> ## Content Index
> Fetch the complete content index at: https://larrysalibra.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Will AI mean less variance in the world?
- URL: https://larrysalibra.com/will-ai-mean-less-variance-in-the-world/
- Published: 2026-10-15T13:01:24.000Z
- Updated: 2026-10-15T13:02:05.000Z
- Author: Larry Salibra
- Tags: ai, day-98, daily

When asked to pick a random number, humans tend to pick the number 7 more than the other numbers. Because LLMs are trained on human output, they take this bias and turn it into what almost amounts to a rule. Go ask your favorite LLM to pick a random number between 1 and 10 and it will probably pick 7\. Grok did:

![](https://storage.ghost.io/c/e4/74/e474b4cb-cbee-4c30-844d-ae523b708967/content/images/2026/10/Screenshot-2026-10-15-at-8.29.53---PM.png)

That got me to thinking: in a world where an increasingly large amount of code and other content - some of which becomes absorbed into culture - is generated by LLM, will we see a similar decrease in variance? Assume for a moment that humans generally like all 3 primary colors, red, blue and green somewhat equally except that red is 10% more popular than blue and green. Does that mean that LLMs will strongly prefer the color red when asked to pick a color from one of those choices?

What will this mean for product design or analysis conducted by AI? Will we increasingly see me-too products that look and function the same just like certain categories of products have converged on whatever Apple makes in that category? (We already have that to some extent...will the "problem" get worse?)

What will this mean for the marketplace for ideas? In a world where people frequently interact with and learn from AI, will we see a similar reduction in variance in the range of ideas both considered and considered acceptable? Will the conventional narrative become an even stronger convention accepted by even more people because of AI's tendency to enhance the biases in its training data?

I'm not sure what the answers are to these questions. Is it that we have already reached peak heterogeneity in the idea market in the roughly three decades between end of mainstream media's monopoly on idea market thanks to the rise of the internet and the rise of generative artificial intelligence? Perhaps we are destined for a future of AI-intermediated culture that will look a lot more like our broadcast television past and a lot less like the past two or three decades.