Alignment Machines

AI has a use case I think will rise more and more: alignment. After inputting enough information, it will eventually know you enough to better find things that fit you. An example would be finding a job or finding a company — but it goes well beyond that.

Matching a person to a company, a school, a person or a task is an alignment problem, and AI is now positioned to solve it because it finally has the missing input: a real model of the person. That model is not built by a 27-page interest form. It's built by everyday conversation — the things you ask an AI because you actually have a problem with them: work, mental state, values, friends. Patterns form out of that, and pattern recognition is what these models are best at. Millions of people are already feeding AI memory as a byproduct of use. That's the enabler.

The request I imagine: "Research the top 10 companies in this region with the highest fit to my personality — and I want hands-on people, not thinkers." The AI already holds the context for "my personality" before the search begins.

Yes, a careers page is written to attract. But there's a huge marketing component everywhere, and you still find real alignment in it. One company foregrounds a 4-day week; another says it's hyperambitious and wants to build the coolest product in the world. Both are marketing — and two genuinely different kinds of alignment, selecting different people. An LLM can read 500+ career pages, values statements, and employee socials; volume of shallow-but-real signal beats a handful of verified data points.

How deep you go is a parameter, not a fixed design: job marketplaces, then company career pages, then ratings portals — mostly unhappy people post there, so a second layer, never truth — and then the big one: people. Employees, their posts, their socials. Can I deal with this person? Do they do sport? Do they rage about the same things I do? That's a question about people, not roles — and it tells you the most.

One trade-off is open: the more context, the better the alignment — but old context lies. Five-year-old chats hold positions I've entirely changed my mind on. Volume trades against staleness, and I haven't settled where the line sits.

Karrierekontext — the concrete form

My first attempt asked the AI to do everything at once — "here's me, find something good" — and that was too general to work. The fix: curate the portals yourself, by hand. Then hand the AI a triple — your written context document, one portal, one profession — and let it work through the results step by step, sorting by your criteria: values, size, location, remote, working hours, even politics.