Reposted by Florentin RochetDr Pyviv @pyviv.bsky.social · 07/10/2026Je suis un mélange de déni, colère et dépression J'essaie de trouver la 4ème voie mais si j'arrive à la distinguer à peu près au niveau individuel (et puis j'ai un poste, je fais ce que je veux au final, c'est beaucoup plus facile), au niveau collectif j'ai du mal à voir 131
Reposted by Florentin RochetDarren Dahly @statsepi.bsky.social · 06/10/2026They hounded Aaron Swartz to death for downloading papers from jstor but society is now supposed to just accept mass theft because this thing can write that TPS report *for* you. en.wikipedia.org/wiki/Aaron_S...en.wikipedia.orgAaron Swartz - Wikipedia 111619547
Florentin Rochet @frochet.bsky.social · 05/10/2026I think I understand better a LLM generated paper than their authors (it is right into my expertise). I read parts of the text where the authors likely misunderstood their LLM and are writing incorrect (but plausible) statements. It is overall much better than last year's LLM papers. Still sad tho. 000
Florentin Rochet @frochet.bsky.social · 03/10/2026However valgrind and gdb are still tools I expect student to play with in my classes: they're useful to help understand computers and programs, even if we may not need them eventually. 100
Florentin Rochet @frochet.bsky.social · 03/10/2026I am writing unsafe code in Rust. it is tedious and I usually shoot myself a few times in the foot with undefined behavior. I don't use valgrind or gdb to debug these cases anymore. I am just telling my local Qwen to look at my mess, and it fixes everything in 5 minutes. 110
Florentin Rochet @frochet.bsky.social · 30/09/2026J'ai l'impression qu'il y a une zone dans laquelle ces outils sont réellement des outils et ne remplace pas l'humain. Il faut être capable de comprendre où cette zone se trouve, et avoir l'autodiscipline de ne pas en sortir. 011
Florentin Rochet @frochet.bsky.social · 30/09/2026Je ressens une certaine satisfaction à améliorer la robustesse de mon code en utilisant de l'IA locale. Ce genre d'utilisation me semble être un net positif: aide à l'écriture de test, de scripts pour des tests d'intégration, à de la documentation. Je ne perd pas la maitrise de mon travail. 100
Florentin Rochet @frochet.bsky.social · 25/09/2026This creep me out, holy shit research.meta.ai/blog/bringing-your-muse-to-liferesearch.meta.aiBringing Your Muse to LifeMuse Realtime Avatar turns Muse Realtime Voice into expressive, interactive avatars. 000
Florentin Rochet @frochet.bsky.social · 22/09/2026Ça me rappelle les 3 mois qu'il m'a fallus pour obtenir le 443 en entrée sur UDP. Il y a certains combats pour lesquels on est pas préparé :') 000
Florentin Rochet @frochet.bsky.social · 22/09/2026I am positively surprised by the number of students this year pushing back on AI. 110
Reposted by Florentin RochetDavid Monniaux @monniauxd.bsky.social · 21/09/2026blogs.mediapart.fr/david-monnia...blogs.mediapart.frL'insécurité informatique, un problème de volonté et de priorités ?Et si la sécurité informatique n'était pas un problème technique, et plus un problème organisationnel et politique ? 384
Florentin Rochet @frochet.bsky.social · 14/09/2026Cette année je passe mes beamers en HTML5/CSS avec des animations créées par la machine à mots pour m'aider à expliquer les machins compliqués. Réelle plus value ou distraction inutile? J'aurai un avis dans quelques semaines 🙃 000
Florentin Rochet @frochet.bsky.social · 14/09/2026The point raised within this letter mathandai.org : there is a misalignment focusing on the result and not the journey to gain expertise, to understand deeply and to innovate. This seems to be valid for any science field.mathandai.orgDeclaration — Math and AIRead the declaration and add your name. 021
Reposted by Florentin RochetVincent ENGLEBERT @vincent-englebert.bsky.social · 11/09/2026La Faculté d'informatique de l'Université de Namur recrute : Un•e chargé•e de cours à temps plein en Sécurité informatique 🔗 jobs.unamur.be/emploi.2026-... #job #softwareengineering #position #university #emploi 001
