkerhetsdiskussioner på onlineforum. Prototypen använder sig av abstraktiv textsummering med zero-shot modellen GPT-3. Prototypen som utvecklades utvärderades genom att mäta de summeringar som skapades med SUPERT. Resultatet från mätningen gav ett värde av 0,269 vid mätning mot de originella texterna och 0,358 vid mätning mot ett dataset som städats från text som inte rör cybersäkerhet. Från dessa värdet dras slutsatsen att utvecklingen av prototypen lyckades.

This paper examines the potential impact of Large Language Models (LLMs) in the press and in the production of books and academic journals. LLMs, such as OpenAI’s GPT-3, are trained on massive text corpora and can predict the next word in a given context through probabilistic methods. They have demonstrated autonomy and versatility in a variety of tasks, including question answering, translation, summarization, text classification, and code generation from natural language instructions. The paper discusses the trends, opportunities, and challenges of artificial intelligence (AI) and LLMs in the publishing industries, as well as the existing research on these topics. It also conducts experiments and operations with GPT-3 to explore its potential benefits and limitations, and offers reflections on the medium- and long-term impact of LLMs in those sectors.

Everyday we are flooded with news from all around the world and this information can be overwhelming. In our study we analyze the possibilities to implement GPT-3 models in the work of news summarization in swedish and automize this process. In this study we also regard the ethic point of view, meaning if we can trust these GPT-3 models and  give them the responsibility to make news summarizations. We studied three different GPT-3 models: ChatGPT, Megatron and GPT-SW3. We used a quantitative survey method where the participants got to rate the news summarizations made by the GPT-3 models. The participants got to rate the news summarizations based on the criterias language, contents and structure. We then took the mean value of the ratings from the survey to see the results. The results showed that ChatGPT was significantly the best of all the three GPT-models on all three criterias, and Megatron and GPT-SW3 performed significantly worse. This shows that these models still need some development to get to the same levels as ChatGPT. Despite ChatGPT being the best performing GPT-3 model it still had its weak sides. We noticed this through one article that had alot of factors included which meant alot of information for the GPT-3 models to condense. Through this study we could confirm that GPT-3 models who are further in their development, like ChatGPT can be used in the work of news summarization but should be used with cautioun of what articles it gets to summarize. This means that GPT-3 models still require human supervision for articles with too much information to condense.

This ethnographic study investigates how ChatGPT meets information needs, andthe difference in functional outcomes when using ChatGPT in a reference transaction compared to a librarian. The theoretical starting point is Taylor's theories on information needs and filters, commands and questions. The study is based on empirical material consisting of observations of libraries and of ChatGPT. The empirical material is analyzed using analytical autoethnography, thick description, and Childer's system for evaluating reference transactions. The purpose of the thesis is to investigate how ChatGPT performs in reference transactions compared to librarians. The study shows that ChatGPT meets information needs in a different way than librarians. The biggest difference in functional results is that ChatGPT's responses are incorrect more frequently than the librarians. ChatGPT not only makes dubious claims, it also hallucinates and recommends books that do not exist. This thesis can be used as a basis for further research, as well as support for public libraries that want to use AI technology in reference transactions.

This essay is a study about AI as an assessment aid in arts. The study commences with a literature review about former research about AI and its abilities, with specific focus on assessment and evaluating pictures. The literature review shows both promise and limitations of the use of AI as assessment aid. The theories used in this study are about the process of assessment, the importance of choosing the right assignment and different types of criteria and their benefits and limitations. By performing two semi-structured interviews with two experienced art teachers in southern Sweden an instruction material was created alongside with reference material consisting of student work and how the teachers assessed each drawing. The two teachers assessed each work and an interrater reliability score was calculated. Then an experiment was conducted where AI and one of the teachers assessed 15 drawings. The interrater reliability was again calculated and the process was repeated several times. The result in terms of the interrater reliability was acceptable, however AI assessed the same work with different grades in the attempts. Thus the conclusion is that there still are some limitations of the use of AI as an assessment aid in arts and that further studies with a larger data set ought to be performed.

