How Large Language Models Are Redefining the Nature of Library Search
Introduction
Historically, it has been difficult for users to search for the data they want in libraries. In the era of traditional search, people could only find the information they wanted through physical directories and indexes. Users still need skills like knowing how to select the right terms and using Boolean logic (such as AND, OR, and NOT) to make their searches more exact. The system works in basic ways, but there is a big difference between the way people search and the way the system works. However, there is a new technology that can change this. This is the Large Language Model, also known as LLM. ChatGPT is a well-known example of an LLM. It has quickly become popular with students and researchers for different tasks (Chigwada and Pasipamire, 2024). LLMs can understand and create text that sounds a lot like the language we use. This will change how we access and use information.
This paper will argue that Large Language Models (LLMs) represent a new and innovative tool with the potential to redefine the nature of traditional library retrieval. The essay will explain how LLMs can generate and synthesize information, creating new ways to find knowledge. It will also analyze the challenges and ethical problems of LLM technology, such as spreading misinformation and bias. The paper will emphasize the necessity of improving AI literacy and conclude that LLMs make the role of librarians more important. Librarians must educate users on the safe, critical and effective use of these technologies.
Problems with Traditional Library Search
When conducting data searches in libraries and online libraries, users must use ‘keywords’ to find the information they seek. Kroll et al. (2024) proposed the basic principle of search, which is that when people use libraries’ online public access catalogs (OPAC), Scopus and other databases, or even general academic search engines such as Google Scholar, keywords will match the content index. While this system of information retrieval is standard, it does not actually align with how humans typically think.
A primary issue is the high level of technical skill it demands from the user. This process demands that users have at least some familiarities with the field of content they are searching for. For example, the commonly used term ‘heart attack’ might not match the article ‘myocardial infarction’. Carroll and Borycz (2024) pointed out that in most cases, to solve this problem, users must not only spend time learning relevant terminology but also master advanced search techniques, including Boolean logic, truncation, and phrase searches. They also state that these skills are not easy for most people to understand and are an important part of the information literacy education that librarians often provide. This requirement makes it hard for people to access information, because the quality of the research results depends a lot on how good the person searching is at using technology.
The second major issue is the lack of a unified data center. Scholarly information is not all kept in the same place. Instead, it is spread across many different databases, catalogues and digital archives. Kroll et al. (2024) provided a search framework in their article: researchers conducting Schopenhauer’s philosophical research may first search physical library catalogs and then switch to professional databases such as PhilPapers to find relevant articles, or they may use Google Scholar to search for broader information. In some cases, they may need to consult specialized archives to obtain historical documents. Each platform may have its own unique search logic, and some platforms’ search systems may be very ineffective. This means that the researcher has to learn and adapt to lots of different systems, doing the same or similar searches again and again. As Kroll et al. (2024) explain, this process can be very tiring and take up a lot of time. This time could be spent on research, but instead it is used for dealing with the many issues involved in adapting to the requirements of different platforms.
A third limitation of traditional search systems is that they can only use a small number of languages. Most academic databases are primarily in English. As Kumar, Yadav, and Sinha (2024) show that there are a lot of non-native English speakers in India, and building a search environment for them is a big challenge. It separates information by language, which makes it hard to do truly international research. Some valuable information may not be written in English, but traditional systems rely on exact matching for retrieval, which fundamentally fails to understand what users truly seek (Chigwada and Pasipamire, 2024). These three main issues will gradually ease with the arrival of LLMs.
How LLMs Change Searching
LLM introduced a fundamentally new approach to the task of searching for information. Traditional search engines require users to think like computers. However, LLMs allow users to search using language that is intuitive to humans. This makes it easier for people to find and organize information. According to Chigwada and Pasipamire (2024) discuss that students in Library and Information Science (LIS) courses have started using tools such as ChatGPT to “ask questions” and receive “explanations” without the need to develop a series of search plans. The emergence of LLMs has enabled search to truly understand information, thereby addressing the limitations of keyword-based search.
