AI Trading Gains Momentum Amid Debates on Reliability and Human Oversight

AI trading bots have surged in popularity, sparking debates over their reliability in managing real money. A 17-year-old from Oklahoma, Nathan Smith, gained attention after using a ChatGPT-powered bot to pick micro-cap stocks, achieving a 23.8% gain in four weeks. His open-source experiment, documented on Substack and GitHub, has inspired similar efforts across Reddit and other online communities, where viral claims—albeit unverified—highlight AI's potential in trading [1].

While many online claims remain unverified—lacking trade logs or receipts—Smith’s transparency has made his project a reference point for AI-driven trading. The trend extends beyond amateur experiments: JPMorgan has rolled out its internal AI platform, LLM Suite, to 60,000 employees. The tool helps summarize regulatory filings, interpret Fed speeches, and generate thematic equity baskets through a product called IndexGPT [1].

Goldman Sachs introduced the GS AI Assistant, a proprietary AI tool based on LLaMA, deployed across 10,000 desktops for tasks such as code generation and model building, reportedly improving productivity by 20% [1]. Bridgewater and Norway’s sovereign wealth fund (NBIM) also utilize AI models for tasks like summarizing earnings reports and monitoring global news flows, with the latter saving an estimated 213,000 analyst hours annually [1].

Third-party platforms are also integrating AI into their trading systems. 3Commas, Kryll, and Pionex offer ChatGPT integration for automated trading, while Tiger Brokers adopted DeepSeek’s AI model to enhance market analysis and trading capabilities [1]. At least 20 other firms, including Sinolink Securities and China Universal Asset Management, use DeepSeek’s models for investment strategies and risk management [1].

Despite growing adoption, experts caution that AI is not yet ready for prime time in real-money trading. While AI can outperform traditional models in predicting crypto price movements, broader research from BCG and Harvard Business School warns that over-reliance on generative AI can lead to poorer performance. Russell Korgaonkar, CIO of Man Group, emphasized that generative AI excels in fundamental analysis but struggles with numerical predictions and price forecasting [1].

Miran Antamian of BookWatch noted the opaque nature of AI reasoning and the limitations of current alignment methods, suggesting that blended approaches with human feedback are necessary for improvement [1]. Gappy Paleologo from Balyasny highlighted that LLMs lack the nuanced judgment needed for high-conviction trades and should be treated as research assistants rather than portfolio managers [1].

Firms also express concerns over model risk, including the tendency of AIs to generate implausible scenarios, misread macroeconomic signals, and hallucinate. This has led many institutions to require human oversight for every AI-generated signal [1].

Even ChatGPT itself acknowledges its limitations when asked directly about trading success. It admits that while profitable outcomes are possible, they depend heavily on sound strategy, risk discipline, and the ability to scale effectively [1].

For now, AI remains a tool that augments human effort rather than replaces it. Decrypt has developed prompts for AI-assisted trading, such as a Degen Portfolio Analyzer and a Personal Finance Advisor, offering color-coded risk assessments and rebalancing suggestions [1]. These tools, available on GitHub, allow hobbyists to experiment with AI-driven financial analysis—though, as Smith’s experience shows, full automation remains a risky proposition.

The debate continues: while AI trading is advancing rapidly, the consensus remains that humans must stay in the loop, especially when real money is at stake [1].

Source: [1]AI Trading Bots Are Booming—But Can You Trust Them With Your Money? (https://decrypt.co/333303/ai-trading-bots-booming-can-you-trust-them-with-your-money)