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

# The gap in shared understanding of LLM capability is widening
- URL: https://www.stirup.si/the-gap-in-shared-understanding-of-llm-capability-is-widening-d16bd93f/
- Published: 2026-10-12T06:19:37.000Z
- Updated: 2026-10-12T06:19:37.000Z
- Description: Andrej Karpathy highlights a growing disparity in how different groups perceive the capabilities of large language models (LLMs). This gap is no longer a simple divide between two groups but a sharp funnel, with only a small fraction of the population experiencing the full potential of frontier-grad
- Author: Satish Das
- Tags: Community, Stirup Research

## Understanding the widening gap in LLM capability perception

Andrej Karpathy’s observations reveal a significant divergence in how different segments of the population perceive the capabilities of large language models (LLMs). This gap is no longer a simple divide between two groups but a sharp funnel, with only a small fraction of the population experiencing the full potential of frontier-grade systems.

According to Karpathy, approximately 6 billion people (around 75% of the global population) have had minimal or no interaction with LLMs. For many in this group, exposure to AI is limited to basic chatbot interactions, which are often perceived as rudimentary tools—similar to a more advanced version of a search engine or a writing aid. This limited experience shapes a general perception of LLMs as tools with narrow utility, primarily useful for simple tasks like drafting emails or searching for information.

A larger segment, around 1 to 2 billion people (20% of the population), are casual users of free-tier LLM products such as ChatGPT. These users may engage with AI tools occasionally, but their experience is still limited to basic functionalities. For many, LLMs are not transformative tools but rather convenient utilities that assist with routine tasks. This group includes professionals outside of technical fields, such as executives and other non-technical roles, who may understand the concept of AI intellectually but do not see it as a core part of their work.

At the other end of the spectrum, a much smaller group—approximately 20 million people (0.2% of the population)—are actively using frontier-grade LLMs in specialized domains like mathematics and coding. These users are witnessing a dramatic shift in how work is done, with complex projects that once took weeks or months now being completed with minimal human intervention. This group is experiencing the full potential of AI, where LLMs are not just tools but transformative agents that can build, translate, and decompile applications with ease.

Finally, a tiny fraction—around 5,000 people (0.00006% of the population)—have access to the most advanced LLM systems internally. These individuals are part of a select group that is witnessing the most profound changes in AI capabilities. They are involved in projects that push the boundaries of what is possible, such as collaborative software development, cyber defense, and even advancements in mathematics and science. These systems are capable of performing tasks that would have taken years of human effort, and the pace of progress is accelerating rapidly.

Karpathy attributes this widening gap to several factors. First, the impact of LLMs scales with the ambition, problem size, and time horizon of the tasks they are applied to. Second, the capabilities of LLMs are more pronounced in digital, verifiable, and economically valuable domains, which are the areas where reinforcement learning is most effective. Finally, access to these systems is highly uneven, with the general public largely limited to free-tier tools, while professionals and researchers have varying degrees of access to more advanced systems.

This disparity in access and perception is creating a situation where the majority of the population is still in the early stages of understanding AI, while a small group is already experiencing the full potential of frontier-grade LLMs. As the capabilities of these systems continue to evolve, the gap in understanding is likely to widen further, creating a significant divide between those who are actively using AI and those who are not.

**Source:** [Hacker News](https://twitter.com/karpathy/status/2109361546505966046?ref=stirup.si). Analysis based on publicly available information; no hands-on testing is claimed.