Still faster than most websites

But I actually think that one if the bigger arguments for single language models is the ability to have more languages. Im from Sweden, so I would like to have swedish on extremly high level, but I wouldnt like to have all other small languages on that level beacuse it would inflate the size. So, I actually think having multiple single language models, make it wider and deeper

Yes, I see the same thing. My working thesis is that if I can keep the codebase modular and clear seperations, so I keep the entire context, while claude code only need to focus on one module at a time, I can keep up the speed and quality. But if I try and give it tasks that cover the entire codebase it will have issues, no matter how you manage context and give directions. And again, this is not suprising, humans do the same, they need to break the task apart into smaller piecers. Have you found the same?

Yes. Spot on. The good thing is that it makes better code if modularity is strict as well.

I’m finding that I am breaking projects down into clear separations of concerns and designing inviolate API walls between modules, where before I might have reached into the code with less clearly defined internal vs external functions.

Exercising solid boundaries and being maniacal about the API surface is also really liberating personally, less cognitive load, less stress, easier tests, easier debugging.

Of course none of this is new, but now we can do it and get -more- done in a day than if we don’t. Building in technical debt no longer raises productivity, it lowers it.

If you are a competent engineer, ai can drastically improve both code quality and productivity, but you have to be capable of cognitively framing the project in advance (which can also be accelerated with ai). You need to work as an architect more than a coder.

But he's not wrong. Training + inference on free customers is the black hole here.

But they are not losing 100x on inference on high paying customers. Their biggest loss is free user + training/development cost

Talking about search on a static docs site, has anyone tried a static pre-generated search like https://lunrjs.com/ ?

Basically every Elixir package's docs include search based on Lunr, as it's included by default by ExDoc[1]. It's quite good.

[1]: https://hexdocs.pm/ex_doc/

We use at endoflife.date, and are fairly happy. We had to add a workaround to get some search terms working (because of how it does stemming, searching for BSD wouldn't show openbsd earlier): https://github.com/endoflife-date/endoflife.date/issues/4924), but it is still not 100% perfect: searching for fusion still doesn't get you coldfusion.

I use https://pagefind.app/ for search on my website. It’s really easy to add to a static site.

I use Lunr on allaboutberlin.com. It's simple and effective, but searching the whole content would have required loading more files than I was comfortable with. Therefore it only searches titles and descriptions.

There are probably workarounds, but it's the only limitation I can think of. Otherwise Lunr just works.

I been using this project (seems to be abandoned now but still works) https://stork-search.net/

Works great.

I think https://develop.kde.org/docs/ is using lunrjs for search, but downloading and parsing almost 2MB file of search data creates some hiccups on website during load.

i use flex search https://github.com/nextapps-de/flexsearch

vitepress seems to use https://github.com/lucaong/minisearch/ and the vitepress docs I have built and used are good.

This must be illegal in so many ways

The pyramid references in the link is from 1992, it even says so on the page. I think that going to war against the recommendations from 1992 feels a bit...dishonest?

Because there have been different FDA food pyramids since then. The one people popularized hasn't been the recommendation for decades

The 90s food pyramid lasted until 2005, so decades is just about correct. Then it was some myolate something or the other.

But people used the 90s food pyramid everywhere and that was the only one popularly known. The myplate stuff, I guess it wasn’t advertised well by the government, who knows.

Can you do it with signal?

Starting to use Opus 4.5 I'm reduces instrutions in claude.md and just ask claude to look in the codebase to understand the patterns already in use. Going from prompts/docs to instead having code being the "truth". Show don't tell. I've found this patterns has made a huge leap with Opus 4.5.

The Ash framework takes the approach you describe.

From the docs (https://hexdocs.pm/ash/what-is-ash.html):

"Model your application's behavior first, as data, and derive everything else automatically. Ash resources center around actions that represent domain logic."

I feel like I've been doing this since Sonnet 3.5 or Sonnet 4. I'll clone projects/modules/whatever into the working directory and tell claude to check it out. Voila, now it knows your standards and conventions.