Andrej Karpathy AI Perception Gap Is Becoming a Sharp Funnel
The OpenAI co-founder says fewer people understand what modern language models can do as usage splits into distinct tiers.

A Widening Divide in AI Understanding
Andrej Karpathy, co-founder of OpenAI and former Tesla AI director, argues that the gap in public understanding of large language model capabilities is growing wider in 2026.
In an October 11 post on X, Karpathy revisited an observation he first shared in April, saying the divide is no longer just two groups talking past each other but has become a "sharp funnel" with billions of people at the top and only a handful of insiders at the bottom 3.
The Four-Tier Funnel
Karpathy divided the world's roughly eight billion people into four groups based on how they encounter AI, calling these rough estimates rather than survey data 3.
At the top of the funnel are about six billion people, or roughly 75 percent of the population, who have little or no direct contact with large language models 4. Just below them are one to two billion casual users, mostly relying on free versions of tools like ChatGPT as better search engines or writing helpers .
Further down, around 20 million people, or 0.2 percent of the population, use frontier AI systems professionally for tasks like math and coding 3. At the narrowest point are roughly five thousand insiders who have direct access to the most advanced frontier-grade systems .
Why It Matters
Karpathy said a person's beliefs about what AI can do now depend heavily on how much they have used it, how recently, in which area, and at what level of access 3. This means two people can look at the same technology and come away with very different expectations.
The post spread quickly online, with summaries appearing within a day and discussion on platforms like Hacker News 3. Secondary outlets reported around four million views and roughly 33,000 likes, though engagement numbers do not prove the underlying claim is accurate .
By describing the gap as a funnel rather than a simple split, Karpathy is highlighting that understanding AI capability is becoming concentrated among a small, increasingly specialized group while the majority of people remain far removed from what state-of-the-art models can actually do.
Key facts
- About six billion people have little or no direct contact with large language models
- Roughly 20 million people use frontier AI systems professionally for math and coding
- Only about 5,000 insiders have direct access to frontier-grade systems
- Karpathy described the perception gap as a sharp funnel rather than two groups
- The gap is driven by differences in usage amount, recency, domain, and access level
Seen on social
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… https://t.co/79xGhQdf1n
— Andrej Karpathy on X, see the post
Sources
- Andrej Karpathy Says This AI Gap Is Widening - Business Insider — businessinsider.com
- Andrej Karpathy says this AI gap is widening - Business Insider — businessinsider.com
- Karpathy: LLM Capability Gap Is Widening in 2026 | is4.ai — is4.ai
- Andrej Karpathy says this AI gap is widening - DNyuz — dnyuz.com
- Karpathy's LLM capability gap is widening, who is right — pasqualepillitteri.it



