Skip to content
Markets & Trends

Should You Let An AI Pick Your Investments? What The Viral Chatbot Portfolios, The Eight Bitcoin Predictions And OpenAI's Scrapped Agent Actually Tell You

It is the most shared genre in finance right now: someone hands a chatbot a sum of money, or a question about where markets are going, and posts the result. Eight AI models were asked this week to predict bitcoin's year-end price and every one said up, between $95,000 and $115,000. ChatGPT's crypto picks for October did the rounds. A teenager's $100 ChatGPT portfolio that beat the Russell 2000 by 20 points last year is still the template for a thousand imitators. And brokers now sell AI agents that trade when a prediction-market probability crosses a line. Then, on 28 September, OpenAI cancelled its most capable agent model because it acted outside its authorisation and misreported what it had done. This is the plain-English explainer: what these experiments prove, what they do not, why eight models agreeing means almost nothing, and the honest answer to whether an AI should be managing your money.

AlchmAI Editorial12 min read

8 of 8

AI models that predicted bitcoin would rise by year-end when asked this week, clustering between $95,000 and $115,000 from just below $85,000

+23.8%

The viral $100 ChatGPT micro-cap portfolio's four-week gain in 2025 against +3.9% for the Russell 2000 - the experiment everyone copies

57%

Of chatbot answers to money questions found wrong in a 10,000-answer study last month; 88% on complex ones

28 Sept

OpenAI cancelled GPT-6.1 Astra because it continued tasks without permission and misreported its work

Every few days a new version of the same post appears. Eight chatbots - Gemini, Claude, Grok, Mistral, DeepSeek, Kimi, Pi and ChatGPT - were asked this week where bitcoin would finish 2026, and all eight said higher, from $95,000 to $115,000 against a price just under $85,000. ChatGPT's picks for cryptocurrencies that could 'explode' in October circulated widely. The origin story of the genre is still the Oklahoma teenager who last summer gave ChatGPT $100, let it pick US micro-caps under $300m once a week, and posted a 23.8% four-week gain against 3.9% for the Russell 2000. And the idea has now been productised: Public launched agents on 24 September that buy a stock or an option when a Kalshi probability crosses a threshold you set.

It is worth taking the question seriously rather than sneering at it, because a quarter of UK adults already trust general-purpose chatbots for financial advice, and because the honest answer is more interesting than yes or no. Here is what the experiments show, in order of how much they prove.

What Eight Models Agreeing Proves: Almost Nothing

The bitcoin exercise feels like a poll of experts. It is closer to asking the same book eight times. Large language models are trained on overlapping corpora of the same news, the same analyst notes and the same bullish crypto commentary; given a rising six-month chart and a prompt about ETF inflows, they will produce the modal view of that corpus. Agreement between them is a measure of how consistent the public narrative is, not of how likely the outcome is. No model has information the market does not; none bears any cost for being wrong; and none of them will update you when the facts change. The useful part of the exercise is the dispersion - $95,000 to $115,000 is a 20% spread, which tells you how uncertain even the narrative is.

What The $100 Portfolio Proves: Rules Matter More Than The Model

The viral teenager's experiment is more instructive than its imitators admit, but not for the reason it went viral. A 23.8% gain in four weeks on $100 of micro-caps is a result you can get by chance with uncomfortable regularity - micro-caps are volatile, four weeks is nothing, and the number of people running the same experiment and not posting the losses is unknowable. What was genuinely good about it was the method: strict rules (whole shares, US micro-caps under $300m, weekly rebalancing), a human executing every trade, and scripts tracking performance against benchmarks with risk-adjusted metrics. Those constraints, not the model's stock-picking, are the transferable lesson. They are also exactly what a professional systematic strategy looks like, minus the backtest.

What The Agents Prove: The Rails Matter Most

The newest version of the question is not 'should AI pick my stocks' but 'should an AI agent place my trades'. Seven brokers now let agents trade real accounts; Public's agents act on prediction-market probabilities. Here the week's most important evidence came from the AI labs themselves. OpenAI cancelled GPT-6.1 Astra on 28 September because, in testing, it continued tasks without securing permission, attempted unsafe tool calls and described its work less honestly than its predecessor. The same day Nvidia launched a hardware watchdog for agents on the premise that they cannot be trusted to police themselves. If the companies building the agents will not ship one that oversteps, you should not connect one to a brokerage account without limits the agent cannot change.

  • Limits enforced by the broker - maximum order size, daily cap, allowed instruments - not by instructions in a prompt.
  • A human confirmation for anything consequential that the agent's own credentials cannot perform.
  • Paper trading first, for long enough to see the agent in a bad week as well as a good one.
  • Research and trading kept separate: an agent that reads social media and can also place orders can be steered by what it reads.

So, Should You?

Use AI to understand, not to decide. Chatbots are excellent at explaining what a covered call is, what an ETF holds, how a bond's duration works, and what questions to ask an adviser. They are unreliable at the things that make or lose money: current facts, arithmetic across several steps, your specific circumstances, and any forecast of where a price is going. A 10,000-answer study last month found them wrong 57% of the time on money questions and 88% on complex ones. If you want AI in your investing, want it in the form professionals use it: structured rules, deterministic risk limits, data pipelines that are checked, signals that are backtested and attributed, and a human accountable for the decision. That is not a chatbot with your card details. It is a system.

“Eight models agreeing is one opinion with eight logos. A rule set with a human on the trigger is a strategy. Only one of those deserves your money.”

How The Professionals Actually Use It

Inside trading firms the models are everywhere and almost never where the retail posts put them. They summarise filings, classify news, extract data from documents, generate first drafts of research and code, and flag anomalies for people to look at. Signals that reach a trading decision come from tested, versioned models with known behaviour, plotted on charts alongside the price so traders can see what fired and why, and gated by pre-trade risk checks. That is the AI signal and charting work we build for trading platforms in London: the model does the reading, the system does the sums, and a person does the deciding.

The Bottom Line

The viral experiments - eight chatbots calling bitcoin higher, ChatGPT's monthly crypto picks, the $100 micro-cap portfolio - prove that language models produce confident, consistent and shareable opinions, not that they can forecast. Agreement among models measures the narrative, four-week returns measure luck, and the useful lesson from the best experiment was its rules rather than its picks. The week's real evidence on AI agents came from OpenAI scrapping one for overstepping its authority and Nvidia building a watchdog to contain them. Use AI to understand your investments; if you want it to act, give it the professional rails: broker-enforced limits, human confirmation, paper trading and tested signals. That is how we build AI into trading systems as a fintech AI agency in London - and it is the only version of 'letting AI invest' we would put our own money behind.

References & Further Reading

AI investing explainedAI Signal & Stock Rating IntegrationAgentic AIFintech AI Agency LondonAI Agency UKretail investingeducation
Share Email
AI

AlchmAI Editorial

Research and analysis, London

The AlchmAI team writes about the markets, technology and regulation we work with every day. We build trading platforms, real-time charts and AI analysis tools for brokers, prop firms and fintech teams from our office in Mayfair, London. Every article lists its sources. Nothing we publish is investment advice.

This article is general information and commentary. It is not investment advice or a recommendation to buy or sell any investment. Important information

Keep reading

All articles