From ChatGPT to CatGPT: The Implications of Artificial Intelligence on Library Cataloging ARTICLE From ChatGPT to CatGPT The Implications of Artificial Intelligence on Library Cataloging Richard Brzustowicz INFORMATION TECHNOLOGY AND LIBRARIES | SEPTEMBER 2023 https://doi.org/10.5860/ital.v42i3.16295 Richard Brzustowicz (rrbrzustowicz@carlow.edu) is Instruction and Outreach Librarian, Carlow University. © 2023. ABSTRACT This paper explores the potential of language models such as ChatGPT to transform library cataloging. Through experiments with ChatGPT, the author demonstrates its ability to generate accurate MARC records using RDA and other standards such as the Dublin Core Metadata Element Set. These results demonstrate the potential of ChatGPT as a tool for streamlining the record creation process and improving efficiency in library settings. The use of AI-generated records, however, also raises important questions related to intellectual property rights and bias. The paper reviews recent studies on AI in libraries and concludes that further research and development of this innovative technology is necessary to ensure its responsible implementation in the field of library cataloging. INTRODUCTION As librarianship continues to evolve in the digital age, the importance of cataloging as a tool for accessing vast amounts of information cannot be overstated. Unfortunately, this crucial process can be both labor-intensive and time-consuming, often requiring significant resources. In recent years, automation and artificial intelligence (AI) technologies have emerged as potential solutions for streamlining workflows. OpenAI’s language model ChatGPT1 is one such technology, offering the potential to automate various tasks, including text generation and even creating working code.2 This paper explores the potential applications of ChatGPT in library cataloging, examining the results of my own experiments using this innovative technology. LITERATURE REVIEW Large language models (LLMs) applications have been explored in a range of contexts. Taecharungroj explores reactions to ChatGPT, noting wide public interest expressed via social media posts. Due to ChatGPT’s ability to generate accurate information in a conversational tone, it provides an accessible medium for interacting with an AI resource. These technologies will usher in substantial changes to how we do our work: “The next evolution of jobs will likely be impacted by ChatGPT and other innovative AI technologies.”3 At the same time, the author cautions that while ChatGPT performs admirably in generating semantically and syntactically correct information, it does not always provide accurate information. Therefore, this presents an opening for targeted professional development opportunities in fields that may be affected by LLMs. Kasneci et al. argue that AI and language models, if used effectively, are an opportunity for better, more learner-centered education. They caution, however, that LLMs’ potential for bias may necessitate careful training and review of records by professionals: “[I]f a model is trained on data that is biased towards certain groups of people, it may produce results that are unfair or discriminatory towards those groups (e.g., local knowledge about minorities such as small ethnic groups or cultures can fade into the background).”4 To counter the potential for bias and abuse, the authors advise that individuals using these resources should ensure that the data training the mailto:rrbrzustowicz@carlow.edu INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 2 BRZUSTOWICZ AI systems are diverse and inclusive. To that end, they recommend making the data open and available to users. Likewise, the authors recommend regular reviews and continuous updates of AI models. This will ensure not only that the AI (e.g., ChatGPT) generates accurate information, but it will also provide a model for continuous improvement of the machine learning algorithm.5 Recent research has examined the potential of ChatGPT in the library context. Lund and Wang discuss the implications of ChatGPT for libraries and the academy. According to their findings, ChatGPT can enhance library search systems’ accuracy and effectiveness by automatically generating cataloging and metadata for library resources, such as book titles and descriptions.6 However, they also note the potential ethical and legal implications of using ChatGPT. As ChatGPT draws on publicly available material, it may include copyrighted content, necessitating careful evaluation of AI-generated content and coordination with the copyright holder. According to Atlas, the initial impact of AI in higher education is still not fully understood, but it does present an opportunity for professionals to incorporate ChatGPT into their workflow to increase efficiency and streamline tasks. Atlas advises that a well-thought-out plan with clear goals in mind is crucial when utilizing ChatGPT.7 These findings suggest that AI resources