Fresh research has asserted large language models, particularly GPT-4 (which empowers certain iterations of ChatGPT and various Microsoft Copilot-branded generative AI products) possess the ability to inspect financial statements with superior precision than humans.
The discoveries from scholars at the University of Chicago suggest significant implications for the future of financial examination and decision-making as AI becomes more widespread.
The analysis also emphasizes the adaptability of generic, multipurpose LLMs such as GPT-4, which can provide similar capabilities as more specialized tools, observing, “We find that the prediction accuracy of the LLM is on par with the performance of a narrowly trained state-of-the-art ML model.”
LLMs excel at examining financial reports
In their examinations, the researchers discovered that GPT-4 surpassed human analysts even in the absence of textual context, showcasing the technology’s precision of 60% in contrast to the 53-57% range of human analysts.
The triumph was not achieved without some preliminary groundwork, though, with the document delving into specifics about the researchers’ utilization of chain-of-thought prompts to develop more suitable and precise responses.
Furthermore, the research unveiled that GPT-4 and human analysts complement each other effectively – while the LLM thrives in areas where humans may be ineffective or prejudiced, humans contribute value where additional context is necessary.
GPT-4’s capacities were ascribed to its extensive knowledge base and theoretical comprehension, allowing it to infer conclusions from data patterns even lacking specific financial training, and although the model exhibited some constraints, advancements have already witnessed the most recent GPT-4o model substantially enhancing efficiency while evolving into a multimodal model.
Despite the skepticism revolving around generative AI’s readiness to supplant human workers, its role as an essential support is increasingly apparent as human workers gear up to integrate with the efficiency-enhancing technology.
