The Digital Déjà Vu: Why Multiple AIs Gave Me the Same "Random" Number

It started with a simple experiment on a Monday afternoon. I was exploring the capabilities of various Large Language Models (LLMs) and decided to give them a trivial task: "Choose a number between 1-50."
I posed the question to ChatGPT. It chose 27. I asked Claude. It also chose 27. I tried Perplexity. Again, 27.A digital déjà vu. It felt too specific to be a coincidence. To break the pattern, I turned to Google's Gemini. It chose 37.

This wasn't a glitch in the matrix or a strange AI conspiracy. It was a fascinating peek under the hood, revealing the core principles of how computers—and the AI models built on them—handle the concept of randomness. What I stumbled upon was a perfect illustration of the deterministic heart of modern AI.

The Illusion of Randomness: A Computer Science Primer

The first thing to understand is a fundamental secret of computing: computers are incapable of true randomness. They are deterministic machines. For any given input, they will always produce the same output.

To create the appearance of randomness, they use something called a Pseudorandom Number Generator (PRNG). Think of a PRNG as a complex mathematical recipe.

1. The Seed: The recipe needs a starting ingredient, a number called a "seed."

2. The Algorithm: The recipe then scrambles this seed through a series of complex, but repeatable, steps to produce a sequence of numbers that appears statistically random.

The golden rule is this: If you use the same seed with the same recipe, you will get the exact same sequence of numbers. Every single time.

This isn't a bug; it's a critical feature called determinism, which is essential for everything from scientific simulations to debugging software.

The AI Layer: Temperature and Other Creativity Dials

For an LLM, a simple PRNG is just the beginning. The final output is governed by sophisticated parameters that control the model's "creativity." The most important of these is Sampling Temperature.

You can think of Temperature as a creativity dial:

• Temperature = 0: The dial is turned all the way down. The model becomes completely deterministic. It will look at all possible next words or numbers and always choose the one with the highest mathematical probability. It’s a strict, predictable newscaster.

• Temperature > 0: The dial is turned up. The model is allowed to take more risks, sometimes picking a less likely option to introduce variety and creativity. It becomes more like a jazz musician improvising on a theme.

A Tale of Two Philosophies: Analyzing the Numbers

This brings us back to 27 and 37. The results I saw were not a coincidence, but a window into competing design philosophies.

The Case of the Convergent 27s

It's highly probable that for my simple, low-stakes query, ChatGPT, Claude, and Perplexity all defaulted to a similar, highly deterministic state. This could be due to:

1. A Fixed Seed or Low Temperature: They likely used a default, fixed seed and set the Temperature to a very low value (or zero). This is computationally efficient and ensures a predictable output for trivial tasks. They all ran the same recipe with the same starting ingredient and got the same result.

2. Model Convergence: The number 27 might be a high-probability answer based on their training data. It appears frequently in mathematics, computer science literature (famously in Donald Knuth's work), and even culture (the "27 Club"). For a model seeking the "most likely" random number, 27 is a surprisingly safe bet.

The Divergent 37

Gemini's choice of 37 signifies a different approach. Its process was stochastic, or non-deterministic relative to the others. This was likely achieved by:

1. A Randomized Seed: Instead of a fixed value, its PRNG was likely seeded with a dynamic number, such as the server clock's time down to the nanosecond when I sent my query. A different seed means a different outcome.

2. A Higher Temperature Setting: Gemini may have a higher default temperature, encouraging it to pick a less obvious—but still valid—answer, thus appearing more genuinely "random" to the end-user.

Conclusion: It's Not a Ghost, It's a Blueprint

What began as a simple question revealed that there's no ghost in the machine. There's a blueprint. The behaviour of these complex AI systems is a direct result of thousands of deliberate, human-made engineering choices.

The 27s represent a philosophy of efficiency, predictability, and safety. The 37 represents a philosophy that prioritizes variability and a more human-like creative spark. Neither is inherently better, but seeing them side-by-side is a powerful reminder that the AI tools we use every day are not magic; they are a reflection of the logic, biases, and design choices of the people who built them.

And that, to me, is far more fascinating than any coincidence.