This study explores the representations of religious concepts generated by Chat-GPT, an AI language model, within the context of educational discourse. Utilizing thematic analysis, the research investigates how Chat-GPT responds to queries related to religion, how it explains and presents religion, examining the alignment of its responses with the Swedish curriculum guidelines. Drawing on theories of social constructivism and educational paradigms, the study elucidates the potential impact of the framing of questions on Chat-GPT’s outputs, particularly in light of implicit biases embedded in educational frameworks. The findings shed light on the intricate interplay between AI technologies and educational context, offering insights into the challenges and opportunities presented by integration of AI tools in pedagogical settings.

Artificial intelligence is one of the long-term trends in the twenty-first century. Historically, the real estate industry has been slow to adopt new technology, but generative AI introduces a range of innovative applications that traditional AI has not addressed. Creating a unique opportunity for the real estate industry to evolve and position itself at the forefront of technological advancements. Despite the promising potential of large language models, research applying the technology on real world problems within real estate sector is almost non-existent. Only a limited number of studies currently exist exploring the area. No applied studies of the technology have yet to be made in Europe to the authors knowledge. The purpose of this study was thus to contribute with an applied study of the technology within the context of facility management. Exploring how generative AI can increase efficiency within facility management by utilizing large language models to analyse tenant matters. Execution consisted of partnering with a real estate company, developing propritary frameworks, technology, and testing these on real world data. A design based researched method was adjusted to fit this study. In total 822 tenant matters where analyzed by a large language model (LLM). The findings show that a large language model can be utilized to analyze tenant matters. Furthermore, that model outputs can be trusted and utilized to improve services for tenants. This study highlights the importance of original data quality, data selection, understanding data inputs and contextualizing instructions for the large language model to achieve successfull automated information extraction. Concluding that analysing tenant matters with generative AI makes it possible to identify and quantify how a real estate company functions, performs, and meets tenants’ needs as a whole —not just from a facility management perspective. / Artificiell intelligens är en av de långsiktiga trenderna under tjugoförsta århundradet. Historiskt har fastighetsbranschen varit långsam med att anamma ny teknik, men generativ AI introducerar en rad innovativa tillämpningar som traditionell AI inte har adresserat. Detta skapar en unik möjlighet för fastighetsbranschen att utvecklas och positionera sig i framkanten av tekniska framsteg. Trots den lovande potentialen hos stora språkmodeller är forskning som tillämpar tekniken, på verkliga problem inom branschen, nästan obefintlig. Endast ett begränsat antal studier existerar för närvarande som utforskar området. Ingen tillämpad studie av tekniken har ännu gjorts i Europa, enligt författarens kännedom. Syftet med denna studie var således att bidra med en tillämpad studie av tekniken inom ramen för fastighetsförvaltning. Utforska hur generativ AI kan öka effektiviteten inom fastighetsförvaltning genom att använda stora språkmodeller för att analysera hyresgäst- ärenden. Genomförandet bestod av att samarbeta med ett fastighetsbolag, utveckling av proprietära ramverk, teknik och testa dessa på verkliga data. En designbaserad forskningsmetod justerades för att passa studien. Totalt analyserades 822 hyresgästärenden av en stor språkmodell (LLM). Resultaten visar att en stor språkmodell kan användas för att analysera hyresgästärenden. Vidare att modellens svar går att lita på och kan användas för att förbättra tjänster mot hyresgäster. Studien framhäver vikten av originaldatakvalitet, val av data, förståelse för datainmatning samt kontextualisering av instruktioner för att den stora språkmodellen ska uppnå framgångsrik automatisk informationsutvinning. Slutsatsen är att AI-analys av hyresgästärenden gör det möjligt att identifiera och kvantifiera hur ett fastighetsbolag som helhet fungerar, presterar och möter hyresgästernas behov—inte bara ur ett fastighetsförvaltningsperspektiv.