Precision versus Heuristics

It’s well known that networking is important. All of the jobs in the last half of my career come from personal connections somehow, for example. And we all know that it’s important to have good networks, and also to watch for bias in how we form and use them.

Why are networks so important? Well, they’re a great filtering heuristic, for one reason. I am not the best {programmer, creative person, engineering leader, whatever} in the world, but I’m a known quantity to my friends. They have a good model in their minds of what the job might be, and a good model of me, so they can perform a fairly high-fidelity match up. Much easier than looking through lots of messy data in the real world.

But…LLMs are pretty good at looking at messy data, and they don’t get tired. So, it’s not super hard to imagine them getting involved in hiring decisions, at least at the level of filtering resumes. But we probably don’t want them just to do dumb, language-level filtering. We want them to have a rich model of what we need and to build a rich model of each person. This might not seem practical now but things I’ve seen and worked on make me believe it’s entirely possible soon.

So now we have a world where we can have a new kind of precision, at scale. We already have precision at scale when the data is regular (even if it’s just a little regular, like credit card transactions, we can do things like spot fraud at scale). The boundary of that precision is exactly where we have trouble ingesting and managing data at scale because of some irregularity or complexity. But LLMs are pretty good at dealing with extracting meaning and intent from natural language and other irregular forms. So that boundary is going to move.

What will that mean? Different things in different contexts, no doubt. There will probably be places we don’t like the additional precision and control (like an employer reading your emails for “tone” or “politeness”) but other domains where we will like it (maybe it makes hiring both better and fairer by enabling employers to look more deeply and thoughtfully at candidates. And maybe it makes employment better by helping you find a job that’s a perfect fit).

In some sense, compute is always about scaling up our ability to think in some particular way. LLMs give us some new ways in which we can scale that thinking. We’ve likely just begun to see the impacts of that.