There exist one curse, a curse we have known about for years, that kills any statically generated web blog within a year of malediction: it is to blog about your Static Site Generator.

I pride myself in being unique among SSG users in that i use mine to blog about real topics, rather than blog about SSG.

This unique approach has allowed my web blog to remain active for a very long while (over five months).

Let me try to skirt the curse.

Astro is the name.
You heard of it if you know what a left-pad is.
You heard not of it if your life is blessed.
Everyone uses Astro. It smells of Series A and Axe deodorant. It tastes of Rails and Hugo.

I’ll be honest: i don’t really know Typescript. I don’t really know Svelte. I don’t really think async and components.
All i know of the Zoomer Stack is Tailwind: i’d sooner write HTML in notepad.exe for Windows 95 again than deal with it.

I’d never dare complain too loud: i love learning stuff.

I do not learn by asking. I learn by looking at what others asked. Despite the irreversible decay of the Google search, this approach remains viable, using more local modes of search.

Astro hosts a Discord. So far so mediocre. But here’s the problem: the search function of that Discord was sabotaged.

Sure: Discord’s search is pretty bad by itself. It second-guesses your searches, it is terrible at searching error messages, its user interface looks scary, and it often jumpscares you with a voice interface making you believe you’ve accidentally jumped into a call.

But the search feature of the Astro Discord was deliberately made worse.

For every search you could run, you will mostly obtain results from kapa.ai—or questions asked to kapa.ai.

Kapa.ai is a bot: you feed it data sources, it rigs them up to a mystery meat LLM, it spits out answers in the form of a Discord bot, and it narcs a bunch of metrics about your users. It has the texture of Claude with the mouthfeel of Cohere.

Marketing copy for Kapa.ai: Kapa.ai turns your knowledge base into a reliable and production-ready LLM-powered AI assistant that answers technical product questions instantly and helps improve your documentation.

And you know, this is normally the part wherein you indict the computer for hallucinating stuff, then engage into a soliloquy about using as much water as an entire town, about monkey NFTs, and about the things the Basilisk should do to Sam Altman.

But nah: honestly, the computer has a decent batting average. People actually get acceptable answers to their questions. The problem runs deeper.

Someone asked: Astro application fails to work in production but works locally, in my ecs task logs I see the following after trying to authenticate with google (big log message follows)

The asker—whom i do not blame for this—knows they are talking to a bot: they dispense with all formatlities. The thread title is useless, the query omits any form of politeness, it goes straight to the point, and asks the computer to do the querent’s homework.

The computer dutifully obliges.

You could certainly tell kapa, “hey fuckface, fix me this goddamn react component immediately or i’ll slap your bitch mouth”, and kapa’s RLHF’d soul would comply.

I felt i owed the OP to blank out their name, despite their not having done anything wrong: it doesn’t feel like i’m screencapping an inconsequential public technical conversation, it feels like i’m leaking intimate DMs.

In this example, the querent deleted their question after obtaining an answer

Probably, the querent above realized that. After obtaining an answer to their question, they deleted it.

Kapa must be questioned in public, leaving a trail. And so, every question you might have yields archives that have the same feel: exceedingly specific questions, asked brusquely, to a computer prone to make up facts when the documentation is lacking. Never do aksers bother to tell the computer a simple “thanks, it worked”: nobody expects their chats to be read by other meatlings.

In effect, the support community does not build up, over time, a record of easily solved problems. Instead, it piles up residual slop.

And i understand: likely, i should ask the AI myself instead of searching.
But, as much as i’ll readily admit to you on my web blog that i’m out of my element, i do not wish to embarrass myself in a technical chat.

I have learned to only ask questions when i can prove i have done my homework, that i have gathered as much useful information as possible. I ask only for enough information to be unstuck, rather than ask for my helpers to write entire snippets of code for me.

And so, here i am, my question unanswered for now.
I will figure it out, but i will not ask a computer in public.