Andrew Yourtchenko
banner
ayourtch.bsky.social
Andrew Yourtchenko
@ayourtch.bsky.social
380 followers 760 following 690 posts
Embedded programming, some Rust, 3D-printing and active mobility. Hacking on fd.io by day at cisco. Release manager for VPP. CiscoLive Europe NOC - automating stuff. Bits of code: GitHub.com/ayourtch ; all posts are entirely only mine.
Posts Media Videos Starter Packs
Really cool! A little UX feedback: iPhone 13 Pro with Safari, on the first scroll the hyperlinked text in the blog seems to expand and then get back to its width. Not sure whether it’s an intentional effect, so thought to let you know !
Reposted by Andrew Yourtchenko
The most successful cities will be the ones that give people safe, affordable, efficient, enjoyable freedom of choice in transportation.
They didn’t specify which human…
agreed, also more well rounded *should* mean better real world coding experience as well. In any case - this is all great news
Stick the chicken into the pressure cooker and boil for a 3 hours. The result is edible in its entirety. (Yes, I am lazy ;-)
The coding benchmark results seemed a bit weaker though ?
Reposted by Andrew Yourtchenko
Evolution du taux de motorisation dans les 20 plus grandes villes de France Métropolitaine entre 2010 et 2021

➡️Les ménages non-motorisés augmentent presque partout grâce aux politiques de mobilité alternatives et de réduction de la place de la voiture

👉Un enjeu pour les prochaines municipales
Reposted by Andrew Yourtchenko
Chinese astronauts become the first to grill wings in space, paving the way for future uncstronauts
and it’s probably very obsolete by now! 😂 The potentially interesting thing is that it’s supposed to be included into the hash calculation forwarding devices load balancing, so you could in principle affect the network path your flow takes ! but then labs.ripe.net/author/joel_... confuses …
IPv6 Flow Label: Misuse in Hashing
When RFC1883 was published in 1995 it marked not the end of the process that produced the IPv6 protocol architecture, but rather was a milestone in the evolution of the IPv6 protocol.
labs.ripe.net
A rare IETF meeting would go without someone inventing yet another use for the flow label! 😂
Agreed. Understanding the max amount of risk for each case, putting a cap on it and giving the ways to undo the wrong where possible. And being transparent to interfacing entities about all of this.
i like the way you describe it; though to me it feels a good bit like judging whether a given email message is spam, purely by its contents. The first 80% are trivially easy, the last 0.1% are extremely hard.
Reposted by Andrew Yourtchenko
asterinas: Asterinas is a secure, fast, and general-purpose OS kernel, written in Rust and providing Linux-compatible ABI. ★3745 https://github.com/asterinas/asterinas
asterinas / asterinas
Asterinas is a secure, fast, and general-purpose OS kernel, written in Rust and providing Linux-compatible ABI.
github.com
I have been using Claude on pyo3 bindings to a rust crate and boy did I remember your post… except in that case there’s no easy answer as I am already Rust :-).. the closest I could find is to start claude inside the activated venv, it seemed to make things a little better.
If people are subscribing to a block list “ai bros”, maybe they just don’t want to be “bridged” ? I know the list, tried it and removed it from the block lists I use, because *i* found it too restrictive for *my* tastes. But someone who wants to eliminate AI topics entirely might find it just right.
Some “design choices” (as per my comment here: github.com/ayourtch/nat...) aren’t pretty - but as that uses my own library, I treat it as an improvement being needed for “ease of use” of it.
Another example: github.com/ayourtch/osi..., which it then folded into the library itself; about three days from start to finish with fun debugging inbetween: stdio.be/blog/2025-05...
A bit hard to capture it well… less prone to LLM getting lost, perhaps. With a couple of JS projects, I ended up with a thoroughly tangled mess with multiple functions repeating, that had to eventually be shredded. Rust is much stricter with that - and great compiler error messages help LLM a lot.
I did quite a few side projects in the past year in Claude+Rust, in summary: happy. Eg: this game took 1.5 days with 0.5 days LLM-coding a buggy JS prototype and then a day of LLM-converting it into somewhat less buggy Rust: stdio.be/circuits/?ur... - Rust edit interactions felt much more “robust”.
I only had one thought - “why?”
With the URLs it is often iffy - often the origins do ban the agent retrieval, not sure if that were the case - it then attempts to retrieve the data in roundabout ways, but often gets distracted with the task itself :-)
I made a thing today… one might even say it’s a game. 🤪 With Claude and GLM-4.6 to speed up the process, and some manual touch-ups when they got stuck. First in Javascript as a quick PoC, then had it translated in Rust / webassembly. stdio.be/circuits/?ur...
Logic Circuit Simulator - A Circuit Simulation Game
stdio.be
If you are using agentic coding tools, did you try having a checkout of the up to date library/docs in a temporary directory within the project and instructing the agent to consult there before writing the code ? Don’t know about google ai, but it worked very well for me with Claude Code.
A re-run next week, maybe ? :-)
This intense cheering will one day look short-sighted.