Reddit user Nathan Smith handed $100 and full trading responsibility to GPT-4o, OpenAI\u2019s latest AI model. In just one month, the AI managed to outperform major market indices \u2014 including the S&P 500. But experts warn: don\u2019t ditch human judgment just yet.<\/p>\n
Gpt-4O Turns $100 Into $125 in Four Weeks<\/p>\n
Nathan Smith\u2019s experiment started with a simple idea: could a publicly available AI model beat the stock market with a small budget and no prior financial expertise? He gave GPT-4o, OpenAI\u2019s most recent release, control over a $100 portfolio and asked it to pick small-cap stocks over a six-month period. At the one-month mark, he reported a 25% gain \u2014 far outpacing the S&P 500, which rose less than 3% during the same period.<\/p>\n
Smith shared a detailed breakdown of the results on Reddit<\/a>, along with visual performance graphs and a GitHub page outlining his methodology<\/a>. The AI model selected stocks daily based on recent performance trends and sentiment data. Smith then manually executed the trades and applied a simple stop-loss rule to minimize downside risk \u2014 selling any stock that dropped below a preset threshold.<\/p>\n According to Smith, GPT-4o consistently recommended stocks that had short-term upside potential, often tapping into biotech and emerging tech companies. \u201cIt was evident early on that the model wasn\u2019t just guessing,\u201d he wrote. \u201cThe picks made sense \u2014 and the results spoke for themselves.\u201d<\/p>\n AI Outperformed Traditional Indices \u2014 but Under Strict Conditions<\/p>\n While the early performance looks strong on paper, it\u2019s important to put the results in context. The portfolio size was extremely small, and trades were executed manually with heavy human oversight. Smith noted that although GPT-4o generated recommendations, he was the one responsible for applying trading rules and interpreting unclear results.<\/p>\n This isn\u2019t the first time AI models have been used to pick stocks. A 2024 study from the University of Duisburg-Essen in Germany found that certain OpenAI models could spot patterns leading to short-term gains. But researchers stressed that success depended on careful prompt design and limited trade volume. In larger-scale simulations, the performance often leveled out or declined.<\/p>\n