Imagine giving an AI a company's balance sheet and income statement, asking it to predict the company’s future, and seeing it outperform professional financial analysts. That’s not science fiction—new research suggests large language models (LLMs) can do just that. Researchers explored how well the powerful GPT-4 model could forecast earnings changes based solely on financial data, finding it remarkably accurate. GPT-4’s success stems from its "chain-of-thought" prompting, which guides the model through a step-by-step analysis mimicking human reasoning. The model breaks down financials, identifies trends, calculates ratios, and even generates narrative explanations. Surprisingly, GPT-4 outperforms not only analysts' one-month-ahead predictions but also forecasts made after three and six months (though those later forecasts incorporate more information). GPT-4 even rivals specialized machine-learning models specifically designed to predict earnings. Notably, it displays unique strengths in analyzing smaller, loss-making companies—cases where traditional models and sometimes even humans struggle. But humans still have an edge when soft information or context beyond the numbers is crucial. While concerns might arise about the AI potentially accessing future information from its vast training data, researchers tackled this issue by using anonymized statements and testing GPT-4 on 2023 earnings—data it couldn't have seen during training. The results? Equally impressive. This suggests that AI-driven financial statement analysis isn’t just a futuristic concept. It's here, providing a potential tool to democratize financial analysis, complement existing methods, and possibly even uncover hidden value in the market. While questions remain about how to best incorporate additional data and refine prompting strategies, the potential for LLMs to reshape financial analysis is undeniable.