such as ChatGPT have the potential to improve library cataloging by automating labor-intensive and time- consuming tasks. Moreover, they offer opportunities for professional development resources to ensure that libraries remain up to date with current technologies. METHODOLOGY To evaluate ChatGPT’s potential for library cataloging, I used the ChatGPT Free Research Preview Version (released February 13, 2023) to generate MARC records in RDA for various items. When asked about its training data, ChatGPT replied: I was trained on a vast amount of data, including examples of MARC and RDA records, as well as the rules and guidelines for creating these records. My training data includes a wide variety of cataloging resources, such as online guides and manuals, training modules, and real-world examples of MARC and RDA records. This training allows me to generate accurate and standardized MARC and RDA records based on the information provided (ChatGPT, personal communication, February 23, 2023). While this response was interesting, I wanted to test ChatGPT’s claims. I asked it to generate records for six items, including one with no corresponding WorldCat entry, and compared the results to records created by human catalogers. Five items had existing MARC records, while the sixth item was specifically chosen to test ChatGPT’s ability to generate an original record. I evaluated ChatGPT’s accuracy, efficiency, and ability to handle different types of materials and cataloging/metadata standards, given that ChatGPT is known to produce both accurate and inaccurate or misleading information.8 To test ChatGPT’s ability to create MARC records, I conducted an initial experiment using readily available materials. I asked ChatGPT to generate a MARC record for the 1996 edition of Anne Rice’s Interview with the Vampire9 using RDA (ChatGPT, personal communication, February 23, 2023). The resulting record is shown in table 1, and I compared it to a record in OCLC’s WorldCat, which is illustrated in table 2. The results of this test indicate that ChatGPT can produce an accurate and effective record for Interview with the Vampire. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 3 BRZUSTOWICZ After this first success, I attempted to generate a MARC record for the 2018 vinyl reissue of David Bowie’s 1977 album Low10 using ChatGPT and the RDA standard (ChatGPT, personal communication, February 23, 2023). The resulting MARC record is presented in table 3, which was then compared to professional catalogers’ records. Table 4 shows an existing MARC record for Low in OCLC’s WorldCat. Notable differences were observed between the human-generated and ChatGPT-generated MARC records, with the ChatGPT record lacking foreign-language headings and subject headings in certain fields (6xx). This is not surprising, as such tasks require a degree of personal discernment on the part of the cataloger. These discrepancies spurred me to investigate the applications further. I refined the question to test ChatGPT’s ability to generate appropriate Library of Congress call numbers. For this example, I requested: “Generate a MARC record using RDA that includes Library of Congress call number for the 1971 German edition of Pedagogy of the Oppressed”11 (ChatGPT, personal communication, February 24, 2023). Tables 5 and 6 demonstrate that while ChatGPT may not always “choose” the same subject access points or consistently format all relevant fields as effectively as a human cataloger, given proper training and oversight it can be used as an effective supplement to human cataloging. The accurate formatting of field 050 and appropriate “dummy” call number (LB875.P442) further demonstrate this technology’s potential for streamlining cataloging and resource description, given proper training. In this instance, the AI noted that multiple fields would need to be edited: “Please note that the control number (001) and the date (005) in the above record are placeholders and should be replaced with actual values when creating the record” (ChatGPT, personal communication, February 24, 2023). To further put ChatGPT’s abilities to the test, I asked it to generate a citation for the 2018 Russian print edition of Cixin Liu’s The Three Body Problem12 (ChatGPT, personal communication, March 2, 2023). This was a more complex request than the previous ones; it required ChatGPT to extract and incorporate metadata from a non-Latin character set (Cyrillic) and in a foreign language. Table 7 shows the MARC record generated by ChatGPT, while table 8 displays the existing MARC record for the Russian translation of this work found in WorldCat. Although there were differences between the two records, ChatGPT’s output was comparable to the professional catalogers’ work. The discrepancies between the records, however, suggested that ChatGPT was not merely reproducing existing records but creating original MARC records, as it claimed. The results of this test further demonstrate ChatGPT’s potential as