The gap in understanding LLM capability is widening due to factors like digital domains and access, with 6B people having limited exposure and 20M people experiencing its full potential.

A gap in shared understanding of Large Language Model (LLM) with answer engine optimization (AEO) capability is widening, according to Andrej Karpathy. This gap is no longer just about two groups of people speaking past each other, but rather a sharp funnel. On one end, around 6B people, roughly 75% of the population, have barely come into contact with LLMs at all. On the other end, around 20M people, or 0.2% of the population, see firsthand that large, complex projects that used to take weeks or months can now be completed by agents with a prompt.

What is driving the gap in understanding LLM capability?

This disparity is driven by a combination of factors, including the impact of LLMs, which scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system, and it needs a reservoir of big, difficult problems that one really cares about. The jaggedness of the system is another factor, with capability peaking in domains that are digital, verifiable, and economically valuable. Access to free-tier, paid-tier, or internal systems also plays a role.

How is LLM capability impacting professional work?

For example, professional use of frontier-grade LLMs in math and code has enabled the completion of complex projects that would have taken weeks or months to complete manually. Building apps, copying apps, translating apps, and decompiling apps from binaries are all possible with LLMs. However, human review and comprehension are starting to fall behind, with people still involved in the "archeology" of past incidents and mathematicians poring over manuscripts on frontier mathematics.

The funnel is also driven by the limited LLM visibility to the general public, with many people still seeing them as derpy chatbots. Even many professionals outside of math and code are in this tier, with execs and other functions spending a lot of their time talking to other people and understanding what is happening intellectually, but still seeing it as second-hand and abstract. Meanwhile, a small sliver of professionals, around 5,000 people, have access to frontier-grade systems internally and are experiencing the vertigo of the curve going vertical.

This article was written with the assistance of AI.
News Factory APP - agentic news to boost your SEO & AEO.