Learning AI, without being left behind

There is a version of this post that starts with the technology. I do not want to write that one. Whatever you think of its merits, or of the morals of the companies selling it, AI has been a disruptive technology. I do not mean that it is good. I mean that it changed things.

For some people it meant losing a job. For others it meant the ground moving under a role they spent years getting good at. Underneath that sits a harder question.

If that is not what I do any more, what is?

For plenty of people the pain was not the technology at all. It was the culture around it: the dismissive tone, the certainty, the hint that anyone with doubts was not clever enough.

I am not going to name anyone, or point at a group and score a point. The technology and the behaviour of some of its loudest fans are different things, and I have no interest in mixing them up.

If you were the one being dismissive, the door is open. The way through is to accept some reasonable adjustments to bullet train speed. I promise that while it is harder, it is more rewarding to lift people up. Nobody has to stay out in the cold, or be told — as a colleague of mine memorably put it — to “get on the bus, or go under it”!

The deadline that never arrived

At several points since the first wave, we were told to get on board now or be left behind. Learn the tools, accept a new label, or be out of date by the end of the quarter. Some of that push was honest enthusiasm. Some was marketing. Some was a way of making other people feel small.

The people who did not rush have not been left behind. The tools changed. The basics did not move nearly as fast as the warnings said. We still argue about the same things: clear names, tidy boundaries, what happens when something fails. Whether the tests can really fail. Whether anyone will be able to read this in six months.

My honest opinion is, that I think nobody in this sector is more than six months of steady work away from being excellent at today’s standards. Six months of deliberate practice closes most of the gaps that get described as structural. The standards move. They just do not move that fast. What moves fast is the noise, and noise is not a course.

What it actually took off

Two things have changed, and they are worth being exact about.

The first is the training wheels. Learning a new subject used to begin with a long, flat stretch of hard work before anything interesting happened. Install this. Learn the words. Work out which guide is out of date. That stretch is still there, but it is much shorter. You can now reach the interesting part in an afternoon, with something real to push against. That is a gift to anyone who learns by doing.

The second is the floor. The worst first try at anything is better than it used to be. That is not the same as saying the best work got better, and I want to be careful not to claim that. AI has not turned us into experts. It has raised the floor. The gap between a beginner and a competent person is now mostly about judgement, not about how fast you can produce a first draft.

The promise of machine learning, without the expert climb

For a long time, machine learning was something you admired from a distance. It asked for maths, and years of patience with notation, before you could do anything useful. Much of that promise is now within reach of people who are not going to take that detour. You can sort things, pull facts out of long documents, group similar items, and sum things up without building the whole machine yourself. You still need to know where it goes wrong. But you no longer need the full degree to begin.

AI as a minion

The most useful model I have seen is also the least glamorous: a minion. You give it an errand. You check the result. You keep the judgement. A delightful head of legal I worked with used it exactly this way, and it was a joy to watch. Not as an oracle. Not as a replacement for their expertise. As a fast, tireless junior who would draft the first pass while they did the thinking that mattered.

I want to be clear that this is not a consolation prize. For a deep expert, the minion model is a valid place to start, and often the best one. You already know what good looks like. You are renting speed, not knowledge.

Chatbots are not the use case

If the only picture you have of AI is a chat window, you have the least imaginative version of it, and you are probably bored of the demos. Chat keeps a person in every step, and people are slow. The interesting uses are quieter and less showy. Sorting a pile of requests. Pulling the useful parts out of documents nobody wants to read. Checking that two lists agree. Spotting when something stopped behaving the way it did yesterday. Work that repeats, where a person or a program can check the answer, and nobody should miss doing it by hand.

Vibes are a starting point, not a destination

Vibes get you moving, and that is worth something. The first time a prompt produces something useful, it feels like magic. That feeling is why people keep going.

The trouble is stopping there. If you only follow vibes, you cannot say why something worked. You cannot tell when it quietly stopped working. You cannot explain it to the person who has to look after it. And refusing to go further is usually fear dressed up as confidence. The hard, unsolved part makes you feel foolish, so it is easier to say the hard part does not matter.

To anyone running on vibes alone: the difficulty is not proof you are doing it wrong. It is proof you are doing it. This work is hard, much of it is unsolved, and that is the fun of it. A field with no open questions would be a job, not a craft. Going deeper is also the kindest thing you can do for the people around you. It turns your enthusiasm into something they can lean on.

Where to start

Nowhere dramatic. Pick a subject you already care about. Give the model an errand, not a career. Check the result the way you would check a junior’s work. Then do it again with something harder, until you find the edge of what it can do. That edge is where the learning is. Six months of that, done on purpose, is roughly the whole course.

The deadline was never real. The interest is.