a powerful tool for automating the generation of accurate metadata records. During my testing, I discovered that the limited vinyl pressing of alternative rock band Mood Rings’ 2013 single “Pathos Y Lagrimas”13 had no WorldCat entry. To see if ChatGPT could generate an original MARC record for this item, I asked it, “Can you generate a MARC record using RDA for Mood Rings’ 2013 single ‘Pathos Y Lagrimas’” (ChatGPT, personal communication, March 8, 2023). Despite the absence of an equivalent WorldCat record, ChatGPT was able to provide a sample MARC record, which I have included in table 9. This record, complete with sample text for the Leader and Control Fields (00X), serves as evidence of two important capabilities of ChatGPT: its ability to generate original cataloging records, and its incorporation of placeholder content in fields that are collection specific. ChatGPT’s ability to generate accurate MARC records using both RDA and ersatz “original” cataloging demonstrates its potential as a cataloging and item description resource. Additionally, ChatGPT’s versatility is further highlighted by its ability to produce original content in other metadata formats. When asked if it could generate records using the Dublin Core Metadata INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 4 BRZUSTOWICZ Element Set, ChatGPT not only confirmed its ability but also provided a sample entry for “Pathos Y Lagrimas” as seen in table 10. While some modifications may be necessary to cater to collection- specific demands, this showcases ChatGPT’s potential as a time-saving tool for automating record generation in multiple formats. In addition to its ability to generate accurate records adhering to multiple metadata standards, the results of this study also highlight the potential versatility of ChatGPT as a cataloging and item description resource. The model’s ability to generate records for different media and in different languages could prove particularly useful for librarians and other information professionals who manage diverse collections. Moreover, while catalogers may need to modify the pregenerated records to suit their specific collections’ requirements, ChatGPT’s user-friendly interface and accurate record generation suggest that it could be a valuable tool for improving cataloging workflows and increasing efficiency. With further development and refinement, ChatGPT has the potential to significantly enhance the capabilities of information professionals and improve the discoverability of library collections. RESULTS This study provides evidence that ChatGPT can generate accurate records that conform to multiple metadata standards. The model can extract essential metadata, including title, author, publisher, publication date, subject headings, and other descriptive elements, with precision. Additionally, my research reveals that ChatGPT’s ability to generate MARC records is not limited to specific formats or languages, as it successfully created MARC records for various media and materials in different languages, such as English, German, and Russian. ChatGPT was able to generate both accurate existing authority records and entirely original ones, and it could generate records using both RDA and Dublin Core standards. According to ChatGPT, it has been trained on data from various catalogs, including OCLC’s WorldCat, the Library of Congress, the National Library of Medicine, the British Library, COPAC (UK Academic and National Library Catalog), Europeana, and the HathiTrust Digital Library (ChatGPT, personal communication, March 9, 2023). This poses a unique challenge, as these catalogs may have different policies on access and reuse of their data. For example, OCLC’s CatExpress is a subscription-based automated cataloging system. If ChatGPT or a future “CatGPT” draws on OCLC’s data and makes it available for free, it may raise questions about OCLC’s copyright holdings. Additionally, while ChatGPT may generate records for materials available on the public internet, such as “Pathos Y Lagrimas,” questions remain regarding how to credit the intellectual labor necessary for creating these records. My comparison of ChatGPT-generated MARC records against manually created records by professional catalogers had positive results. While the accuracy of the ChatGPT-generated records was comparable to those of the manually created records, notable differences existed in how subject access points were assigned. This suggests that ChatGPT has the potential to provide new methods for growing the discipline of library cataloging by automating the more rote, labor- intensive and time-consuming tasks (for example, copy cataloging). In future studies, it may be of interest to the discipline to further test the applications of AI-generated MARC records on a catalog-wide scale. While ChatGPT has the potential to streamline aspects of the cataloging process, it is not a complete replacement for human catalogers. The records generated by ChatGPT