TimByte

It would be interesting to measure not just the accuracy, but how many people actually decided to double check the answer

Obviously without a proper RAG pipeline or web search it is just going to hallucinate with maximum confidence. It would be more interesting to see the results with top tier models that have proper alignment to refuse to answer when token probability is low. As it is they just proved that people tend to trust well written text in a chat ui

The interesting taxonomy may be less about the source of the names and more about what properties they have

Somewhere in Cloudflare's DNS system, a random number generator briefly became the office matchmaker

A surprising amount of infrastructure design seems to be hiding patterns that humans will otherwise turn into unsupported features

The web has its own problems, but at least bad HTML is often still visible enough to repair. A custom canvas-like desktop UI can be basically a black box to assistive tech

Yeah, blind seems unavoidable there, since it's literally about a blind client

The "invisible" point is interesting, although I'm not sure I'd read too much into it. Visual metaphors are so baked into English that avoiding all of them can start to feel a bit performative unless the wording is actually excluding someone

Sidewalks have to be usable by people who can't hear, people moving slowly, kids, older people etc. If a cyclist is on a sidewalk and can't safely pass without the pedestrian reacting instantly, they're the one creating the problem

The "read only" example is such a good illustration of how accessibility problems often aren't one big broken thing, but a bunch of tiny reasonable-seeming decisions stacked together

The most interesting number is missing here, and that is the token distribution by use case. If 60-70% was eaten up by PDFs, agents and automation instead of people actually sitting in Claude Code, then it is a completely different story

Maybe the real lesson is that public education has always had both impulses: emancipation and formation on one hand, conformity and state needs on the other

I think both can be true, depending on which layer of the system you're looking at

That may be why the rhetoric of education stays lofty while the day-to-day machinery often feels much more pragmatic

Mandatory education is still probably better than the alternative but it does seem to create a constant tension: the system has to serve students who want very different things from it

The funniest part is that the article about AI slime polluting the internet was itself generated from start to finish by a neural net to promote some startup.

But the problem they're so clumsily trying to monetize is absolutely real. GitHub is rapidly turning from a place with battle-tested solutions into a dumpster fire of hallucinations. And no crutches like MAREF are gonna fix that because platforms profit from showing growth in repo and commit counts even if it's all dead plastic code.

This is a good reminder that "green" does not have to mean perfectly manicured

I don't think the point is that London literally lacks pavement trees. As you say, the London plane is almost part of the city's visual identity in many areas. The interesting thing to me is how uneven the experience can be

The green space exists, but access to it is often something you have to deliberately route yourself through

What struck me is that the "three trees from your window" part sounds almost trivial until you actually test it

To be fair, you're comparing different use cases

It depends on the type of task and the developer's level. For typical CRUD tasks, I'd say AI is often already faster and better than an average developer. But as soon as ambiguous requirements or architectural trade-offs appear, the situation changes

Speed itself also changes the equation. If I can try out five implementation approaches in the same time I used to spend writing just one, the overall quality of the project could very well improve.

If programming were just about typing text, the profession would have disappeared already. In practice most of the work involves understanding requirements, analyzing constraints, finding trade-offs, and fixing problems before they hit production. AI is still noticeably worse at all that than it is at generating code.

At the end of the day it's just a tool. If someone enjoys building things, AI expands their capabilities rather than taking them away. Right now one developer can put together what used to require a small team. To me that sounds like an argument for learning it, not against it

Agreed. Shorting requires timing the market perfectly, but buying assets after a crash just requires having liquidity and patience. Historically, that's often been the safer play

After the dot-com bust, infrastructure assets turned out to be one of the most undervalued asset classes. Maybe in a few years people will look back on GPU clusters the same way they looked at dark fiber back then

I think a lot of people are mixing two different ideas: AI is overhyped and AI company stocks are about to crash. The first might well be true, but the second might not be. The dot-com bubble burst too, but the internet didn't go away. The question is more about who will be left standing once the era of endless capex growth ends.

There's no practical point to this. System calls and hardware interactions are deterministic. We don't need a probabilistic model to efficiently handle interrupts or write blocks to disk. Ai is great for the user interface layer, but at the kernel level, hardcoded C or Rust will always be faster and more reliable

The only viable scenario for that kind of ji approach would be a profiler that analyzes load patterns and recompiles kernel modules with optimal compiler flags for a specific use case. Dynamically generating kernel code from scratch would kill the system with compilation and ast verification overhead

That's the whole point of an operating system - to manage local hardware. That's its reason for existing. The moment task switching starts depending on rtt to a cloud provider's servers, it's just a thin client with delusions of grandeur. Fifty milliseconds per system call, and no amount of tflops on the backend is going to save you from that feeling of working through a proxy in 2005

If you're on macOS, take a look at Orion from the team behind Kagi Search. It runs on WebKit, is really light on battery usage, doesn't come bundled with crypto stuff or AI agents, and still supports Chrome and Firefox extensions natively

I think a lot of people aren't actually against AI itself. Personally I just want to choose when I need a chatbot and when I want a normal list of links. Over the last few years, that line has started getting pretty blurry

Your second idea is pretty much the direction the classic solution takes: walk along the curve incrementally, avoid square roots, and use symmetry to fill in the other parts