Subject. The article considers LLM models and their key elements in the context of economic research. Objectives. The aim is to develop a tuple model that formalizes the main components of LLM models, and a methodological approach that defines the essence of the use of LLM models based on GPT-4o in the interpretation of economic texts. Methods. Based on the analysis of literature on functional and structural interactions of LLM components, a tuple model was developed: LLM = (Data Preparation, Data Processing, Text Generation), enabling to present the mechanism of operation of artificial intelligence algorithms in text generation. Results. The findings show that using the tuple model helps understand how the key elements of LLM models interact with each other, which hyperparameters of the model allow influencing its creativity in forming answers to questions. The paper offers a methodological approach and proves the effectiveness of its application on pretrained models based on GPT-4o for "temperature" hyperparameter regulation. Conclusions. The developed approach and the tuple representation of LLM models can be effectively used for further study, including in the framework of economic processes. The results of the analysis can contribute to the creation of strategies aimed at optimizing the use of LLM models in various sectors of the economy, and improving decision-making based on data analysis.
Keywords: economics, tuple model, data analysis, artificial intelligence
References:
Bubeck S., Chandrasekaran V., Eldan R. et al. Sparks of Artificial General Intelligence: Early experiments with GPT-4. arXiv, 2023, abs/2303.12712. URL: Link
Firdaus M.F., Wibawa J.N., Rahman F.F. Utilization of GPT-4 to Improve Education Quality Through Personalized Learning for Generation Z in Indonesia. IT for Society, 2023, vol. 8, no. 1. URL: Link
Nazarov D.M., Begicheva S.V. [Application of large language models in educational process]. Biznes. Obrazovanie. Pravo = Business. Education. Law, 2024, no. 3, pp. 430–436. (In Russ.) URL: Link
Yang Wu, Shilong Wang, Hao Yang et al. An Early Evaluation of GPT-4V(ision). arXiv, 2023, abs/2310.16534. URL: Link
Ermilov D.I. [Chat GPT: A new technology for communicating with clients in the banking sector]. Finansovye rynki i banki = Financial Markets and Banks, 2023, no. 7, pp. 18–22. URL: Link (In Russ.)
Yizhen Zheng, Huan Yee Koh, Jiaxin Ju et al. Large Language Models for Scientific Synthesis, Inference and Explanation. arXiv, 2023, abs/2310.07984. URL: Link
Burnell R., Hao H., Conway A., Orallo J. Revealing the structure of language model capabilities. arXiv, 2023, abs/2306.10062. URL: Link
Janik R. Aspects of human memory and Large Language Models. arXiv, 2023, abs/2311.03839. URL: Link
Manning C.D., Clark K., Hewitt J. et al. Emergent linguistic structure in artificial neural networks trained by self-supervision. Proceedings of the National Academy of Sciences, 2020, vol. 117, iss. 48. URL: Link
Brown T.B., Mann B., Ryder N. et al. Language models are few-shot learners. arXiv, 2020, abs/2005.14165. URL: Link
Singhal K., Azizi S., Tao Tu et al. Large language models encode clinical knowledge. Nature, 2023, vol. 620, pp. 172–180. URL: Link
Cheng Kunming, Sun Zaijie, He Yongbin et al. The potential impact of ChatGPT/GPT-4 on surgery: Will it topple the profession of surgeons? International Journal of Surgery, 2023, vol. 109, iss. 5, pp. 1545–1547. URL: 10.1097/JS9.0000000000000388
Jojic A., Zhen Wang, Jojic N. GPT is becoming a Turing machine: Here are some ways to program it. arXiv, 2023, abs/2303.14310. URL: Link
Oppenlaender J., Linder R., Silvennoinen J. Prompting AI Art: An Investigation into the Creative Skill of Prompt Engineering. arXiv, 2023, abs/2303.13534. URL: Link