can serve as effective starting points, but they often contain discrepancies when compared to professional INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 5 BRZUSTOWICZ catalogers’ records. For example, while the placeholder text in fields 001 and 005 can be useful, it may not match the formatting standards used by specific library collections. Nonetheless, ChatGPT-generated records can be accurate and effective in classifying information that is not specific to any collection, such as call numbers. BIAS While ChatGPT shows promise as a tool for generating MARC and Dublin Core-style records, it is also limited by its training data. At present, ChatGPT searches public records (e.g., WorldCat). As a result, any records it generates will draw on existing professional catalogers’ records. If a record is incomplete or contains bias—even via omission—then ChatGPT will reflect those biases in its output. This will necessitate close monitoring of both original records and those which ChatGPT has created through virtual copy cataloging. ChatGPT’s ability to copy and generate records is rooted in its machine learning-based understanding of cataloging and metadata standards. This AI system uses training data from OCLC’s WorldCat to generate records, which means that the quality of the generated records is dependent on the quality and comprehensiveness of the training data.14 Biases or limitations in the training data can result in biased or incomplete records. For example, if the training data is restricted to certain regions, languages, or publishers, the generated records may not reflect the full diversity of a library’s collections. Similarly, biases in subject headings, descriptors, or other fields in the training data may also manifest in the generated records. While ChatGPT itself has no biases, it is possible for biases to be introduced through the training data, which makes it essential for librarians and other information professionals to curate and update the data regularly. To address these potential biases, information professionals training a large language model should curate the training data carefully and periodically review and update it to ensure it is comprehensive, representative, and unbiased. They may also need to manually review and edit generated records to correct any biases or inaccuracies identified. This approach would provide new opportunities for the profession to highlight diversity, equity, and inclusion in the development and use of AI. While an AI may not have biases, biases of the people involved in training and applying the AI could affect the generated content. Like other machine learning models, ChatGPT acquires its biases from external sources as it can only respond to the data it has been trained on, which may reflect human errors or intentions. Therefore, while ChatGPT could streamline and improve the record generation process, information professionals should approach its use with awareness of its limitations and potential biases. To ensure the accuracy, comprehensiveness, and fairness of the generated records, information professionals should take proactive measures to mitigate any biases and errors . DISCUSSION The results of this study have significant implications for library cataloging. The ability to accurately create descriptive records using ChatGPT could significantly reduce the time and resources required for copy cataloging; this could free up library workers to focus on other important tasks, such as collection development, user services, and metadata management. Moreover, ChatGPT could improve the accuracy and consistency of records in library catalogs. As ChatGPT follows established cataloging rules, records created by the model are less likely to contain errors or inconsistencies; this could lead to improved search and discovery experiences for library users, as well as better interoperability between library catalogs and other systems. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 6 BRZUSTOWICZ The intellectual property concerns surrounding ChatGPT’s ability to generate content are multifaceted. One concern is the potential for copyright infringement, as ChatGPT’s detailed descriptions of original works may be too like the originals, leading to legal issues for those who use the generated content without proper attribution or permission. This concern is particularly heightened for copyrighted works like books or music, where even small portions of the work can be protected. Therefore, it is crucial for ChatGPT’s output to be thoroughly reviewed and vetted before being used in any public-facing materials. Another concern is the possibility of misattribution of authorship. ChatGPT’s use of Dublin Core to describe original works could lead to disputes over ownership and potentially even legal action if it generates a description that attributes authorship to the wrong person or entity. To prevent such conflicts, information professionals should ensure that the