Abstract: LLM (large language model) families have a big powerful as one of many AI models that has the potential to revolutionize research and scientific, including general purposes. That is, current LLMs assist not only general purposes but also domains such as scientific research and engineering, the prediction of economic trend, health diagnosis, and policy decision. It also enables graphic works, data analysis, coding workface. However, unfortunately, the characteristics and application possibilities of the several structure of LLMs did not work through study on application areas. Current LLMs as well as GAI (General AI) models respond to question of user by re-organization these data after learn vast data (Text, Number, Image). That is, inference algorithms of current LLMs do not have the cause on why this task happen. To recover this problem, causal inference is quite important as core technology. The causal inference by structures of LLM such as standard LLM, offline causal RL with backdoor adjustment (Confounder control), offline causal model with explicit confounder, and backdoor adjustment intervention must fully study through step by step of case because these responses is quite different from structures. This paper offers how to apply and what structures of LLMs is useful for user case, depending structure of LLM causal inference. The learning capabilities of LLM is quite different from causal inference model of LLM on application. To research on purpose, what is the core approaches by causal inference model, what is the tuned parameter and structure, and how they obtain practical data for application. These methodologies of LLM causal inference on application should be provided through research and simulation of real-world LLM applications across the target. The prediction of Korea economic growth based on the causal inference of LLM should be performed through a review of current S. Korean situation and data on purpose because we do not fully study and effectively use by the causal inference of LLM for research target. This paper shows the prediction about GDP growth of S. Korea, China, and world, and stock market trend as example, and compares these graphs to see the results by the causal inference of LLM.

Keywords: LLM, ChatGPT, Causal Inference, Korean GDP growth prediction, Korean Stock market prediction.

Downloads: | DOI: 10.17148/IARJSET.2026.13222