Unlike traditional search tools, which give users a list of relevant documents, LLMs can process information from their vast training data and provide a direct response, summarize a topic, break down a complex idea or compare and contrast concepts. Carroll and Borycz (2024) provide an example where students say they use LLM to ‘summarize longer papers and condense information’. They also mentioned that LLM can be used to brainstorm topics, ideas, and even help create structured “keyword lists” for databases. Other studies also show that many students use ChatGPT to get background information on topics, to understand what text means, to write paragraphs and to start research projects (Chaudhuri and Terrones, 2024). LLMs are more than a passive index; they actively participate in the research process.
Additionally, LLMs excel at identifying connections between different information fragments. Unlike traditional libraries, LLMs are trained using a vast array of materials, and the training process itself involves seeking out connections. Kroll et al. (2024) pointed out that this means they can answer complex problems that are difficult to solve using only keywords. Users can inquire about the connection between two seemingly unrelated fields, and LLM can generate responses pointing to related topics and literature. They also mentioned that LLM can decompose complex languages used in specific fields and explain difficult concepts in simple and understandable language. This is very useful for research that combines different subjects and for helping people who are not experts enter new research areas.
Finally, LLMs are able to overcome language barriers, a problem that traditional systems are almost unable to solve. Users can submit queries in their native language and request the LLM to search for information in other languages. Kumar, Yadav and Sinha (2024) note that LLM can summarize the results in the user’s native language with instant translation, allowing users to quickly assess the practicality of non-native language materials for their research. This represents a significant step toward breaking down information silos and integrating global academic resources. They have come up with a plan to use LLMs to deal with the tricky nature of language in library services in certain countries, like India. Based on the many advantages of LLMs, Chigwada and Pasipamire (2024) stated that many information science experts believe that LLMs are important for the future of library services.
Big Problems and the Need for AI Literacy
LLMs have a lot of benefits that many people are excited about. However, they also have some big problems that can disrupt academic and research work. Users can get incorrect or even dangerous information if they use these tools without thinking. Burtsev et al. (2024) believe that overestimating the capacity of the LLM can lead to ‘unreliable applications’. According to Chigwada and Pasipamire (2024), students and information professionals state that the most significant challenges are the quantity of false information, the inclusion of biases, and complex ethical issues related to academic integrity. Therefore, just using LLM as a new technology in libraries is not enough. It is a challenge for teaching and learning, which means we need to develop a new way of thinking about critical skills. This is often referred to as “AI literacy” (Lo, 2024).
LLMs have a problem where they “hallucinate”. Burtsev et al. (2024) emphasize that the so-called “hallucination” is about models writing text that’s easy to read, seems real, and is well organized, but which is partly or completely made up. They mentioned that LLMs can make up facts, quote text that doesn’t exist, and create complex false arguments. This is because their data is not extracted from a database but rather the model itself is a text generator that’s goal is to predict the next possible word based on previous information. Chigwada and Pasipamire (2024) found in their study of Library and Information Science (LIS) students that they are concerned about the possibility of obtaining biased information leading to the spread of misinformation. They also stated that it is important to ‘verify the information generated by LLM from reliable sources’. Therefore, when using LLM, checking the facts is essential.
An associated challenge is the limitations and biases of the LLM knowledge base. Their ability to generate limitations and biases depends on the data they receive for training, which may have weaknesses. Carroll and Borycz (2024) mentioned that the data may be outdated, which means LLM may lack understanding of the latest developments. This is a serious flaw for researchers. In addition, Lo (2024) emphasizes that these data are often taken from the Internet and therefore contain biases, which LLM can therefore replicate or even reinforce in their responses. The root cause is that the process is a ‘black box,’ similar to the ‘black box’ in aviation, where aircraft manufacturing and flight data are kept hidden. The fundamental reason is that this process is a ‘black box’, similar to the ‘black box’ of the aviation industry - aircraft manufacturing and flight data are hidden. Lo (2024) illustrates that many commercial LLMs are the same: users do not know how they are made or how they work because companies do not share these details. This means that it is impossible to check or see any biases in the source. These biases can lead to serious academic honesty and responsibility risks.