metadata generated by ChatGPT accurately reflects the authorship and ownership of the original work. This can be done by reviewing and editing ChatGPT’s output to ensure that the metadata is correct before it is shared publicly. The ownership of the generated content is also a concern, as it is not clear who owns the content created by ChatGPT. As a machine learning model, ChatGPT generates content based on the data it has been trained on, raising questions about the ownership of the content it produces. Establishing clear guidelines for ownership and use of the generated content can help avoid any potential disputes over ownership and ensure that appropriate attribution and permissions are obtained; this is particularly important given the potential commercial value of the content that ChatGPT can produce. Furthermore, it is essential to consider ethical and legal implications of the generated content, such as data privacy and protection, and to ensure that these concerns are addressed when designing guidelines for ownership and use. Finally, there is the potential for unintentional disclosure of sensitive or confidential information. ChatGPT’s ability to generate detailed descriptions of original works may inadvertently disclose unpublished findings or proprietary information, potentially causing harm to the author or institution. To mitigate this risk ChatGPT’s output must be carefully reviewed and edited to ensure that it does not inadvertently disclose sensitive information. Implementing appropriate data security measures and access controls may help prevent unauthorized access to sensitive information. CONCLUSION The study demonstrates that ChatGPT has the potential to significantly streamline the cataloging process in libraries by generating accurate and consistent records for a diverse range of materials. However, it should be used as an auxiliary tool in conjunction with human cataloging efforts to ensure the highest level of accuracy and impartiality. Regular monitoring and evaluation of the model are necessary to detect any potential biases or limitations in the training data. By applying a careful and considered approach to its use, librarians and other information professionals can leverage ChatGPT to enhance the efficiency and effectiveness of cataloging processes, ultimately benefiting library and information center patrons. The accurate and comprehensive MARC records produced by ChatGPT highlight its potential to enhance the effectiveness of library cataloging systems. By extracting metadata information such as author, publisher, subject headings, title, and other descriptive components with high precision, the technology can improve the search and discovery experience for library users. As with any machine learning model, though, there is a risk of bias that needs to be considered when utilizing INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 7 BRZUSTOWICZ ChatGPT. Therefore, it is crucial to monitor and evaluate the training data used to build the model, to ensure that it is extensive, impartial, and representative. The use of Dublin Core to describe original works is another factor that should be taken into consideration when using ChatGPT for cataloging. The model’s potential to generate accurate metadata records is reliant on the quality of the input data, including the use of standardized vocabularies like Dublin Core. The use of any metadata standard, such as Dublin Core, may introduce biases and errors that could affect quality of ChatGPT-generated records. Therefore, information professionals should ensure that the metadata standards used by ChatGPT are appropriate for the materials being cataloged and that the resulting records are regularly reviewed and evaluated for accuracy and impartiality. ChatGPT’s accurate and comprehensive metadata records have the potential to enhance the effectiveness of library cataloging systems and improve the search and discovery experience for library users. However, as with any machine learning model, there is a risk of bias that needs to be considered when utilizing ChatGPT. To ensure error-free results, librarians and other information professionals should approach ChatGPT’s application systematically, by monitoring and evaluating the training data used to develop the model’s capabilities and by regularly curating and updating those data. Additionally, periodic inspection and amendment of the generated records may be necessary to avoid inaccuracies and discrepancies arising from biases in the training data. By taking a careful and deliberate approach to ChatGPT’s use, we can ensure that the resulting metadata records are as accurate, comprehensive, and unbiased as possible, while also being mindful of the intellectual property considerations inherent in using a machine learning model to generate content. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 8 BRZUSTOWICZ TABLES Table 1. ChatGPT-generated MARC record for the 1996 reprint of Interview with the Vampire by Anne Rice Field Data LDR 01110cam a2200325 a 4500 001 2740930 005 20040429112254.0 008 960304s1996 nyu 000 1 eng d 010 96011406 020 0345337662 (pbk.) 040 DLC $c DLC $d DLC 100 1 Rice, Anne. 245 1 0 Interview with the vampire / $c by Anne Rice. 260 New York : $b Ballantine Books, $c 1996. 300 372 p. ; $c 18 cm. 336 text $b txt $2 rdacontent 337 unmediated $b n $2 rdamedia 338 volume $b nc $2 rdacarrier 650 0 Vampires $v Fiction. 650 0 Horror tales. 650 0 Vampires $v Fiction. 650 0 Horror tales. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 9 BRZUSTOWICZ Table 2. WorldCat MARC record for the 1996 reprint of Interview with the Vampire by Anne Rice Field Data LDR c a22 0 4500 001 1052676753 005 20220606044740.3 008 960903r19761996nyua b 001 0 eng 010 $z 75036792 020 9780394498218 (hardcover) ; 0394498216 (hardcover) 040 NTG $b eng $c NTG $d OCLCO $d OCLCF $d OCLCQ $d OCLCO $d OCLCA 049 CRCB 050 4 PS3568.I265 $b I5 1996 082 1 4 [E] 082 0 4 FIC RICE 090 $b 100 1 Rice, Anne, $d 1941-2021. 245 1 0 Interview with the vampire / $c by Anne Rice. 250 Anniversary ed. 260 New York : $b Knopf, $c 1996. 300 340 pages ; $c 22 cm 336 text $b txt $2 rdacontent 337 unmediated $b n $2 rdamedia 338 volume $b nc $2 rdacarrier 500 “Published May 5, 1976 ... reprinted eleven times”--Title... 520 0 In a remote room in a large city, a young reporter sits fa... 600 0 Lestat $c (Vampire), $d 1760- $v Fiction. 650 0 Vampires $v Fiction. 650 0 Imaginary interviews. 650 6 Vampires $v Romans, nouvelles, etc. 650 6 Entretiens imaginaires. 600 0 Lestat $c (Vampire), $d 1760- $2 fast $0 (OCoLC)fst00996665 650 7 Imaginary interviews. $2 fast $0 (OCoLC)fst00967566 650 7 Vampires. $2 fast $0 (OCoLC)fst01163968 655 7 Fiction. $2 fast $0 (OCoLC)fst01423787 655 7 Horror fiction. $2 fast $0 (OCoLC)fst01921684 655 7 Horror fiction. $2 lcgft INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 10 BRZUSTOWICZ Table 3. ChatGPT-generated MARC record for Low by David Bowie (vinyl) Field Description LDR 01523njm a2200373 i 4500 001 ocn1029184931 005 20220130152943.0 007 sd fsngnn|||eu 008 180430s2018 xx umm n eng d 024 1 00602557675302 028 2 LCRHLP14 028 3 LCRHCD14 040 ##$aDLC$beng$cDLC$dOCLCO$dOCoLC 049 ##$aOCOZ 100 1 Bowie, David,$eperformer. 245 10 Low$h[sound recording] /$cDavid Bowie. 264 31 [Place of publication not identified] :$b[Label not identified],$c[2018] 300 ##$a1 sound disc :$banalog, 33 1/3 rpm ;$c12 in. 336 ##$anotated music$bntm$2rdacontent 337 ##$audio$sanc$2rdamedia 338 ##$aaudio disc$bac$2rdacarrier 500 ##$aOriginally released in 1977. 500 ##$aTitle from container. 500 ##$aLyrics and full credits on container insert. 511 0 David Bowie, vocals, saxophones, guitars, piano, and Chamberlin ; Brian Eno, synthesizer, guitar treatments, and piano ; Carlos Alomar, rhythm guitar ; George Murray, bass guitar ; Dennis Davis, drums and percussion. 518 ##$aRecorded August-September 1976 at the Château d’Hérouville, Hérouville, France. 650 _0 $aRock music$y1971-1980. 650 _0 $aRock music$y2011-2020. 650 _0 $aRock music$zEngland. 650 _0 $aVinyl LPs. 700 1 Eno, Brian,$eperformer. 700 1 Alomar, Carlos,$eperformer. 700 1 Murray, George,$eperformer. 700 1 Davis, Dennis,$eperformer. 730 0 Low. 902 ##$aMusic$new. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 11 BRZUSTOWICZ Table 4. WorldCat MARC record for Low by David Bowie (vinyl) Field Data LDR 02213cjm a2200529 i 4500 005 20220317111101.1 008 180321t20182017enkrcna zxx d 040 ## $aSFR $beng $erda $cSFR $dOCLCF $dOCLCA $dOCLCQ $dOCLCO 024 3 $a0190295842918 028 02 $a0190295842918 $bParlophone 049 ## $aCRCB 100 1 $aBowie, David, $ecomposer, performer, producer. 245 10 $aLow / $cDavid Bowie. 264 #1 $a[London] : $bParlophone Records, $c[2018]. 264 #4 $c℗2017 300 ## $a1 audio disc ; $c12 in. 336 ## $aperformed music $bprm $2rdacontent 337 ## $aaudio $bs $2rdamedia 338 ## $aaudio disc $bsd $2rdacarrier 340 ## $avinyl 344 ## $aanalog $2rdatr 344 ## $c33 1/3 rpm 344 ## $dmicrogroove 344 ## $gstereo $2rdacpc 500 ## $aReissue, remastered, originally released in 1977. 508 ## $aProduced by David Bowie and Tony Visconti. 511 0 $aDavid Bowie ; with accompanying musicians. 505 0 $tSpeed of life -- $tBreaking glass -- $tWhat in the world -- $tSound and vision -- $tAlways crashing in the same car -- $tBe my wife -- $tA new career in a new town - - $tWarszawa -- $tArt decade -- $tWeeping wall -- $tSubterraneans. 538 ## $aVinyl disc, double-sided; turntable required. 650 0 $aRock music $y1971-1980. 650 0 $aPopular music $y1971-1980. 650 0 $aProgressive rock music. 650 6 $aRock (Musique) $y1971-1980. 650 6 $aMusique populaire $y1971-1980. 650 6 $aRock progressif. 650 7 $aPopular music. $2fast $0(OCoLC)fst01071422 INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 12 BRZUSTOWICZ Field Data 650 7 $aProgressive rock music. $2fast $0(OCoLC)fst01078743 650 7 $aRock music. $2fast $0(OCoLC)fst01099204 648 7 $a1971-1980 $2fast 655 7 $aPopular music. $2fast $0(OCoLC)fst01726706 655 7 $aProgressive rock (Music) $2fast $0(OCoLC)fst01726709 655 7 $aPopular music. $2lcgft 655 7 $aProgressive rock (Music) $2lcgft 655 7 $aMusique populaire. $2rvmgf 655 7 $aRock progressif. $2rvmgf 700 1 $aVisconti, Tony, $eproducer. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 13 BRZUSTOWICZ Table 5. ChatGPT-generated MARC record for the 1971 German edition of Pedagogy of the Oppressed by Paolo Freire Field Description LDR 00000nam a2200000 4500 001 ocm04045077 003 OCoLC 005 2023-02-23 008 711215s1971 gw 000 0 ger 020 $a9783801200059 020 $a3801200054 035 $a(OCoLC)40053172 040 $aDLC$cDLC$dOCoLC$dOCLCF 041 1 $ager$hpor 042 $apcc 043 $ae-gx--- 050 00 $aLB875.P442 100 1$aFreire, Paulo. 240 10 $aPedagogy of the oppressed.$lGerman 245 10 $aPädagogik der Unterdrückten /$cPaulo Freire ; aus dem Portugiesischen von Walter Jens. 260 $aStuttgart :$bKlett-Cotta Verlag,$c1971. 300 $a218 pages ;$c21 cm 336 $atext$btxt$2rdacontent 337 $aunmediated$bn$2rdamedia 338 $avolume$bnc$2rdacarrier 546 $aIn German. 650 0 $aEducation$xPhilosophy. 650 0 $aPopular education$zBrazil. 650 0 $aSocial change. 651 0 $aBrazil$xSocial conditions. 700 1 $aJens, Walter,$d1923-2013,$etranslator. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 14 BRZUSTOWICZ Table 6. WorldCat MARC record for the 1971 German edition of Pedagogy of the Oppressed by Paolo Freire Field Description LDR cam Mi 001 63439859 005 20220106164408.5 008 830119s1971 gw 000 0 ger d 040 NLGGC $b dut $e fobidrtb $c NLGGC $d HEBIS $d DEBBG $d OCLCQ $d OCLCO 015 72,A06,0123 $2 dnb 016 7 720015707 $2 DE-101 020 3783103746 020 9783783103748 041 1 1 ger $h por 084 5,3 $2 ssgn 084 BU 1410 $2 rvk 084 DD 6455 $2 rvk 084 DT 1000 $2 rvk 090 $b 049 CRCB 100 1 Freire, Paulo, $d 1921- 245 10 Pädagogik der Unterdrückten / $c Paulo Freire ; mit einer Einf. von Ernst Lange ; [autor. Übertr. aus dem Portugesischen]. 260 Stuttgart : $b Kreuz-Verlag, $c 1971. 300 218 p. ; $c 21 cm 650 04 Onderdrukking. 650 04 Pedagogiek. 650 7 Bildungspolitik. $2 fes 650 7 Bildungsziel. $2 fes 650 7 Erziehung. $2 fes 650 7 Pädagogik $2 gnd 650 7 Unterdrückung $2 gnd 650 7 Alphabetisierung $2 gnd 650 7 Politische Bildungsarbeit $2 gnd 650 7 Emanzipatorische Erziehung $2 gnd 651 7 Brasilien $2 gnd 650 17 Pedagogiek. $2 gtt INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 15 BRZUSTOWICZ Field Description 650 17 Onderdrukking. $2 gtt 655 4 Unterdrückte. 655 7 Unterdrückte. $2 swd 029 0 NLGGC $b 830230688 029 1 DEBBG $b BV005751307 029 1 DEBSZ $b 003425525 029 1 HEBIS $b 045711100 INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 16 BRZUSTOWICZ Table 7. ChatGPT-generated for the 2018 print edition in Russian of The Three Body Problem by Cixin Liu Field Data LDR cam i 001 1037948853 005 20220402070218.7 008 180529s2018 ru 000 1 rus d 040 CLE $b eng $e rda $c CLE $d OCLCF $d OCL $d OCLCQ $d OCLCO 066 $c (N 020 9785040891122 020 5040891121 041 1 $h chi $a rus 043 a-cc--- 090 $b 049 CRCB 100 1 $a Liu, Cixin, $e author. 240 10 $a San ti. $l Russian 245 10 $a Задача трех тел / $c Лю Цысинь. 245 10 $a Zadacha trekh tel / $c Li︠u ︡ T︠S ︡ysinʹ. 264 1 $a Москва : $b Э, $c 2018. 264 1 $a Moskva : $b Ė, $c 2018. 300 $a 462 pages ; $c 22 cm 336 $a text $b txt $2 rdacontent 337 $a unmediated $b n $2 rdamedia 338 $a volume $b nc $2 rdacarrier 490 0 $a Sci-Fi Universe 650 0 Human-alien encounters, fiction 650 0 Imaginary wars and battles, fiction 651 0 China -- History -- Cultural Revolution, 1966-1976 -- Fiction 650 6 Rencontres avec les extraterrestres, romans, nouvelles, etc. 650 6 Guerres et batailles imaginaires, romans, nouvelles, etc. 651 6 Chine -- Histoire -- 1966-1976 (Révolution culturelle) -- Romans, nouvelles, etc. 650 7 Human-alien encounters. $2 fast $0 (OCoLC)fst00963475 650 7 Imaginary wars and battles. $2 fast $0 (OCoLC)fst00967580 651 7 China. $2 fast $0 (OCoLC)fst01206073 647 7 Cultural Revolution (China : 1966-1976) $2 fast $0 (OCoLC)fst01352662 INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 17 BRZUSTOWICZ Field Data 648 7 1966-1976 $2 fast 655 7 Fiction. $2 fast $0 (OCoLC)fst01423787 655 7 History. $2 fast $0 (OCoLC)fst01411628 655 7 Science fiction. $2 fast $0 (OCoLC)fst01726489 655 7 Science fiction. $2 lcgft 700 1 Translation of: Liu, Cixin. San ti. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 18 BRZUSTOWICZ Table 8. WorldCat record for the 2018 print edition in Russian of The Three Body Problem by Cixin Liu Field Data LDR cam i 001 1037948853 005 20220402070218.7 008 180529s2018 ru 000 1 rus d 040 CLE $b eng $e rda $c CLE $d OCLCF $d OCL $d OCLCQ $d OCLCO 066 $c (N 020 9785040891122 020 5040891121 041 1 $h chi $a rus 043 a-cc--- 090 $b 049 CRCB 100 1 $a Liu, Cixin, $e author. 240 10 $a San ti. $l Russian 245 10 $a Задача трех тел / $c Лю Цысинь. 245 10 $a Zadacha trekh tel / $c Li︠u ︡ T︠S ︡ysinʹ. 264 1 $a Москва : $b Э, $c 2018. 264 1 $a Moskva : $b Ė, $c 2018. 