We examine investor behavior and trading strategies in the age of social media and AI-powered trading. Applying large language models to 77 million messages from 800,000 users on a leading social media platform, we classify posts into technical analysis (TA), fundamental analysis (FA), and other strategy categories. TA posts increase substantially during periods when fundamental news is scarce. High TA sentiment is associated with lower future returns, less informative aggregate retail order flows, and a higher likelihood of herding by Robinhood traders. A state-of-the-art AI strategy exploiting price trends generates significant profits only when trading against TA sentiment. In contrast, non-TA posts, especially those related to FA, exhibit sentiment that positively predicts future returns and tend to increase the informativeness of retail order flows. Our findings highlight the heterogeneity of retail trading strategies and suggest that a key source of AI profitability is the exploitation of retail investors’ tendency to herd on technical signals.