When I was studying as a CS undergrad in 1998 at the University of Montreal, every student had a Linux account with a special folder in their $HOME where you could put stuff which would be instantly served on the web. It took me a while to see the beauty of it, because at that time, I didn’t know much about the web and Linux. Fast-forward in 2026, and I now work as a professor at an online university, and it’s strangely difficult to host anything online in that environment. Actually, it’s next to impossible. The only thing I have is an extremely slow and clunky web app to edit my home page. And whenever I change something using it, I’m always scared that I’m going to break or delete something. The only mechanism that is available for publishing stuff is based on Moodle, but it’s meant for the courses, and it’s not particularly user-friendly. So me and my (very few) colleagues, who want to publish stuff online in a more flexible way, are forced to use external platforms. But of course the University sees that with suspicious eyes. To someone who doesn’t know about it, Github might appear like a potentially “dangerous” platform. Who knows? Code is scary, because hackers are scary! ...

For some weird reason, I always had some kind of slight “mental hesitation” with the meaning of data encoding versus decoding. Which one goes in what direction? To be honest, I have the same kind of weirdness with other concepts: daylight saving time for instance (are we gaining or losing an hour? I can never tell, sometimes even for many days after a change). So I wanted to create a diagram to illustrate the dichotomy between encoding and decoding, for a course I’m creating on software engineering. So one way to “create with AI” would be to ask one: “Can you please create a diagram to illustrate the difference between data encoding and decoding”. ...

Most often when we think about the difficulties associated with the translation of a language, we are thinking in one direction: you are learning a new language, and it is difficult to translate your own language and thoughts into this new one. This difficulty feels extremely real and visceral, and engages generally with many aspects of your overall cognitive experience: it’s difficult to remember words, grammar rules, it’s embarrassing to make mistakes, frustrating to not feel understood, etc. ...

J’expliquais récemment à un ami que ChatGPT, dans son essence, est « juste » un modèle de prédiction du mot suivant, celui qui vient après une suite d’autres mots. Ainsi, quand on lui demande « Quelle est la capitale de la France ? », il ne répond pas (vraiment) à la question : il complète plutôt une séquence de mots sur laquelle il a été entraîné, en profondeur et avec une très grande efficacité. ...

I was explaining to a friend recently that ChatGPT, to its core, is “just” a model to predict the next word, the one coming after a bunch of other words. So when you ask it “What is the capital of France?”, it does not (really) answer your question, it completes a sequence of words on which it has been trained, deeply and efficiently. So considering that, it might seem that ChatGPT is in a situation that would be akin to you, if someone tells you a bunch of words you don’t understand (in a foreign language say) and then, someone else gives you a card, on which you can find some words to pronounce, as a reply (in a language that you don’t understand but can read, let’s say). ...

With VS Code I am writing a lot of Markdown text these days (for instance this Hugo blog) and I’m using two extensions to help with my word processing needs: Code Spell Checker, which has a bunch of dictionaries for many languages (I use French). Auto Correct, which is much more obscure extension (its code has not been updated for many years), which allows to automatically change a word to another one, as I write, based on a list of entries managed in .vscode/settings.json. ...

Imagine a situation where a majority effectively has a quiet veto over a change. Not because the status quo truly serves them, but because, for each individual person, it feels simpler to go along with it. Speaking up costs energy. It risks conflict. It can feel embarrassing, or socially unsafe. So something persists that does not really benefit the majority, and yet is still easier, moment by moment, to maintain. ...

Or.. the era of turbocharged goalpost moving The ARC benchmark was designed by François Chollet to serve one goal : be sufficiently difficult, demanding so that it cannot be “hacked” by some “cheating” AI techniques, LLM or whatever. But it must do so in a rigorous, systematic and simple to define way, it cannot be vague or ambiguous. It must be (relatively) easy for a human, but hard for a program. And when I first looked at it, I admired its simplicity and purity: the problems are simple but deep, and it’s clear that for many of them, you need to grasp something that goes beyond mere pattern recognition or superficial pattern matching. They seem to require some seriously deeper thinking. And if, like me, you thought for a minute about how you’d try to tackle them, in a programmatic way (ML or otherwise), it was quite easy to become convinced, that this is quite a good benchmark. And at first, what happened, on Kaggle, for instance, was exactly that: nobody could get even remotely decent results, the problem set really felt like a tough nut to crack. From there, the temptation was great, to suggest the idea that whenever ARC would be cracked, AGI would have arrived! ...

Some models of reality are bolder than others Digital physics is the body of mathematical and philosophical work treating the universe and the way it works as a giant digital computer. This is often associated with cellular automata, and names like Konrad Zuse, John Von Neumann, Stephen Wolfram, etc. What I find fascinating about this field is that the models it suggests are making very deep metaphysical claims: if they are true, it means that the underlying structure of the world is much different than we think, and radically simpler in a sense. Take the lattice gas automaton for instance. A version of it is an hexagonal cellular automata with very simple collision rules, not more complicated than the famous Rule 30 or 110, for 1D cellular automata. The impressive thing about it is that a simulation running this rule with many particles can be shown to approximate the Navier-Stokes equations, which are the classical complicated mathematics to describe the dynamics of fluids. Following Wolfram, I find it very appealing to consider the idea that the world is not somehow running “hidden mathematics”, somewhere and somehow, to solve some complicated equations in a seemingly magical way, but rather, that things are radically simpler, in that the world is simply implementing a set of trivially simple rules. The world is not concerned with, or made with mathematics, mathematics just emerges, with inherent and irreducible complexity, from extreme simplicity. ...

Computer Science as a term should simply capitulate and give way to AI In French the term “informatique” feels slightly better, as a label to describe the field, than “Computer Science” feels in English. But this is a rare occurrence for French, because most of the other terms, like “technologie de l’information”, and “science des données”, feel awkward and far from their “real” cultural counterpart, the thing in itself that we do, when we do it. ...