$88M AUM. +111% since last year using Grok to pick stocks in the WWIII portfolio. “LLMs are bad at stock picking.” You’re right. They are. But here’s what people miss: we don’t use LLMs to pick stocks. We built a SYSTEM that uses LLMs. Completely different thing
Ask Grok: “Should I buy RKLB?” You’ll get: – Outdated info or brief web search – Vague language (“could be a good opportunity”) – No accountability – No way to track if it’s right This is useless. And this is what 99% of people do.
A system isn’t “asking AI for stock picks.” A system is: – Injecting real-time data – Forcing structured outputs – Comparing against benchmarks – Tracking predictions over time – Iterating on what works The LLM is a component. Not the system.
Raw LLMs don’t know BKSY’s current price or backlog, sure, it may search but it is unreliable. Our system: 1. Pulls live financials 2. Fetches news 3. Contains geopolitical context 4. Injects ALL of this into the prompt Now the LLM has something real to analyze.
Raw LLM says: “BKSY has strong growth potential” Our system forces: – Expected return percentages Vague → quantified. Untrackable → measurable.
November 17, 2025. System outputs: RKLB: BUY, +42% expected 12M return KTOS: BUY, +38% expected 12M return 8 weeks later: RKLB: +90% KTOS: +56% You can’t improve what you don’t measure.
Each prompt trick compounds: One prompt becomes a research report that rivals Wall Street. Not because the LLM is smart. Because the SYSTEM extracts what the LLM knows.
We run this on all stocks every quarter. Designed by @alejandroll10 and @ai_prof_funds AI vs. Street. Every single name. Looking for disagreements. $88M AUM. +111% since Jan 2025.
Source: https://x.com/grkportfolio/status/2011258910276198711
Can Grok outperform politicians in investing? We partnered with a Wharton PhD to test whether Grok could beat the S&P 500, and, more interestingly, politicians. It turns out Grok can actually outperform Nancy Pelosi, and of course, the S&P 500. Here’s how we did it.
Every month, we ask Grok to forecast everything that can affect the stock market: Inflation, economic growth, unemployment, etc. But Grok can also forecast other events that we cannot with traditional methods
For example, Grok can reason and forecast about Supreme Court decisions, conflict escalations, tariffs, or other policy events. After that, we need to research thousands of firms in depth.
We find it best to split the task; we run thousands of parallel Grok calls using code. For each Grok instance, we investigate every firm in depth, considering recent price movements, news, earnings, catalysts, and anything that could affect the stock’s performance.
We then provide a fresh Grok instance with all the information from the previous steps and ask it to construct a 15-asset portfolio that should outperform the S&P 500. Grok may conduct additional research and is not limited to any particular assets. For example, it sometimes includes gold as a hedge.
Grok is up 33% since last February, vs 15% for the SP 500. You can join hundreds of users copying its trade via @joinautopilot. The portfolio is designed by @ai_prof_funds @alejandroll10
Source: https://x.com/grkportfolio/status/2015177148886122655
We’re up 111% since Jan 2025 using AI to pick stocks. People want to do this, but often make mistakes. You ask an LLM to analyze a stock. It searches, finds analyst reports, and gives you… the analyst consensus. 5 min wasted for a fancy Google search Here is how we fix it
“Analyze IRDM with current data.” LLM searches the web. Finds news, analyst ratings. Returns: “Analysts have a HOLD rating with a $29 price target. The stock faces headwinds from Starlink competition.” Cool. You could’ve googled that in 10 seconds. Where’s the insight?
You don’t want AI to parrot consensus. You want AI to DISAGREE with consensus – and explain why. That’s where alpha lives. But LLMs won’t do this by default. They’re trained to summarize, not to challenge. You have to force it.
The line we add: “State the Wall Street analyst consensus. Then provide YOUR independent view. Where do you agree? Where do you DISAGREE? Explain specifically why your analysis differs from the Street.” Now the LLM has to think, not just summarize.
Without this line: “Analysts rate IRDM a HOLD. Target $29. Concerns about Starlink.” With this line: “Street says HOLD at $29. I DISAGREE. Rating: BUY. The 52% drawdown overweights Starlink fear and underweights the $738M Space Force contract catalyst. Expected return 12 months:
November 2025. IRDM at $16.54. Wall Street: HOLD. Starlink fears. Mixed sentiment. Our AI: “I disagree with the HOLD. The selloff is overdone. Analysts are missing the government contract stability. BUY.” It found what the Street was missing. January 2026: Up 17% since then and climbing.
The best opportunities are where AI disagrees with consensus: IRDM: Street said HOLD → AI said BUY → Up 17% But agreement matters too: RKLB: Both said BUY, ~42% target → Up 90% Agreement = validation Disagreement = opportunity
We run this on all stocks every quarter. AI vs. Street. Every single name. Looking for disagreements. $88M AUM. +111% since Jan 2025.
Source: https://x.com/ai_prof_funds/status/2011254899271684244