This does suggest that subjects who are bought into and understand the purpose behind what they’re doing, and are attentive to how the specific tasks they’re doing tie into the bigger picture, are more likely to be actively engaging their judgement as they go. And subjects who are just trying to follow the tasks as given to them are sort of washing their hands of the outcomes as long as they’re following the directions (which is, ironically, causing them to fail at following the directions too).

A AI assistant would work more like Planet Fitness where the goal is to figure out how to convince you to keep paying them while using the facilities as little as possible.

A big part of that might just be steering you towards repos of existing solutions to the problem you’re trying to solve rather than helping you vibe code a solution yourself. Over time they’ll be able to accrue a whole pile of canned functions that’s all automatically documented and audited and it’ll be able to plug and play those rather than having to rewrite.

The security implications of this give me a headache to contemplate to be honest.

Yeah it should be factored in, but it’s a different set of implications for long term sustainability. They don’t actually need to test and optimize a new menu every day or week. If they decide to just stick to the same one longer they can get way more return from each dollar spent on development. It’s just that right now the rate of improvement you get with training is really high and nobody can afford to fall behind their competition.

I suspect that once the models hit a point of “good enough” for certain use cases companies will start putting R&D focus in other areas that may be less expensive. Like figuring out how to run more efficiently, UI/UX conventions that help users get what they’re trying to accomplish in fewer steps, various kinds of caching of requests, etc. So the cost to serve tokens over time should only come down, and will probably start coming down more rapidly as the returns to model training slow down.

That’ll probably be a while though, because each successive model tends to be a lot better than the last.

This is why they were freaking out about DeepSeek just taking the trained model weights and slapping an interface on it.

This is basically how social media was when you needed a computer to go online. You’d sit and sift through your feeds and there’d be message chains you respond to. You’re not really doing anything else while you’re doing that and you’re putting it out of mind once you step away. When Twitter first started getting big it was sort of a joke that people are talking to you while on the toilet. The idea that you were only ever half engaged with anything you’re doing was remarkable enough to be worth pointing out instead of taken for granted.

It’s just a lot more focused and intentional when you’re dedicating time and headspace to the task instead of “microdosing” on connection via a dopamine lottery. Even if you took away the ads and the interpolation of creator-content crowding out the connections with people you actually know, I think designing for an infinite scroll just inherently makes the thing less human-centered. It sets it up so you’re interacting with these atomized bits of ‘content’ rather than people.

There’s reasons why, because of what a company does, it’s able to offer you something for markedly cheaper than you’d get it on the market, so the arbitrage you can do by getting it at that discount may be worth it, like getting a free meal or drink when you work at a restaurant. But in most cases, money can be exchanged for goods and services and $1 of currency is always going to be more valuable than $1 worth of some commodity.

I think a lot of online publications could be modestly sized businesses with a very respectable annual revenue range in the tens of millions or even hundreds of millions, but they erode their own brand value chasing tantalizingly higher numbers that SEEM in reach but actually aren’t.

A lot of print magazines, like Vogue or even Field & Stream, are like 60% or more full page ads. But if you’re reading something like New Bride magazine you’re probably actively shopping for wedding dresses and flowers and such, so the advertising ideally works as part of what makes the magazine valuable for the reader and the advertiser.

The real problem is that the finance and business folks are addicted to performance metrics and they preferentially put their money towards things that can be represented as graphs because it’s hard to argue with a graph. Jon Gruber has a vague sense for what sort of audience he has and what they’re into, so he can pitch advertisers on the idea that by advertising with him they’re going to reach an audience of Apple enthusiast technologists who presumably care about design and UI/UX and whatever other intuitions he has about his readership. But none of that is a quantitative metric, so only a small market is open to putting money into it.

If the government just went on a building binge of housing to be sold at market rate, or even set an upper bound before qualifying to buy them at a middle class income, it’d work out fine. That’s basically how Singapore does it only they couple it with somewhat aggressive policies to encourage people to downsize their living situations once they’re empty nesting to free up family dwellings for people with families. We probably wouldn’t need to do that second part since we’re not a claustrophobic island, and could just count on natural turnover.

The instinct to see a bureaucratic system working poorly and resolving to opt out instead of fixing it is precisely why everything sucks right now. You can’t just work and vibe and spend your way out of living in an advanced industrial society.

The ones who are around the capital are the ones doing policy or back-office work that’d be the equivalent of a corporate HQ, and they have benefits from being concentrated in a region where there’s a lot of other Federal back office work because they can recruit from a talent pool of people who are experienced in the particular requirements of Federal work. It’s the same sort of agglomeration effects that drives finance to concentrate in NYC or the film industry to concentrate in LA.

Taking the USDA as an example, it doesn’t literally operates farm. The vast majority of what it’s doing is functionally insurance and financial services. The stuff that needs direct interaction with farms like inspections or scientific research are done out of field offices.

If America has an equivalent of The Capitol from Hunger Games where a bunch of absurdly wealthy and out of touch elites pull the strings on how the rest of the country lives for their own benefit, I’d submit Wall Street or Silicon Valley before I posit DC.

The stuff LLMs will democratize will be a lot more impactful than nice posters for car wash fundraisers though. So in that sense it will be different, but I don’t think it will crack the market for proficient experts in the field in the same way photoshop didn’t destroy graphic design and CAD didn’t destroy drafting. It may get rid of the market for a lot of the second-tier bootcamp grad talent though, so I wouldn’t be getting into that right now if I could help it.

I think I have a conversation at least weekly where I have to explain to someone that using an LLM to convert COBOL to Java (or whatever) will not actually save much effort. I don’t know how many ways to explain that translating the literal instructions from one language to another is not actually is not that hard for someone fluent in both and the actual bottleneck is in understanding what sort of business logic the COBOL has embedded in it and all the foundational rearchitecting that will involve.