←
Cedric Lejeune2,394 posts
Cedric Lejeune avatar

Cedric Lejeune

@CLVPTY

Agentic transformation at HOLLI/TREKLAIR "Every data tells a story". holli.st

◷ Joined August 2015▧ 95 Photos and videos
Mickaël Ahouansou avatar
⟳ Cedric Lejeune retweeted
Mickaël Ahouansou@optimikee·Nov 17

BFM : « Je voudrais être rassurante. Seuls 6 % des employeurs pensent que l'IA pourra concurrencer directement l'humain dans leur métier principal d'ici cinq ans. » Pardon, mais cette rassurite, c'est des conneries. En mars 2026, les 21-29 ans signaient déjà environ 6 200 CDI de moins par mois qu'en novembre 2022 dans les métiers les plus exposés à l'IA (HEC et Banque de France). Ces 6 %, c'est ce que les employeurs anticipent. Les embauches, c'est ce qui se passe déjà. Les métiers exposés, ce sont ceux dont l'IA peut le plus facilement faire les tâches. L'étude cite par exemple le traitement de texte, la traduction, le journalisme, les études économiques et commerciales, les opérations bancaires. Dans ces métiers, les embauches en CDI des 21-29 ans sont environ 28 % sous leur niveau de novembre 2022, le mois de l'arrivée de ChatGPT. Contre 8 % dans les métiers les moins exposés. Aux États-Unis, Stanford trouve le même décrochage chez les 22-25 ans : un écart d'environ 19 % avec les métiers moins exposés, surtout par moins d'embauches. Et l'écart va dans le même sens à l'intérieur d'une même entreprise, ou en écartant l'effet de la hausse des taux d'intérêt. On peut encore discuter la cause. Le changement, lui, est mesuré. Et comme il passe par les embauches, personne ne le voit venir. Pas de licenciement retentissant. Juste des offres qui ne sont pas publiées, et des jeunes qui attendent leur premier CDI. Si vous cherchez votre premier poste, si votre enfant finit ses études, ou si vous recrutez des débutants, c'est déjà votre sujet. Je vois le mécanisme dans mon propre travail. Une stratégie de référencement (SEO) ambitieuse qui me demandait deux semaines, je la livre en trois jours depuis deux mois, avec la dernière génération de modèles. Et j'ai beaucoup moins besoin de la retravailler. Une équipe qui livre trois fois plus vite n'a pas besoin d'embaucher pour faire trois fois plus. Pourquoi les débutants d'abord ? Les chercheurs avancent une piste : ils font davantage de tâches que l'IA accomplit bien, sans l'expérience qui les rend difficiles à remplacer. Mon explication va dans le même sens : l'IA progresse d'abord là où on peut vérifier vite qu'elle a raison. Un compte rendu de réunion fait par un logiciel comme Granola, un email, un rapport, un rapprochement de documents : ça se vérifie vite. C'est la couche simple du travail, celle par laquelle commencent les débutants. Regardez votre semaine : combien d'heures y passent ? Si vous avez de l'expérience, vous vous dites que ça ne vous concerne pas. Les chiffres vous donnent raison : en France, l'écart s'atténue après 30 ans et disparaît presque après 50. Pour l'instant. Les chercheurs le rappellent eux-mêmes : c'est dans les postes de débutant que s'acquièrent les compétences des cadres et des dirigeants de demain. Si vous recrutez, c'est votre vivier de futurs cadres qui se joue. Et la couche simple de votre semaine, c'est celle que l'IA apprend en premier. La couche difficile, comme confier à l'IA un dossier entier et vérifier son travail, demande encore beaucoup de pratique. Mais chaque nouvelle génération de modèles en rend une partie accessible. Mon pari : la zone touchée va monter avec les modèles. Ceux qui s'y mettent maintenant auront de l'avance quand elle sera à la portée de tous. L'édito conclut lui-même qu'il va falloir s'y mettre, et vite. Sur ce point, il a raison. Mais on ne prépare pas les gens en les rassurant avec un sondage. Les gens cherchent les preuves du changement alors qu'elles sont déjà là. Ils regardent juste aux mauvais endroits.

Post attachment
BFM@BFMTVNov 17

L'ÉDITO ÉCO - IA - Les premiers emplois menacés

1:49
Daniel Schneidermann avatar
⟳ Cedric Lejeune retweeted
Daniel Schneidermann@d_schneidermann·Nov 16

Extraordinaire photo.