Florentin Rochet @frochet.bsky.social · 10/09/2026blog.apnic.net/2022/05/24/t...blog.apnic.netTCPLS: Modern Transport Services with TCP and TLS | APNIC BlogGuest Post: TCPLS takes advantage of decades of performance optimizations made to TCP and has also a few other tricks up its sleeves. 000
Florentin Rochet @frochet.bsky.social · 10/09/2026I am re-reading an old blogpost of mine from 2021. It contains em dashes and its text construction feels a bit AI. I think I contributed to this monster 💀 100
Reposted by Florentin Rochet🏴☠️ MC Pablo ∴ @p4bl0.net · 10/09/2026La licence informatique de l'Université Paris 8 a pris des mesures officielles, entérinées dans ses modalités de contrôle des connaissances et des compétences et validées en CFVU, contre le recours aux IA génératives, pour le bien de nos étudiant·es : informatique.up8.edu/licence-iv/o... 3378
Reposted by Florentin RochetGro-Tsen @gro-tsen.bsky.social · 08/09/2026Basically, math papers should now no longer be about ‣ I've solved this problem / proved this conjecture but rather about ‣ I found and understood something instructive / enlightening / surprising / beautiful / useful, and I want to share it with you and explain it. 513323
Reposted by Florentin RochetVincent ENGLEBERT @vincent-englebert.bsky.social · 08/09/2026🫣 Je rappelle : signal.org pour une alternative plus saine.signal.orgSignal Messenger: Speak FreelySay "hello" to a different messaging experience. An unexpected focus on privacy, combined with all of the features you expect. 011
Florentin Rochet @frochet.bsky.social · 04/09/2026Cette semaine infernale est prisque finie ... Après les corrections des copies ce weekend 100
Florentin Rochet @frochet.bsky.social · 28/08/2026Je pense que malloc() navigue en utilisant des meta-donnés autour d'allocations précédemment free() pour chercher à recycler. Des métadonnées d'une précédente allocation seraient corrompues? Unsorted est le nom d'une des bins dans laquelle free() place de l'info sur l'allocation liberée. 000
Florentin Rochet @frochet.bsky.social · 28/08/2026Times to teach low-level code optimization given how much RAM is projected to cost? (yeah, any reason to ditch python) 000
Reposted by Florentin Rochet🏴☠️ MC Pablo ∴ @p4bl0.net · 25/08/2026Your executable is a SQLite database Complètement fou, mais très amusant et intéressant. fzakaria.com/2026/08/23/y...fzakaria.comYour executable is a SQLite databaseI have been probably obsessed with two things in the last few years: Nix as a tool to explore innovative ideas that require the capability to rebuild the world and replacing ELF with SQLite as an exec... 084
Florentin Rochet @frochet.bsky.social · 25/08/2026People wonder whether ox-alpha that showed up on OpenCode as a stealth model and free of charge is either a new version of Gemini or some chinese model. Questioning it about what happened the 4th of June 1989 should do the trick? 000
Reposted by Florentin RochetXavier Devroey @xdevroey.wsocial.eu · 21/08/2026These weeks are the prep courses in informatics. 4 sessions of 2h each for 2 groups with @bvdose.bsky.social. Too much work on top of resit exams + MSc thesis defenses (guess who is presiding over almost all of the juries this year? ... yep, @bvdose.bsky.social and yt). #aintnorestforthewicked 111
Florentin Rochet @frochet.bsky.social · 20/08/2026The SIGCOMM chairs were very kind to let me present remotely as I could not go in the US. More cool stuff related to secure communication will be coming in the future :-) 020
Florentin Rochet @frochet.bsky.social · 18/08/2026At reasoning output of ~50 tok/s, I can follow the reasoning and even found myself gamifying its chain-of-thoughts, trying to understand the problem behind an obscure bug faster than it does. Frontier models don't show chain of thoughts to avoid distillation, and that makes them boring, depressing. 000
