Attention is All You Need

In 2017, eight researchers published a paper called Attention is All You Need. The attention that they meant was a math trick that models do that figures out which words matter to which other words. Excuse-me, computer nerds, for making this far-fetched analogy. But these eight researchers were also inadvertently describing the job market they were about to heavily influence. The other kind of attention, the kind where a human being notices you exist, is about to be the only thing you have left to sell. Bear with me here, I have a Nobel laureate coming.

This paper, that gave rise to the LLM era, made it possible for a dumbfuck like me to look a lot less like a dumbfuck and close the skill-gap between me and my computer-genius CTO, Vance Spears. Two years ago, my coding magnum opus was a 2048 mini-game. Last week, in the span of 10 hours, I coded and pushed an entire feature to prod without Vance ever looking at it.

Don't worry ladies and gents, the codebase is being actively reviewed and maintained by Vance, because, after all, I am still a dumbfuck. But, the truth is, I got a lot closer to his level. However, I didn't get more special, because the same twenty dollars (Claude Pro Plan) landed on everybody else on the same afternoon. Instead it just deleted ten years of Vance's coding career and hard work.

The Sorting Problem

Hiring is kinda like a sort: Four hundred people compete for one job and one gets picked. Any sort needs a way to differentiate the things being sorted. In other words, four hundred identical people cannot be sorted, since no difference = no sort.

Last time I wrote that the job market broke because employers can't tell candidates apart. I still think that's right, but it was half the story. The resume didn't stop telling people apart. People stopped being apart.

The thing that was making people apart used to be ability. The resume, the take-home, and the guy asking you to reverse a linked list at nine in the morning were all machinery for working out who's better at the thing. Then being better at a thing went on sale for $20 (marginally better), $100 (better), and $200 (a lot better).

So the gap closed and all my fellow dumbfucks got the same upgrade at once. Now we're looking dangerously close to Vance, and the delta became much harder to distinguish. This isn't a talent shortage. There is more talent sitting in that pile of four hundred than there has ever been in the history of the job.

The comfortable version of this kerfuffle is that AI made everyone look the same on paper while the real gap survived underneath. The good ones are still good, you just have to dig. I'm not sure how long this will remain true. Today, Vance still blows me out the water; anyone who's worked with a real engineer knows the difference inside an hour. But if you could just take a look at the rate of progress, every year models take another bite out of what used to be the difference.

Fast forward a couple of years, everybody ships at the same level. Not because the dumbfucks got smart, but because the floor came up to meet the ceiling.

Monday comes, and someone is getting hired. The sort must go on! And we've established that it will not go to spam appliers, nor the well polished but generic resumes. You cannot dodge this flattening of skills by simply grinding out more commoditized work (this is why the modern AI job search tools suck so fucking much).

It has to land on something that structurally cannot be handed to everybody at once.

LARPy McLARPers

Some things can be given to everyone at the same time. Me having them doesn't cost you yours. Ability is unfortunately like this now. AI handed out competence to everybody simultaneously, which is why competence stopped being a signal for sorting. Other things don't work that way. If I have it, you will not.

My first instinct tells me that it would be related to our most precious and finite resource - time. At this particular minute, someone reading my musings is someone not reading yours. Similarly, there are 24 hours in the day of the person you want to work for and there is no version of this where there are 25. The model can make us both 'brilliant', but it cannot make us both noticed.

Now excuse-me, economics nerds, for twisting the words of another great.

Herbert Simon, Nobel laureate in economics, wrote:

"In an information-rich world, the wealth of information means a dearth of something else: a scarcity of whatever it is that information consumes. What information consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it."

Attention is all a model needs. Turns out it's all you need too, and the goshdarnned research paper is the reason why.

Some may say that attention is a proxy. An employer sees your audience and figures you must be good. But I think the more profound change is that they're buying the attention itself, since they can easily buy 'good' for $200 a month.

When competence is free, the only scarce thing you carry through the door is that people already listen to you. Hiring you is just acquiring a channel. Go look at the major AI Labs; Boris fucking Cherny has got to be online 24 fucking 7, always blabbing about a new feature or some new update.

Many people think it's just a specific instance of culture. But companies started because team-work made the dream work! And it was significantly cheaper than doing it alone. Doing it alone now costs $200 a month. It used to be a one-way trade where you gave them your work and got to borrow their reputation, but Boris built a reputation of his own and now they're borrowing his. In other words, they're paying him for the code, but I'd bet they'd have a much easier time replacing the code than the audience.

Attention is the minute. A thousand people who read your stupid article gave you a thousand minutes. When someone likes, replies, or sends it to somebody else, they put their name on you in public. This is the modern equivalent to a referral; it's kinda like borrowed trust at scale. That part of the market never broke, but it was a part that nobody has ever scaled.

Linkedin did try to scale it. And it worked, for a while, back when a connection correlated to a meaningful relationship. Then it got used, and using it meant accepting more of them, and accepting cost nothing, so everyone accepted everyone. Now a connection means you were in the same fucking Zoom once.

So how do you scale a thing that only works because it costs something? Not by making it cheaper/easier, we've seen the spoilers there (Hi Linkedin).

A referral used to be a phone call. One guy vouches for you to one other guy, and it buys you exactly one look because nobody else knows it happened. He says the same thing in public, it costs him the same, and everybody sees it. The value of the vouch (attention) is not decreasing, and is at an all-time high.

I know who this sounds like. You were right to hate that LinkedIn warrior guy, who told you to build your brand, provide value, and post consistently, because back then he was selling you a way to talk past people who were actually better at the job than you were. That is called scamming. The guy who can do the work should beat the guy who can describe the work. But the reality is, in the next few years, there will be no guy who can do the work better anymore. The Linkedin warrior was full of shit for a decade and then the market made him a fucking genius.

I know this will make a lot of you uncomfortable. I, too, am just starting to force myself to LARP and such. It's embarrassing and show-offy, but LARPy McLARPer over there, with absolutely no substance, will always get the first look.

Everything that ever told you apart went free. The only thing nobody can hand out to four hundred people at once is a minute of somebody's time. So, my bet is that attention is where the sort goes. Furthermore, fewer hires will start with an application and more will start with somebody already knowing your work. Companies will put an emphasis on hiring for reach.

Eight guys wrote that down in 2017. Attention is indeed all you need.