saved
The LLM capability gap is now a sharp funnel
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
Andrej Karpathy revisits his April post on the AI understanding gap and argues it has widened into a funnel. Most people barely touch LLMs, casual users see derpy chatbots, about 20M professionals in math and code watch agents finish months of work, and roughly 5,000 insiders see agent swarms advancing science and security.
ideas
- Four tiers of exposure. Karpathy's napkin math: about 6B people with little contact, 1–2B casual free-tier users, about 20M professional frontier users in math and code, and about 5,000 people with internal frontier access.
- Ambition drives impact. A question with a paragraph answer barely stresses the system; the gains show up on big, hard, long-horizon problems people care about.
- Jaggedness shapes who notices. Reinforcement learning on verifiable, revenue-driven environments makes capability peak in digital, verifiable, valuable domains like math and code.
- Access is the third axis. Free, paid, and internal tiers see very different systems, and insiders already see review and comprehension falling behind.
quotes
“It's now less "two groups of people speaking past each other" and more a sharp funnel.”
“Capability peaks in domains that are digital, verifiable and economically valuable.”
“And a small sliver of professionals are experiencing the vertigo of the curve going vertical.”