These challenges demonstrate the importance of librarians to today’s world. Lo (2024) pointed out that it is necessary to change the teaching methods of information literacy; Merely teaching users how to use search tools is not enough, libraries need to help people engage in more critical thinking, especially when understanding the complex and often misleading information generated by AI. He also emphasized that librarians must master how to use artificial intelligence to help users critically analyze their outputs and understand ethical issues related to LLM. It is important to teach students how to concretely analyze facts, identify potential biases, and distinguish between text generated based on statistical data and validated information. Carroll and Borycz (2024) proposed the “5I” framework (Incomplete, Inconsistent, Incoherent, Illogical, and Indulgent) which effectively encapsulates the limitations of students’ own perceptions and provides a practical way of determining whether or not data is usable. This effectively helped the librarian to identify a specific framework. Overall, librarians will play an increasingly important role in future library searches.
Conclusion
In conclusion, it is clear that Large Language Models will be a breakthrough development in the field of information retrieval. Unlike traditional systems, big language models can understand, translate, process, and synthesize natural language, providing a more human like user experience. Many examples in the article confirm that people have high expectations for this new technology, as they no longer need to learn complex search commands.
However, technological breakthroughs are not a simple process. LLM has also brought new challenges. These challenges include the generation of illusions and false information, limitations, and biases. People need to spend more energy dealing with the many problems brought by LLM. Therefore, the rise of LLM does not necessarily mean the decline of libraries or librarians. On the contrary, the arrival of LLM will give librarians more missions.
The new task of the library is to lead the development and teaching of AI literacy. The core role of librarians is changing, from primarily being proficient in various advanced searches to being able to proficiently distinguish the availability of AI information. As Lederman and Mahowald (2024) argue, LLMs currently lack “intention”, which prevents them from becoming librarians. This also resulted in LLMs being unable to replace the responsibilities of librarians for a long time. Undoubtedly, LLMs will become the next generation of tools, but they will live forever in the next generation until many problems are solved.
References
Burtsev, M., Reeves, M. and Job, A. (2024) ‘The Working Limitations of Large Language Models’, MIT Sloan Management Review, Winter, pp. 8-10.
Carroll, A.J. and Borycz, J. (2024) ‘Integrating large language models and generative artificial intelligence tools into information literacy instruction’, The Journal of Academic Librarianship, 50(102899), pp. 1-10.
Chaudhuri, J., and Terrones, L. (2025). Reshaping academic library information literacy programs in the advent of ChatGPT and other generative AI technologies. Internet Reference Services Quarterly, 29(1), 1-25.
Chigwada, J. and Pasipamire, N. (2024) ‘Perception and Use of Large Language Models by Library and Information Science Students’, International Journal of Librarianship, 9(3), pp. 75-89.
Kroll, H., Kreutz, C.K., Jehn, M. and Risse, T. (2024) Requirements for a Digital Library System: A Case Study in Digital Humanities (Technical Report). Available at: arXiv:2410.22358v2 [cs.DL].
Kumar, M., Yadav, A. K., and Sinha, A. Harnessing Large Language Models for Enhanced Library Services: A Strategic Frame Work for Indian Libraries.
Lederman, H. and Mahowald, K. (2024) Are Language Models More Like Libraries or Like Librarians? Bibliotechnism, the Novel Reference Problem, and the Attitudes of LLMs. Available at: arXiv:2401.04854v3 [cs.CL].
Lo, L.S. (2024) ‘Evaluating AI Literacy in Academic Libraries: A Survey Study with a Focus on U.S. Employees’, College & Research Libraries, July, pp. 635-668.