AI trading bots are booming – but can you trust them with your money?

When 17-year-old Nathan Smith handed a Chatgpt-driven trade bone a portfolio of micro-CAP shares, it yielded a profit of 23.8% in four weeks on the performance of the Russell 2000 and launched him from the countryside of national Oklahoma to Viral Reddit-Sterdom.

Smith’s trip from rural high school to Peak R/Wallstreetbets Poster Boy is part of a larger movement that blooms on the internet with traders who build Stock-Pick systems around ready-made large language models.

The internet is littered with viral claims about AI trading success. One Reddit -Post recently caught fire after he had claimed Chatgpt and Grok, a “flawless, 100% Win Rate” achieved more than 18 transactions with quite large profit. Another account gave $ 400 to chatgpt with the aim of becoming the world’s first AI-made billionaire “

However, none of the posts has provided verification – there are no tickers, trade logs or coupons.

However, Smith attracted attention, precisely because he documents his journey on his substit and shares his configurations, prompts and documentation on Github. This means that you can replicate, improve or adjust his code at any time.

AI-driven trade is not only a Reddit fantasy Meer-it quickly becomes Wall Street Reality.

From amateur codes who use open-source bots to investment giants such as JPMorgan and Bridgewater Building tailor-made AI platforms, a new wave of Markt tools promises faster insights and hands-free profits. But while personal experiments spread viral and institutional tools quietly, experts warn that most large language models still do not miss precision, discipline and reliability that is needed to exchange real money on a scale. The question now is not whether AI can exchange – it is whether someone should leave it.

JPMorgan rolled an internal platform from the name LLM Suite, described as a “chatgpt-like product” for 60,000 employees. It pars speeches, summarizes the registrations, generates memo concepts and feeds a thematic ideas motor called indexgpt that builds tailor-made theme-based stock baskets.

Goldman Sachs calls his chatbot the GS AI assistant, built on his own LLama-based GS AI platform. Now on 10,000 desktops in engineering, research and trade agencies, it is reportedly generating up to 20% productivity gains for codewriting and model structure.

The Bridge Gewater research team built his investment analyst Assistant on Claude, used it to write Python, generate graphs and summarize an income commentary – tasks that a junior analyst would do within days, done within a few minutes. The sovereign Wealth Fund (NBIM) of Norway uses Claude to control the news current in 9,000 companies, which saves an estimated 213,000 analyst hours per year.

Elsewhere, platforms such as 3Commas, Kryll and Pionex Chatgpt integration for trade automation, according to Phemex. In February 2025, Tiger Brokers integrated the AI model of Deepseek, Deepseek-R1, in their chatbot, Tigergpt, improving market analysis and trade options. At least 20 other companies, including Sinolink Securities and China Universal Asset Management, have adopted Deepseek’s models for risk management and investment strategies.

All this raises an obvious question: have we finally come to the point where AI can make good financial bets?

Is AI-Assisted Trading finally ready for Prime Time?

Multiple studies suggest that AI, and even chatgpt-improved systems, can exceed manual and conventional machine learning models when predicting crypto price movements.

BCG and Harvard Business School, however, warned against too high dependence on generative AI, which stated that GPT-4 users conducted 23% worse Then users who shun AI. They see Jibes with some other professionals.

“Only because you have more data does not mean that you add more returns. Sometimes you just add more noise,” said Cio Russell Korgaonkar of Man Group. The systematic male Group systematic has trained chatgpt to digest papers, write internal python and sort ideas on the watch lists – but you still have to do a large part of the heavy lifting before you even think about using an AI model reliably.

For Korgaonkar, generative AI and typical aids for machine learning have different applications. Chatgpt can help you with fundamental analysis, but will suck on price forecasts, while the non-generative AI tools are unable to tackle fundamentals, but be able to analyze data and perform pure technical analysis.

“The breakthroughs of Genai are on the language. It is not particularly useful for numerical predictions,” he said. “People use Genai to help them in their work, but they don’t use it to predict markets.”

Even for fundamental analysis, the process that leads an AI to a specific conclusion is not always reliable.

“The fact that models have the opportunity to hide underlying reasoning suggests that disturbing solutions can be avoided, indicating that the current coordination methods are insufficient and require enormous improvement,” said Bookwatch founder and CEO Miran Antamian Decrypt. “Instead of just reprimanding ‘negative thinking’, we need blended approaches of iterative human feedback and adaptive remuneration functions that actively shift over time. This can greatly help in identifying behavioral changes that are masked by fines.”

Gappy Paleologo, partner at Balyasny, pointed out that LLMS still does not miss “real-world earthen” and the nuanced judgment that is needed for high condemnations. He sees them best as research assistants, not portfolio managers.

Other funds warn about the model risk: these AIs are inclined to present unbelievable scenarios, macro language and hallucinate to read incorrectly to insist on the audit of the human-over-the-loop for each AI signal. And what is even worse, the better the model is, the convincing it will be to lie, and the harder it will be to admit an error. There are studies that prove this.

In other words, so far it is extremely difficult to get people out of this comparison, especially when money is involved.

“The concept of monitoring more powerful models with the help of weaker, such as GPT-4O is interesting, but it is unlikely that it is sustainable for an indefinite period of time,” Antamian said Decrypt. “A combination of automated and human evaluation of experts can be more suitable; looking at the reserved level of reasoning may require more than one guided model to supervise.”

Even Chatgpt itself remains realistic about his limitations. When asked directly about making someone up to a millionaire by trade, Chatgpt reacted with a realistic prospect – acknowledges that, although it is possible, success depends on having a profitable strategy, disciplined risk management and the ability to effectively scale.

Yet for hobbyists it is nice to tinker with things like this. If you are interested in exploring AI-assisted trade without full automation, Decrypt Has developed its own instructions, only for fun – and probably clicks. Our Deg Portfolio Analyzer provides personalized, color -coded risk assessments that adapt to whether you are a degenerate trader or a conservative investor. The Framework integrates fundamental, sentiment and technical analysis during the collection of user experience, risk tolerance and data from investment time line.

Our Personal Finance Advisor promptly is intended to provide analysis of institutional quality with the help of the same methods as large investment companies. When tested for a Brazilian stock portfolio, the concentrated exposure risks and currency mermatches identified, which generates detailed re -balancing recommendations with specific risk management strategies.

Both prompts are available on Github for anyone who wants to experiment with AI-assisted financial analysis, although the experiment of Smith shows, the most interesting results sometimes come from having the wheel completely taken and simply perform what the machine says.

Not that we would ever advise someone to do that. Although you may not have a problem to give $ 100 to chatgpt to invest, there is no chance that you will see JP Morgan do that. Yet.