Reposted by Florentin RochetSteven Murdoch @steven.murdoch.is · 18/08/2026I spoke to The Guardian about Claude’s content watermarking. As long as it only modifies how the existing random number generator works, it should have no measurable effect on quality. www.theguardian.com/technology/2... 141
Florentin Rochet @frochet.bsky.social · 18/08/2026Qwen 3.8-27B is stronger than frontier models that were state of the art a few months ago. I played with it last weekend, and it can do miracles while running fully in my 2,5k€ computer, cut off from the Internet. I still don't know exactly how to feel about it. 110
Reposted by Florentin RochetOlivia Guest · Ολίβια Γκεστ @olivia.science · 24/10/2025Katrien Beuls: "When ChatGPT launched at the end of 2022, the then rector, Rik Van de Walle, swiftly announced that students could use the chatbot for their master's thesis. A bad move. Or at least the beginning of a lot of trouble.” apache.be/2025/10/24/b...apache.beBelgian AI scientists resist the use of AI in academiaSeveral AI scientists have published an open letter calling for a ban on AI use by students. 14511
Florentin Rochet @frochet.bsky.social · 17/08/2026Now I am going to put this one on my office door: 000
Florentin Rochet @frochet.bsky.social · 17/08/2026Microsoft is making this logo one of the most well known. 110
Florentin Rochet @frochet.bsky.social · 17/08/2026I get why people are writing their paper with AI. But paper reviews by AI? I don't get it, why being salty-by-proxy when we can use our own fingers to tipitap salty remarks. 000
Reposted by Florentin RochetLukasz Olejnik @lukaszolejnik.bsky.social · 17/08/2026How about disagreeing with this @financialtimes.com take? Adopting open-weight models does not equate adopting standards and governance. The models can be modified anyway so whatever “standards” / “governance” are baked in may be bent, cut out, replaced. That’s not how technological influence works 2102
Florentin Rochet @frochet.bsky.social · 12/08/2026Il y a un mémoire dans ma liste qui consiste à exploiter la robustness pour exfiltrer de l'information. Il y a pas mal d'attaques qui sont possibles sur base de cette idée. La dernière en date que j'ai pu voir est celle sur l'exfiltration de chain of thoughts en.wikipedia.org/wiki/Robustn...en.wikipedia.orgRobustness principle - Wikipedia 100
Florentin Rochet @frochet.bsky.social · 12/08/2026S'il a une petite LED, il doit être possible de la faire clignoter (très vite) pour émettre de l'information sans que l'agent secret paranoïaque s'en aperçoive. Mais si ce monsieur est parano, il aurait cassé la LED aussi?🤔 110
Florentin Rochet @frochet.bsky.social · 11/08/2026I wonder whether this would impact VPN usage. These watermarks are the result of the "EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content", so it may not apply if your IP address is outside of Europe? 000
Florentin Rochet @frochet.bsky.social · 11/08/2026I suppose many watermarks in published papers will soon appear in plain sight. Many commits in open source projects too. 200
Florentin Rochet @frochet.bsky.social · 11/08/2026I feel like they were experimenting on text watermarks for some times 😅 100
Florentin Rochet @frochet.bsky.social · 11/08/2026So apparently Claude now watermarks generated content (From August 2, 2026 onward) TL;DR "imperceptible watermark directly into the text itself", and "signed metadata" attached to file modifications/creations support.claude.com/en/articles/...support.claude.comHow Claude marks AI-generated content | Claude Help Center 100
Florentin Rochet @frochet.bsky.social · 11/08/2026The feedback sentiment for the idea was positive. Avoiding a memcpy() over the whole stream by-design is "interesting", and a "clever hack" (quoting participants), but the working groups also wonders whether the complexity is worth it, and what current usage of QUIC v1 suffers from memcpy()ies. 000
Florentin Rochet @frochet.bsky.social · 11/08/2026The overall idea was presented in Vienna at the QUIC working group meeting. Video is here: youtu.be/Pzy_4svJ4vY?...youtu.beIETF 126: QUIC (QUIC) 2026-07-22 14:00YouTube video by IETF - Internet Engineering Task Force 110