It's always easier to blame the model and convince yourself that you have some sort of talent in reviewing LLM's work that others don't.
In my experience the differences are mostly in how the code produced by LLM is prompted and what context is given to the agent. Developers who have experience delegating their work are more likely to prevent downstream problems from happening immediately and complain their colleagues cannot prompt as efficiently without a lot of hand holding. And those who rarely or never delegated their work are invariably going to miss crucial context details and rate the output they get lower.
Have owned a couple iPads starting ~2010 -- mainly for reading pdfs, and comics in electronic form. Occasionally drawing / jamming some tunes - almost all via 3rd party apps. There are still plenty of decent apps in the ecosystem, even though their eventual obsolescence is as good as built in, and a lot of stuff I previously loved the platform for has now been gone for years with no replacement. Native apps have never been great at pretty much anything, with a notable exception of Garage Band which is an absolute banger for its money. Books is... passable I guess?
But the reading pdfs part is important -- and really hard to beat for me, the iOS drag/scroll/pinch/zoom UX perceived responsiveness is still unmatched IMO. It would take some real creativity beyond liquid glass to enshittify this aspect out.
I find the last part really surprising. That seems, and feels (I do have an iPad), standard to me. But markup is, to put it lightly, fucking terrible.
I lean towards the iPad's success mostly being brand name and advertising because I've never experienced the "just works". I understand that from the non techie people, but not when talking to nerds
Wish it was only the keyboard enshittified. Literally everything became worse with the update, I had to google how to turn off the silly transparency (Accessibility Settings -> Display -> Reduce Transparency) so that the battery that used to happily last for the entire day on iOS 18 does not die in a matter of some 4 hours. And don't even get me started on now-always-lagging home screen swipes and the Safari overhaul madness! Wanna close the active tab? That will be three taps, thank you very much. Oh, you want them taps to register _every time_, too? This basic phone UX used to be Apple's major USP over Android, now fewer and fewer reasons to stick to this ecosystem.
The subscription revenues is a decent chunk of your lifetime value (LTV) as a customer, but it's not all of it. The goal here is to squeeze as much value from you aside from that as possible, measured mostly by two things, really: the direct ad revenue, measured by dollars that go on the balance sheet, and the indirect "engagement" value measured by the KPIs (think daily, weekly, monthly active users) that go into the quarterlies. The more time you spend on the platform, the more "things" you have got used to interacting with (aka day-to-day, week-to-week "retention"), the more they can potentially "sell" to you -- and it's not just ads / youtube subscription upsells, it can be and often is other "products" on the same platform: their music streaming, their search, their documents and emails, maps, drive, etc. etc. And it just so happens that the short format is _really, really_ engaging for many folks.
The more time you spend in the mall, the fuller are the bags on the way out, be it out of chance, habit, or convenience.
That's right, but it's not just products that they can "sell" you. It's all about your data, which is worth much more than any upsell opportunity.
Whether the user pays for YouTube Premium or not, they still have access to your behavioral data, your interests, they can easily determine your location, and so on. All of these data points contribute to your profile, which is a literal gold mine for their entire business. How much value they extract from it exactly is likely something not even Google knows. But given that it can be exchanged on dark data broker markets in perpetuity, the price can only go up.
It's a goddamn racket that needs to be made visible and subject to thorough public and legal scrutiny.
> How much value they extract from it exactly is likely something not even Google knows. But given that it can be exchanged on dark data broker markets in perpetuity,
Companies like Google and Meta don't sell your data, on dark markets or otherwise.
They keep it in-house for advertising targeting purposes.
If they sold it to other companies it would reduce their competitive advantage. It's not even worth it for them.
Google doesn't want to sell your data. They want to keep it internal as much as possible so their ad platform is valuable.
Maybe you're right as I haven't seen the studies you're claiming prove this. I based my comment off my own experience using AdWords around 2013-2016 and then again around 2019-2021.
Whether they're directly doing business with data brokers or not is not the point. They're indirectly profiting from the profiles they build by selling access to them via their advertising platform. It's just a roundabout way of doing business, as is common in advertising.
Besides, even if they're not selling these profiles, they will end up on data broker markets one way or another. Whether their lack of security allows companies to export it, as in Meta's case, or simply by using their tools to gather as much information about people as possible.
The reality is that nobody outside of these companies, and likely only people in executive positions, knows how they operate internally. They have an army of PR and legal people to do their bidding. Whatever practices the public thinks these companies are or aren't involved with is mere guesswork, but one thing is certain: they don't maintain their size and power by keeping their hands clean. But then again, I'm probably on the wrong forum for this line of thinking.
> Whether they're directly doing business with data brokers or not is not the point.
That was literally the point I responded to.
> They're indirectly profiting from the profiles they build by selling access to them via their advertising platform
That’s very different than the “selling your data” line that keeps getting repeated.
There’s a motte and bailey game that gets played every time this topic comes up. The argument starts with claims they’re selling your data, then when that’s revealed as a false claim the argument pivots to something else with strained arguments that it’s equally bad.
> The goal here is to squeeze as much value from you aside from that as possible, measured mostly by two things, really: the direct ad revenue, measured by dollars that go on the balance sheet
There are no ads on a sub, this doesn’t make any sense as such to the parents comment.
You don't get ads on YouTube with a premium sub, your activity data (views, for how long, what topics, what times of the year, of the day, so on and so forth) is still collected, and appended to your profile, the same profile that is used by AdSense to show you ads around the rest of the web.
Jonathan Blow's The Witness is a notable example (minor spoiler alert)! Past a certain point in the game, it becomes REALLY challenging to just walk through the IRL woods without over-concentrating on things.
Great point! After playing that game I and a few friends were trading real-world photos of spots where we'd found examples of the in-game thing you're talking about.
In JetBrains editors it's possible to highlight mutable variables, at least in the languages where the distinction exists. My go to setting in Kotlin is to underscore all `var`'s, for two reasons:
- this makes them really stand out, much easier to track the mutation visually,
- the underscore effect is intrusive just-enough to nudge you to think twice when adding a new `var`.
Nothing like diving into a >3k lines PR peppered with underscores.
Mathematics trains a lot of skills that are generally applicable in engineering. Decomposing complex problems into non-trivial sequences of manageable steps, being able to prove that the design works, spotting appropriate invariants to build type hierarchies/abstractions around, communicating it all in an intentional and comprehensible way where each of the next steps follows from some of the previous, etc., etc.
Yeah, the `numpy` version still looks relatively cryptic (like, "line > 0" is still fine, but the numpy arrays broadcasting rules can quickly get out of hand) compared to the author's Javascript example, or any decent collections API in a typed "enterprise" language like C#/Java/Scala for that matter. Here's my personal favorite, a Kotlin version:
diffs.countIf { line ->
line.all { abs(it) in 1..3 } and (
line.all { it > 0} or
line.all { it < 0}
)
}
In my experience the differences are mostly in how the code produced by LLM is prompted and what context is given to the agent. Developers who have experience delegating their work are more likely to prevent downstream problems from happening immediately and complain their colleagues cannot prompt as efficiently without a lot of hand holding. And those who rarely or never delegated their work are invariably going to miss crucial context details and rate the output they get lower.
reply