Nicolas Rinaudo

@benjamingeer@piaille.fr je pense qu'il y a un juste milieu déplaisant mais acceptable à viser: ne pas rejeter les LLMs en block, au moins pendant l'entretien, mais ne pas non plus promettre des les utiliser pour générer du code. Il est possible, si l'on est forcé d'utiliser des LLMs (et donc de faire fi de ses objections ethiques, ecologiques, morales...), de les utiliser pour. par exemple, se documenter sur des techs qu'on ne souhaite pas apprendre mais devons utiliser. Ou comme moteur de recherche (si tant est qu'on va *systématiquement* valider les résultats à la source). Etc... Après évidemment se pose la question de, serait-on heureux dans une boite qui essaye de forcer les LLMs à tous les niveaux - parce que si c'est pour le code, on peut imaginer aussi que les emails vont etre générés, la documentation, que de nombreuses conversations vont etre "oui mais ChatGPT dit que...", ... Mais c'est parfois préférable à avoir faim...

Oooh lovely: https://github.com/scala/scala3/pull/26246

I'm looking at a little bit of code that, in order to do basic auth in a (very simple) web app, requires understanding Kleisli, IO, and a bunch of other rather arcane concepts.

I love to hate on Spring Boot as much as the next guy, but it's hard to justify favouring the Scala complexity (and the, I'm not exagerating, days of work to get this working) over the somewhat black magic of Spring Boot that gets you something working in an hour and you can move on with your life and never look back.

@NohatCoder@mastodon.gamedev.place I actually think that one is less evil than Google's usual stuff. At least in my case, all services were unlinked by default, which I'm going to assume is a legal requirement because there's no way my opsec is that good, so this email is in fact trying to get you to opt in to the evil stuff. They probably do not want it in the spam folder.

@hipsterelectron@circumstances.run I would have appreciated it if you had, of course, but I can also just wait until the migration to git-page is fully complete :)

@hipsterelectron@circumstances.run I mean I'm not going to complain too hard, it's a massive effort and they're, to the best of my knowledge, pillaging my work to feed their models or ad customers...

@hipsterelectron@circumstances.run ah interesting, that's just what I want to do. What documentation did you follow to achieve this?

Anybody using codeberg pages? I want to move my blog there, but the documentation is confusing - unless I'm mistaken, it documents the git-page method, but tells us to use the legacy v2 method if we want custom domains, and the legacy v2 method isn't documented?

Very glad i started lifting weights again, but i now have muscle pain in some very odd places. Do i really use my armpits? Do i have muscles there!?

@IsomorphicDude@fosstodon.org no, that never happens to me with men. Ahem.

I just naturally, without thinking, called someone I was on the phone for tech support "young man", and I don't think I'll ever forgive myself for that.

Oh this was frustrating. I didn't realise have of the lecture was missing.

Does anyone know where I can find the rest of Bauer's lecture on algebraic effects, the one that's linked on the Eff resource pages (https://github.com/yallop/effects-bibliography?tab=readme-ov-file#2018)?

The mental model that works for me starts from: we’re simply trying to prevent some variables from escaping their intended scope. That’s it. The way this happens is clever and sometimes a little convoluted, but it’s *not* to do with capabilities. It just so happens that we don’t want capabilities to escape their intended scope. But there are other such things.

I’ve been fielding questions on capture checking and have reached the conclusion that the documentation forces you to think about them a certain way, and that’s the wrong mental model.

It’s unfortunate that they’re so tied to capabilities in their vocabulary when they solve a different, orthogonal problem. One that is useful for capabilities, certainly, but distinct from them.

While everybody’s gushing about Project Hail Mary, it might be worth pointing out this tidbit from the book.

@RosaCtrl@social.vivaldi.net you have no control over whether you panic or not, isn't that the entire point? I'm glad you're less anxious now, and that I could at least try to help a little.

@RosaCtrl@social.vivaldi.net no it was a very serious article on hacker news, they were genuinely impressed.

I'm half amused, half appalled by the reactions I'm seeing to Claude's code leak.

Goes from "this is the most impressive piece of code I've ever seen" to "well it *is* april the 1st, surely this must be a joke".

Just published the slides to my "Effects as capabilities" talk from BOBKonf and Scalar: https://nrinaudo.github.io/talks/capabilities.html

@dwardoric@chaos.social i have another one in 2 weeks, and i’m hoping to do better. This one was on trails in the forest with lots of ups and downs, the next one is a lot more flat around the Fontainebleau castle

49 minutes 22. Not too bad!

Showing up for a 10k race quite casually, and all the athleticism clubs in the area are here taking this sooo seriously and warning up like crazy. Starting to feel a little intimidated

@josch@floss.social @mhoye@cosocial.ca was that ever in doubt?

@mhoye@cosocial.ca so, don't get me wrong, I agree with you wholeheartedly and am making this argument regularly at work. But it's important to view the other perspective: an LLM will follow your instructions far, far faster than a junior developer. Hell, it will follow my own instructions far faster than I would.

It's hard to make the argument that speed is not the important metric here - to show that LLM-generated code is at best average, by definition. That a human gets better every single time they perform a task, in a way LLMs don't. That a trained junior can train other juniors, and if you do it right, you get exponentially more developers, where an LLM stays the same.

The problem is that:

• managers in our industry often care less about quality than about speed of delivery, because you can always fix later (which means, really, they'll have made a bundle and left before consequences must be dealt with).
• the asshole developer, which is unfortunately (in my experience) the majority of us, cares about feeling better than the rest, not about making everybody better.

@cellio@indieweb.social oh right i thought it was your question, sorry! I’m happy to reply with what i just said here, but i don’t have much more to say than that, not sure it’s worth it

@cellio@indieweb.social i don’t necessarily have the theoretical background for this, but aren’t dependent types types indexed by values? Scala match types aren’t that, they’re a lot closer to haskell type families