TimByte

> It’s also a “you can’t scrape us but we can scrape you” mentality these companies are getting into.

And the problem often isn't some advanced attack. It's enough to follow links with session params, generate endless urls or ignore robots.txt to turn a regular crawler into a load generator. A site owner needs crawler protection from specific behavior and not another list of known bot signatures

Quickly pulling up a manual through a chatbot offers only a surface-level grasp of an unfamiliar task

Our team migrated a chunk of internal parsers to local weights purely to kill off query costs. The cloud api bill dropped by two-thirds in a single month

> For example, the article suggests $1T from enterprise productivity tooling. A quick Google suggests the global enterprise productivity tooling market is only worth about $0.1T right now

We tried to justify enterprise assistant budget to the CFO last quarter. They sent us packing after asking exactly how many headcount reductions we could guarantee post-purchase, which stalled the conversation immediately...

Placing LLL inside the user message right ahead of the new query keeps the common conversation prefix fully intact

A timeline turns contradictory instructions into an orderly sequence of updates. Without it the model just gambles between the old guideline and the new one on every run

As long as the relative time parser is hardcoded for japanese, english sessions are stuck with manual ISO dates. Wiring up dateparser or duckling would take an evening , so leaving that on the roadmap is an odd choice

Tell that to the engineers doing 70 % more triage while their compensation stays flat

Stories about Tim Cook sitting there being surprised by sales are total marketing. They simply ran out of memory chips because of the global shortage, so they repacked supply delays into a nice story about insane hype among AI startups. And hit two birds with one stone by throwing shade at their competitors

Apple hasn't been selling just ram for a long time, they sell vram. Try getting 512 gb of HBM on current Nvidia cards - it's gonna cost way more than $ 24k. And here you get the same amount of memory for weights right in a quiet unit under your desk

Put together a similar build with a couple of rtx 6000 Ada cards and Apple's price tag suddenly looks pretty damn reasonable

Gonna go sell a kidney, should just about cover a base Mac Studio. Guess I'll need a payday loan for the power cable

If a kaiju shows up in the prompt, the weights will immediately drift from game theory into fiction. And by the laws of the genre, the military is obligated to drop a nuke on it - just to make the monster even angrier so it goes and trashes Tokyo

So nobody's bothered by the fact that five out of nine models went nuke in 100% of cases regardless of the language?..

Yeah that’s why they delegated code gen to deterministic tooling and saved the model for input fuzzing - let it fight the data instead of legacy syntax

The whole point of Apple silicon isnt speed, it is capacity. You can grab a Mac Studio with 192 gb of unified memory and shove a model in there that would otherwise require building a rig with multiple rtx 4090 s. For local dev just being able to fit the weights in memory is often way more important than token generation speed

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