A few weeks ago I had the pleasure of participating in the Argentinian radio program “Piloto de Prueba”, translated as “Test Pilot”, in the radio station of Universidad Nacional de La Plata—my alma mater. I don’t make a habit of being in public media, so this was an enriching experience. The interview focused on analyzing recent news on AI, particularly about the tempatively imminent dangers of this technology.

My opinion, in general, is less apocalyptic than what’s been published on mainstream media. Some comments projecting numbers, such as ten percent of the global population getting killed due to the misuse of AI, are irresponsible. They do not have a scientific base. This sensationalism causes the average citizen, the one for which computers internals are a mystery, to get unnecessarily alarmed. This doesn’t mean that we need to take a passive attitude towards this technology; equally to every disruptive innovation, we need to learn how to channel it for the betterment of society while taking sufficient precautions against its dangers.

take the long way to the summit
avoid shortcuts to the top
slow down and observe
how the falcon hovers and dives

when necessary
go back on your steps
and choose a different path
but don’t forget to keep walking

take the long way to the summit
the sun and the moon
will be there
for you

I’ve been self-hosting part of my Internet footprint for over a decade. This website, my mail system, a personal wiki, my code repositories, you name it. I do this for two reasons: to be in control of my own data, and to learn by doing. After all:

“For the things we have to learn before we can do them, we learn by doing them.”

– Aristotle, The Nicomachean Ethics

My services are generally steady, as I’m not doing anything overly complicated. I run the stable release of Debian in a tiny virtual machine, and use packages from the official archives with unattended security upgrades. However, every now and then I have to deal with some instability. This is fine; it’s part of the trade-off of self-hosting.

During the last (southern hemisphere) summer I wrote the Emacs package Org Tempus to enhance the editor’s time tracking capabilities. I’ve been using it constantly since then, fixing obscure race conditions and timer bugs. Even with the aid of LLMs, it took several iterations to get things right.

The usage of LLMs still produces bloated code, also known as AI slop. However, there are techniques that can be used to prevent it. I’ve experimented with a few LLM “skills” to simplify codebases, and I was pleased with the result. Code is a liability, so cutting down on lines is desirable.

It’s been a long while since my last Debian contributions log. Life gets busy, interests change, and sometimes it’s necessary to take a break to come back more energized and with new ideas. A few months ago I started giving some love to all the packages I maintain. I was curious to see how much the workflows have changed since the advent of LLMs. Packaging work is mostly mechanical and well defined, making it quite fitting for these tools to process.