MarketGPT was inspired by research conducted at the University of Florida, where ChatGPT was tasked with predicting market performance based on news headlines and achieved over a 500% annual profit: https://ssrn.com/abstract=4412788 Motivated to replicate these results, I developed MarketGPT. This AI model analyzes headlines and abstracts of financial news articles to predict which stocks are most likely to experience a change in value the following day. Each day, it identifies a number of such stocks, which are then published in a mobile app. You can trade these stocks using a simple procedure that requires only two trades per stock, one when the trading opens and one before it closes for the day. The model was trained on data from 2018 to 2022 and tested on 2023 news, where it achieved a 609.75% profit from a total of 564 stocks. You can find a detailed analysis of the results on my website: https://www.marketgpt.app/perfor... I have strived for transparency with the model's results: all stock predictions are archived on the website and can be reviewed to verify historical performance. For current predictions, there is a free 7-day trial available, allowing users to confirm that the stocks published in the archive match those selected during the day. One limitation of the current model is that it only provides "Sell" predictions. This is because the market tends to react more strongly and predictably to negative news, a phenomenon also observed in the original Florida research. I aim to enhance the model in the future to include "Buy" recommendations as well.