Can artificial intelligence predict the stock market? A groundbreaking new study suggests it might be possible, and with remarkable accuracy. Researchers at University College London explored the potential of large language models (LLMs) like OPT, BERT, and FinBERT to analyze financial news and predict stock returns. They discovered that the GPT-3-based OPT model significantly outperformed traditional methods, achieving a 74.4% accuracy in predicting market movements. This isn't just theoretical; a simulated trading strategy based on OPT's insights yielded a staggering 355% return over a two-year period, dwarfing standard market portfolios and highlighting the transformative potential of LLMs in finance. The study analyzed nearly a million U.S. financial news articles, meticulously filtering for relevance and novelty. By fine-tuning these LLMs to understand the nuances of financial language, the researchers created a system that could interpret news sentiment and its impact on stock prices with unprecedented precision. While previous studies have used simpler dictionary-based approaches, this research demonstrates the power of LLMs to delve deeper into the complex relationships between news and market behavior. The superior performance of OPT, attributed to its larger parameter space and advanced training, opens exciting new avenues for AI-driven investment strategies. While the Loughran-McDonald dictionary model, a common tool in financial research, showed limited predictive power, the LLM-based strategies, especially OPT, delivered significantly higher returns. This suggests that LLMs can uncover hidden market signals within news data that traditional methods miss. The research not only offers a glimpse into the future of AI-powered investing, but also raises important questions for regulators and policymakers about the responsible use and oversight of these powerful technologies as they become increasingly integrated into financial markets.