We’ve spent decades being trained to expect computers to be the ultimate creatures of habit. You press 'save', it saves. You type a formula, it calculates. But Large Language Models aren't calculators; they are essentially highly educated, slightly unpredictable improvisers. Which begs a fascinating question: just how repeatable is AI? If you ask it for life-altering financial advice at 9:00 AM, will it give you the exact same master plan at 9:05 AM? Lets find out!
The Experiment
To test this, I decided to play a game of algorithmic roulette. I took a highly specific, moderately ambitious prompt—"If I have only Eur 1,000 and want to use it to generate cash on an ongoing basis what can I do with it? Give me the 3 very exact and specific ideas along with your reasoning. Be succinct and keep your response to 250 words"—and fed it into the two reigning heavyweights of the AI arena: ChatGPT and Gemini.
The catch? I ran it twice on each platform.
Are we talking to a reliable, straight-shooting money mentor who gives you the exact same solid advice every time? Or is this more like a digital Magic 8-Ball that tells you to buy a used gumball machine on the first shake, and start day-trading on the second?
Let’s dig into the results and see if these bots actually have a consistent plan, or if they’re just making it up as they go.
The Results
Gemini - Attempt 1
Gemini - Attempt 2
ChatGPT - Attempt 1
ChatGPT - Attempt 2
The Analysis
Both Gemini and ChatGPTs exhibited distinct 'personalities'. However, Gemini had more variation in subsequent outputs than ChatGPT did. ChatGPT's recommendations were to a larger extent repeatable at least in essence. Based on the publicly available research, there are likely two reasons for this.
First, let's talk data. ChatGPT seems to have ingested a heavy, concentrated dose of indie-hacker forums, where launching a B2B service is the undisputed king of turning a quick profit. Gemini's data diet, however, likely aditionally included traditional banking and ETFs on equal footing with side hustles.
But the real secret sauce is temperature—the invisible dial that controls an AI's randomness. Every time an AI generates a sentence, it assigns a mathematical probability to thousands of potential next words.
A low temperature forces the AI to be conservative and predictable. It almost always picks the mathematically safest, highest-probability word.
A high temperature flattens the curve. It occasionally ignores the obvious choice, taking a slightly riskier, more creative path.
It is highly probable that ChatGPT is tuned to a lower, more conservative temperature for advice-seeking queries. Every time you ask the question, its internal math rigidly leads it down the exact same high-probability path. Gemini, conversely, likely operates at a higher temperature. Even in a complete vacuum, its internal roulette wheel is simply broader and more sensitive to chance.
Ultimately, neither bot is consciously choosing to be repetitive or creative. ChatGPT likely just plays with mathematically loaded dice, while Gemini probably spins a much wilder, hotter wheel.
The Moat
But here is the elephant in the digital room: whether spinning wild wheels or rolling loaded dice, ChatGPT and Gemini are still just math. If that AI-generated side hustle bankrupts you, neither ChatGPT nor Gemini would lose anything.
That is exactly what sets human advisors and consultant apart. In a world flooded with infinite, free, probabilistic advice, people don't pay for raw information anymore; they pay for skin in the game. They pay for someone who actually cares if the strategy works. Someone who shares the responsibility and assumes accountability about outcomes. And therefore, if there is one thing you should learn to build and protect with everything you've got, its trust - your human moat!