Talk notes · 26 August 2026
What AI is, how it learns, and why the cost of knowing things is collapsing — explained at the kitchen table, to a fourteen-year-old who wanted to know which jobs are left.
The cost of knowledge is trending to —
Illustrative, not real model output — but this is the whole mechanism.
Premise
That is the plainest definition available: our attempt to make software act as humanly as possible. The old way of building it was rule-based — when A happens, do B — with every branch written by hand by someone who had already thought of the case.
The new way reads natural language. The machine interprets what you mean rather than executing what you typed. That single change is what makes everything after it possible.
Mechanism
A baby learns to say yes to milk through repetition. From then on the brain runs one calculation constantly: if I do X, how likely is Y? Reward, punishment, repeat.
Machine learning is the same shape. Feed a system enormous amounts of written text — Reddit, blogs, forums, documents — and it builds probability chains: this word is most likely followed by that word. A large language model is that process run across essentially all written language. Whatever you put in, what comes out is the highest-probability answer.
The parenting parallelChildren are shaped by rewards, punishments and repetition whether anyone intends it or not. Homeschooling just makes the choice deliberate: you decide what to instil. Training a model is structurally identical — the same conditioning, without emotion or soul.
Consequence
The machine answers questions, organises data, explains complicated things and writes code. The interesting part is not any one of those. It is the price. You no longer need to hire the programmer, the researcher or the subject-matter expert to get moving. US$20 a month now stands in for decades of accumulated specialist knowledge.
Four accountants. Forty years each. Eight hours a day.
Work a model can now do — the picture that makes the point
Both things are on an exponential curve at once: how much knowledge is available, and how much labour is available to act on it. Most planning assumes those two stay scarce.
Response
That was the question from the fourteen-year-old, who is mostly interested in making money. Jensen Huang's answer has been: be a plumber. Physical work, done in a specific place with your hands, is the hardest thing to automate.
The more useful answer is not a job title. He is already building a dropshipping app with Claude — pulling manufacturer data out of China, ranking it by price and value, putting the good stuff behind a subscription. The tool is not the constraint any more. Knowing what to build with it is.
The habit
Always challenge, modify or question what the machine gives you. The highest-probability answer is not the same thing as the right one — and telling the difference is now the job.