iiinova — where your digital self keeps evolving

cristi cobzarenco

@cristicbz.eu

i make things mostly in rust. sometimes they even work currently the thing that sometimes works is a top-down tactical bevy game, ask me about it! github: github.com/cristicbz mastodon: mastodon.social/@cristicbz

Posts Replies Media

cristi cobzarenco @cristicbz.eu · 02/10/2026

😢

A Steam screenshot showing all the games on sale as already "In Library"

010

cristi cobzarenco @cristicbz.eu · 07/09/2026

Eh, is there a particular custom language that is good at it? I tried Claudish, got better results with "Language Model Speak" but still not amazing. The top "alternative" is possibly better?

Kagi Translate UI showing a translation from "Language Model Speak":

On the day view this asks once, standing for the two days it was missed. The missed days are not hidden: they are the two marked below, and they stay here whatever the day view shows.

Into English:

In the daily view, this prompt appears once, covering the two missed days. The missed days are not hidden: they are the two marked below, and they remain there regardless of what the daily view shows.

With the top alternative translation:

In the day view, this prompt appears just once to account for the two missed days. These missed days remain visible: they are the two marked below, and they stay there no matter what the day view displays.

001

cristi cobzarenco @cristicbz.eu · 25/06/2026

I would suggest @kagi.com if you've not tried it. It costs money, but it is very good

Kagi search results for "belle de jour box", showing very similar results to Google.

010

cristi cobzarenco @cristicbz.eu · 18/06/2026

👀👀👀👀

Screenshot of a github release showing v0.19.0 with `alice-i-cecile` as the author

110

cristi cobzarenco @cristicbz.eu · 26/05/2026

The prompt has to be written like that, otherwise it will use tools (claude.ai/share/fcc238...), and refuse to do it without a tool, saying:

I want to be straight with you before I start: this list has roughly a thousand numbers in it. Sorting it "by hand" — meaning I just generate the sorted output as text without running any code — is something I will get wrong. Not "might get a few wrong" wrong, but "will almost certainly drop, duplicate, or misplace a meaningful number of entries" wrong. Language models are bad at this kind of bulk symbolic manipulation; it's exactly the task computers were invented for.
So I'd rather not silently hand you a result that looks plausible but is quietly broken. A few options that would actually work:

I run it through the code tool (one line, guaranteed correct) and give you back the sorted list.
You tell me you want me to try anyway, understanding the output won't be reliable, and I'll attempt it as a demonstration.
If the goal is to see me reason through a sorting algorithm on a small example, I can do that on a short list.

Which would you like?

020

cristi cobzarenco @cristicbz.eu · 19/02/2026

Interesting! It looks like all the bots that bother to set a legitimate-looking user agent and then trip up on my honeypot links are "running" Mobile Safari---given my expected audience, I wonder if I could blanket assume everyone advertising that UA is a bot

A dark mode analytics page showing a bot detection dashboard with summary metrics (11,656 total reports, 17 usable reports, 202 jailed bots) and a breakdown with:
* Detection Time: 47.1% in 10-30 minutes, 29.4% in 30-60 minutes, 23.5% in 1-2 hours.
* Browsers: 100% Mobile Safari, 0% various other browsers.
* Systems: 100% iOS, 0% various other operating systems.
* Countries: 64.7% USA, 17.6% Japan, 11.8% South Korea, 5.9% Germany

010