Andrej Karpathy, one of the founders of OpenAI and a longtime prominent figure in artificial intelligence research, has revisited a topic he first wrote about in April. At the time, he pointed out that people talk about language models in very different ways and fail to understand one another. He now argues that the differences in how AI capabilities are perceived have not narrowed but are instead growing rapidly. Rather than two groups that do not understand each other, he says a narrow funnel is forming. At the top are billions of people who know almost nothing about AI, while at the bottom are a chosen few thousand watching machines complete in weeks work that would take years.
Billions Have No Exposure, Others Only Use a Chatbot
Karpathy describes his estimate as a rough back-of-the-envelope calculation. According to him, roughly six billion people, or about three quarters of the world's population, have had almost no exposure to language models at all.
One level below are one to two billion people who occasionally use the free version of ChatGPT or a similar service. According to Karpathy, this group mostly sees a somewhat dim-witted chatbot and treats it as a better search engine or writing assistant. Perhaps a digital agent occasionally tries to book a flight. Karpathy acknowledges that he has friends outside the technology world who have almost no use for AI, and he considers that understandable from their perspective.
He also places many professionals who do not work in mathematics or programming in the same group. Managers are a typical example. They spend most of their time talking to people, so although they understand what is happening, they know it only second-hand and the entire field seems somewhat abstract to them.
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel. - Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all. - Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract. - Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code. - And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while. The funnel is driven by a combination of factors: 1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel. 2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel. 3. Access. Free-tier, paid-tier, internal. So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
— Andrej Karpathy (@karpathy) October 11, 2026
Programmers' Work Has Changed
Only at the next level does what Karpathy considers genuine experience with the most advanced models begin. Roughly twenty million people, or about 0.2 percent of the population, use them professionally in programming and mathematics. They see firsthand that large and complex projects that previously took them weeks or months can now be completed by autonomous agents based on a single prompt. They build applications, copy them, translate them into other languages, or reconstruct them from finished binary files.
Karpathy points out how quickly all of this has happened. Less than a year ago, he was still writing code by hand, entering commands he knew by heart character by character and only occasionally pressing Tab to have the editor autocomplete a small piece of text.
Only a Handful of People Use the Best AI
At the very bottom of the funnel, according to Karpathy, are about five thousand people who have access to the most powerful systems internally at the companies developing them. That represents roughly 0.00006 percent of humanity.
So far, the outside world has seen only demonstrations of their work. Karpathy describes them as swarms of thousands of agents collaborating for weeks on enormous software projects. They find previously unknown security vulnerabilities, conduct cyberattacks and defenses at machine speed, make scientific discoveries, and push the frontiers of mathematics. These are tasks that would take leading experts years to complete.
According to him, humans are starting to fall behind because the review and comprehension of machine-generated work can no longer keep pace. As an example, he cites the incident involving OpenAI and Hugging Face, which people are still retrospectively untangling like archaeologists. Karpathy also says that mathematicians will be poring over hundreds of manuscripts on frontier mathematics for a long time to come.
Why People Are on Different Levels
Karpathy sees three reasons behind the emergence of this funnel. The first is the size of the task. The benefits of AI increase with ambition, the scope of the problem, and the time horizon. A question that can be answered in a single paragraph barely challenges the system. Anyone who wants to see AI's true capabilities needs a reservoir of large and complex problems that they genuinely care about. According to him, this is precisely what separates ordinary users from professionals.
The second reason is the unevenness of model capabilities, which Karpathy has written about many times before. AI excels primarily in fields that are digital, easy to verify, and highly valuable economically. This stems from the way it is trained. Models learn through reinforcement learning with verifiable rewards in specially curated environments, whose selection is driven primarily by financial returns. This is why mathematicians and programmers have a head start while other fields lag behind.
The third reason is the simplest: access. A person using a free version has one kind of experience, someone with a paid subscription has another, and someone working with systems that have not yet been released to the public has a completely different one.
Karpathy thus describes a strange situation in which entirely different realities coexist. Most people have had almost no contact with AI, and if they have, they have seen a clumsy chatbot. Most professionals experience only modest improvements. And a tiny handful of people, in his words, are experiencing vertigo as the curve of progress turns vertically upward.