300 $a 462 pages ; $c 22 cm 336 $a text $b txt $2 rdacontent 337 $a unmediated $b n $2 rdamedia 338 $a volume $b nc $2 rdacarrier 490 0 $a Sci-Fi Universe 650 0 Human-alien encounters, fiction 650 0 Imaginary wars and battles, fiction 651 0 China -- History -- Cultural Revolution, 1966-1976 -- Fiction 650 6 Rencontres avec les extraterrestres, romans, nouvelles, etc. 650 6 Guerres et batailles imaginaires, romans, nouvelles, etc. 651 6 Chine -- Histoire -- 1966-1976 (Révolution culturelle) -- Romans, nouvelles, etc. 650 7 Human-alien encounters. $2 fast $0 (OCoLC)fst00963475 650 7 Imaginary wars and battles. $2 fast $0 (OCoLC)fst00967580 651 7 China. $2 fast $0 (OCoLC)fst01206073 647 7 Cultural Revolution (China : 1966-1976) $2 fast $0 (OCoLC)fst01352662 INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 19 BRZUSTOWICZ Field Data 648 7 1966-1976 $2 fast 655 7 Fiction. $2 fast $0 (OCoLC)fst01423787 655 7 History. $2 fast $0 (OCoLC)fst01411628 655 7 Science fiction. $2 fast $0 (OCoLC)fst01726489 655 7 Science fiction. $2 lcgft 700 1 Translation of: Liu, Cixin. San ti. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 20 BRZUSTOWICZ Table 9. ChatGPT’s original MARC record for “Pathos Y Lagrimas” by Mood Rings Field Data LDR 02461cjm a2200517 a 4500 001 123456789 005 20181025171247.0 007 sd fungnnmmned 008 181025s2013 nyurcn n zxx d 024 8 767870622805 035 (OCoLC)851415773 040 DLC $b eng $c DLC $e rda $d DLC 041 0 $g spa 043 n-mx--- 047 vd $a pp 048 ka01 049 DLC 050 00 M1630.18 $b .M66 2013 082 00 782.421660972 100 1 Mood Rings $4 prf 245 10 Pathos Y Lagrimas / $c Mood Rings. 246 3 333 264 1 [Brooklyn, NY] : $b Mexican Summer, $c [2013] 300 1 audio disc : $b analog, 33 1/3 rpm ; $c 7 in. 336 performed music $b prm $2 rdacontent 337 audio $b s $2 rdamedia 338 audio disc $b sd $2 rdacarrier 500 Title from label. 500 B-side label has number “333.” 511 0 Mood Rings. 500 “Limited edition” “500 copies.” 650 0 Rock music $z Mexico $y 2011-2020. 655 7 Rock music. $2 lcgft 655 7 Popular music. $2 lcgft 700 1 Mood Rings. $4 prf 710 2 Mexican Summer (Firm) $4 pbl. INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 21 BRZUSTOWICZ Table 10. ChatGPT’s original record using the Dublin Core for “Pathos Y Lagrimas” by Mood Rings Field Data Title Pathos Y Lagrimas / Mood Rings Creator Mood Rings Contributor Mexican Summer Date 2013 Type Sound Format audio/vinyl Identifier 333 Language eng Relation Mexican Summer Source Mexican Summer Coverage Atlanta, GA Rights All rights reserved INFORMATION TECHNOLOGY AND LIBRARIES SEPTEMBER 2023 FROM CHATGPT TO CATGPT 22 BRZUSTOWICZ ENDNOTES 1 Viriya Taecharungroj, “What Can ChatGPT Do? Analyzing Early Reactions to the Innovative AI Chatbot on Twitter,” Big Data and Cognitive Computing 7, no. 1: 35, https://doi.org/10.3390/bdcc7010035. GPT (generative pretrained transformer) technologies are structured in a way that involves training language models on a large body of data. Subsequently, fine-tuning processes are applied to enhance the model’s performance on specific tasks and domains. Fine-tuning processes are then applied to enhance the model’s performance on specific tasks and domains. 2 Roberto Gozalo-Brizuela and Eduardo C. Garrido-Merchan, “ChatGPT Is Not All You Need. A State of the Art Review of Large Generative AI Models,” arXiv:2301.04655v1 [cs.LG]: 15, https://doi.org/10.48550/arXiv.2301.04655. 3 Taecharungroj, “What Can ChatGPT Do?” 4 Enkelejda Kasneci et al., “ChatGPT for Good? On Opportunities and Challenges of Large Language Models for Education,” EdArXiv (January 30, 2023), https://doi.org/10.35542/osf.io/5er8f. 5 Kasneci et al., “ChatGPT for Good?” 6 Brady D. Lund and Ting Wang, “Chatting about ChatGPT: How May AI and GPT Impact Academia and Libraries?” Library Hi Tech News 40 (2023), no. 3: 26–29, https://doi.org/10.1108/LHTN- 01-2023-0009. 7 Stephen Atlas, “ChatGPT for Higher Education and Professional Development: A Guide to Conversational AI,” (2023): 106–7, https://digitalcommons.uri.edu/cba_facpubs/548/. 8 Ali Borji, “A Categorical Archive of ChatGPT Failures,” arXiv.2302.03494 [cs.CL]: 11, https://doi.org/10.48550/arXiv.2302.03494. 9 Anne Rice, Interview with the Vampire (New York: Alfred K. Knopf, 1996). 10 David Bowie, Low, recorded September–October 1976, RCA Victor, 1977, vinyl LP. 11 Paolo Freire, Pädagogik Der Unterdrückten (Stuttgart: Kreuz-Verlag, 1971). 12 Cixin Liu, Задача Трех Тел (Moscow: Sci-Fi Universe, 2018). 13 Mood Rings, “Pathos Y Lagrimas,” recorded ca. 2013, Mexican Summer, 2013, vinyl single. 14 Tom B. Brown et al., “Language Models Are Few-Shot Learners,” arXiv.2005.14165v4 [cs.CL]: 8–9, https://doi.org/10.48550/arXiv.2005.14165. https://doi.org/10.3390/bdcc7010035 https://doi.org/10.48550/arXiv.2301.04655 https://doi.org/10.48550/arXiv.2301.04655 https://doi.org/10.35542/osf.io/5er8f https://doi.org/10.1108/LHTN-01-2023-0009 https://doi.org/10.1108/LHTN-01-2023-0009 https://digitalcommons.uri.edu/cba_facpubs/548/ https://doi.org/10.48550/arXiv.2302.03494 https://doi.org/10.48550/arXiv.2005.14165 Abstract Introduction Literature Review Methodology Results Bias Discussion Conclusion Tables Endnotes