L’avenir c’est construire des outils grâce a l’IA, qui sont imprécis et dont les résultats seront corrigé par l’humain. Jusqu’à ce qu’on progresse et rafine nos méthodes pour ensuite être de plus en plus précis dans les résultats. Nous retournons à notre réel status : techniciens.

Built a small automatic differentiation engine with dynamic computational graph support deep learning framework in Rust from scratch. Gets to 98% accuracy on MNIST dataset in just 1 sec of training! github.com/vaibhawvipul...

👇👇👇

On est lundi un peu d'humour

I had problems with the Scala compiler, I just noticed that I had configured -Xss2m for the IDE... LOL

When working with async, you often need to call a function that would block the runtime. Turns out not all blocking functions are the same Tokio defines two kinds of blocking functions: IO bound and CPU bound. A close read of the tokio docs tells you to avoid CPU bound tasks in spawn_blocking

Am I the only one who has noticed that there are a lot of inactive X accounts here? In a way, it's good because it allows you to meet new people :D

🔥 JVM performance matters. Check out our article recapping this year's session “Keeping Your Java Hot by Solving the JVM Warmup Problem.” with @speakjava.bsky.social! 👉 devnexus.com/posts/keepin... Devnexus '26 will bring even more insights like this. 🎟 Get super early bird tickets until Sept 1!

Concurrency is still a challenge for many, making Approachable Concurrency very welcome. To discover how and why it can improve your way to Swift 6, we’re honored to host one of the true authorities on the topic, @mattie.bsky.social ! 👉 https://swiftconnection.io

Linux developers plan to drop legacy initrd support Almost all systems already rely on initramfs, so the change should simplify the kernel code www.phoronix.com/news/Initrd-...

- Without knowledge of each company’s semantics, AI agents are useless - Successful “Agentic BI” requires data intelligence, when the systems understand the company’s data and its unique business concepts - Agentic AI require: infrastructure, data and semantics www.databricks.com/blog/deliver...

if you’re willing to throw another population under the bus to save yourself, you’re not actually willing to fight fascism. fascism isn’t a bear, it can’t be escaped by simply outrunning someone slower than you. the only way it can be defeated is through solidarity.

One of the keys to productivity is using a toolchain that doesn't slow down compilation too much and generates a feedback loop..

Linux kernel 6.16.3 fixes bugs in ext4. 6.16 adds TCP zero-copy for faster networking, OpenVPN DCO for higher VPN throughput, Nvidia Blackwell GPU support, Intel TDX for secure VMs, USB audio offload, and major ext4/XFS improvements lwn.net/Articles/103...

If there are no major technological changes in the sector tomorrow, how will you develop your career? :)

C'est pas moi qui l'ai dit :D

I'm not a fan of benchmarking your own solution because obviously, you choose the scenario, how you run it, and how you measure it. But a lot of @kestra.io users ask for it, so I did it: kestra.io/docs/perform... 1/5

I'm having troubles using GraalVM from a Nix development environment in VS Code. All other JDK distros work fine. It seems to be just the "graalvm-ce" Nix package to cause issues. I'm not sure if it's a Nix package issue or it's a bug in the VS Code Java extension. Any clue?

Before, we had tech people who influenced sprint planning, defining what needed to change or not in the code base (functional/technical). Now it's an LLM agent. So who do we yell at when it all goes wrong?

Natural language is becoming a new coding language. Before, we had to hire coding experts who “understood” the domain; today, domain experts no longer need us, they are autonomous 🤷‍♂️

Kind of glad that AI is making us give up the idea that code should be self-documenting.

One tangible problem with AI hype is that there is a degree of provocative hypothesis that is necessary in order for people to think a little differently about old problems. But then the hype goes so far that it becomes detached from technical reality, and thus, disqualifies itself.

En quittant Twitter, je viens de remarquer que je n’écoute quasiment plus de podcast tech. Je pense que c’est dû au fait qu’ils n’apparaissent plus dans ma timeline. Je suis le seul dans cette situation ?

Chapter 8: Memory and Disassembly of "Building a Debugger" (in Rust) by @tartanllama.xyz is done! This was a very straightforward chapter, I don't have anything to say about it. As hoped/expected, the iced_x86 Rust lib handles all my disassembly needs with aplomb. github.com/drmorr0/drbug

It's interesting to think that AI makes us more productive because we produce more code and features To be honest, I feel like we've all become juniors again to some degree We've added code to a code base, but is it really useful maintainable.. ? You know those discussions we all had 15 years ago?

We spend hours on complex tech interviews (reverse B-trees, fancy algos) but struggle to compose 3 API calls without breaking. Maybe before playing smart, we should master basic composition and transaction patterns?

I laughed a lot especially during the presentation part www.youtube.com/watch?v=i7aQ...

When we start tracking the execution time of our favorite CLIs, we notice all the time we lose every day/week/month... This should have a significant impact on whether or not we adopt certain practices and tools. As usual, we don't measure enough.

If there were a single lesson I'd drill into new software engineers it'd be this: You can be technically right and politically and systemically dead wrong.

Should I try this in Kotlin/Scala world ???... hum... Breaking FP rules should be funny :D bsky.app/profile/alle...

We did a good job today generating the bytecode for the POJOs of the JFR events. Next steps: sending them to an ORM :D I discovered a lot of strange things in bytecode generation, and I must say that using an LLM to browse through certain parts of the documentation helps speed things up.

Y a peut être un momentum a prendre sur l'inovation est les projets OSS, avec l'AI on peu avancer tres rapidement sur tous les projets qu'on a dans un tiroir et les publier. Bien ou pas, je n'ai pas vraiment d'avis mais si ca peut aider certaines personnes a produire du soft libre : good !

Let's see .. how about a little experiment involving direct support for typescript erasable syntax in workerd. This is not expected to be enabled in production anytime soon (or necessarily at all)... Just experimenting with possibilities. github.com/cloudflare/w...

They should add a feed where I can choose which sources I want (maybe by URL if they don't already have it), organize them by categories of my choice, and show their stories in chronological order. Feel like that would be a hit. Call it Reader or something. I dunno, I'm not great at branding.

The Pursuit of Physical Memory Contiguity in Datacenters dl.acm.org/doi/pdf/10.1...

Expérience, passer de IntelliJ (scala) pour VSCode (+ metals). Je suis fatigué des lags d'indexation, checktypes et etc

we live in a world so obsessed with structure—so much so that we often forget how much of it depends on chance. we crave predictability—systems we can understand and control. but sometimes randomness isn’t a flaw; it’s a feature. somedaystack.com/posts/skipli...

💥Did you know that final is…not so final? Checkout the first video from #JVMLS 2025 and find out what steps are being considered by #Java to address final’s limitations. inside.java/2025/08/12/j...

Ambiance dustri.org/b/my-experie...

L’IA et le vibe coding va solutionner la maintenance de tous nos projets OSS délaissés 🤷‍♂️