The stock-picking abilities of AI models have recently come under scrutiny, with concerns about their accuracy and reliability. Simon Brown, a financial expert, delved into the performance of two popular AI models, Bing and Google, and discovered some alarming results.
Brown first tested the models’ ability to calculate the compounded annual growth rate of stocks. While the formulas used were correct, the data provided by both Bing and Google was completely unreliable and nonsensical. This was surprising considering both search engines have access to accurate share price data. Instead, the models seemed to randomly select price points that had no resemblance to reality.
Furthermore, when Brown asked for summaries of the balance sheet and debt situations of various companies, both AI models produced gibberish. Even a simple question about British American Tobacco’s debt and its expiry profile proved to be a challenging task for these models.
Nonetheless, Brown did uncover some uses for Bing, which he prefers over Google. By treating it more like an advanced search engine than an intelligent assistant, he found valuable information for generating an investment thesis on electric vehicles (EVs). Asking for a list of exchange-traded funds related to EVs provided him with useful data, including details about the largest stock holdings and expense ratios.
Additionally, Brown used Bing to gain insights into the EV industry, including research reports, major players, and suppliers. This allowed him to develop a better understanding of the industry’s dynamics and risks, serving as a starting point for further, manual research.
Despite these benefits, Brown emphasizes the need for caution when using Bing or any AI model. He notes that it’s essential to double-check all answers provided and recognizes that this can be time-consuming. However, failing to do so renders the entire exercise useless.
In conclusion, while Bing does offer some value as an advanced search engine, relying solely on AI models for accurate financial information remains risky. As investors and researchers, it is crucial to approach these tools with skepticism and do thorough manual verification.