Photo prise par la FCPE pendant la visite du ministre Geffray au lycée Mandela à Nantes. Costa-Gavras aurait pu faire un film de SF dystopique avec de telles images

Post attachment
Arnaud Bertrand avatar
⟳ Cedric Lejeune retweeted
Arnaud Bertrand@RnaudBertrand·Nov 17

By my own count, if one includes these latest sanctions, the EU has now sanctioned 8 European citizens for speech: - 3 German citizens: Alina Lipp, Thomas Röper and Hüseyin Doğru - 3 French citizens: Xavier Moreau, Adrien Bocquet and Sylvain Afoua (aka Egountchi Behanzin) - 1 Lithuanian citizen: Ernest Mackevičius - 1 Irish citizen: Chay Bowes The sanctions include freezing these people's assets and making it a crime for anyone in the EU to pay them, hire them or even lend them a single euro. They also include a travel ban, which leads to some absurd situations: Doğru - based in Germany - was invited by some members of the EU parliament to Brussels to plead his case but he couldn't go due to the travel ban. As one of the MEPs who invited him - and made 40 phone calls to try, unsuccessfully, to get him cleared to come - put it: "It really was like something out of a Kafka novel" (berliner-zeitung.de/article/…). And, probably worst of all, all of this was decided without any charge, trial or court ruling - simply as a unilateral decision of government ministers. Yesterday's sanctions, for instance, were decided at the Foreign Affairs Council meeting of EU Foreign Ministers and were simply signed by Kaja Kallas alone on behalf of the Council (eur-lex.europa.eu/legal-cont…). Put it all together, and this violates at least half a dozen of what the EU keeps saying are the fundamental values and principles it supposedly fights for: 1) Separation of powers: here we have government ministers - members of the executive branch of their countries - acting as prosecutor, judge and jury 2) Presumption of innocence (Article 48 of the EU’s own Charter of Fundamental Rights): they’re punished without guilt ever being established, and without being given a chance to plead their case before the decision 3) Right to a fair trial (Article 47): there was no trial at all, let alone a fair one 4) Freedom of expression and of the media (Article 11): they were sanctioned - explicitly so - for what they said and published. For instance, in these latest sanctions, Chay Bowes was sanctioned for, among other things, the speech crime of "participat[ing] in the spreading of anti-Western and anti-Ukraine disinformation at the UN Security Council" (eur-lex.europa.eu/legal-cont…). Sanctioned in part for a speech at the UN, a forum that exists precisely so that clashing views on international matters can be heard, quite the message to send from the EU... 5) Freedom of movement (Article 45): they're under a travel ban 6) Right to work and to property (Articles 15 and 17): their assets are frozen and nobody may employ or pay them You can think whatever you want about these 8 people. But ask yourself this: even if you are pro-Ukraine, pro-EU, anti-Russia and you view these people as malignant Russian operatives... what is the EU fighting for here, exactly? This is more profound than you may think: we're speaking about the very foundation of modern European civilization, built on the very idea that political power alone cannot decide what is true. I'm not exaggerating. What we’re witnessing is a return to the pre-Enlightenment playbook, when monarchic or religious authorities defined the truth, deviations from it were called heresy, and heretics were excommunicated - cut off from the community, so that no one could trade with them, employ them or help them. In Europe's darkest times, heretics were even famously burned at the stake - like Joan of Arc or Giordano Bruno. It took literal centuries of religious wars and debates, and the likes of Voltaire, Kant and Mill, to replace it with a basic idea: that no single power, however legitimate, gets to be the final judge of what is true, and that you answer for your words only before a court, under laws written via a process separate from the power that does the accusation. You can find countless works of European literature on this very matter. Take Spinoza, excommunicated in 1656 by the Portuguese-Jewish community of Amsterdam, with a decree ordering that "no one should communicate with him" nor "accord him any favor". He went on to write the Theological-Political Treatise, one of the founding texts of the Enlightenment, with an entire chapter arguing that "in a free state every man may think what he likes, and say what he thinks." Or Mill, two centuries later in On Liberty, Chapter 2: "All silencing of discussion is an assumption of infallibility." Or Kafka’s The Trial, the story of a man accused by an authority nobody can quite locate, of something nobody will explain. How is what Europe is presently doing anything other than going back to that very system, under new appellations? The authority: government ministers. The heresy: "disinformation". The excommunication: assets frozen, banned from travelling and nobody may pay them, hire them or lend them a single euro. It’s basically Spinoza’s excommunication decree on EU letterhead, with a bit of Kafka sprinkled in given that the man invited to the European Parliament to contest his punishment couldn’t attend because of the very punishment he wanted to contest. You don’t have to like these 8 people, or even care about them, to see the problem. Heck, this system was developed precisely for the people nobody likes: consensual ideas never needed protecting. No, the real question here is this: do you want Europe to be the heir of Spinoza - or of Voltaire, or of Galileo, or of any of the countless Europeans who suffered or even died for thinking differently - or of their persecutors?

Today the EU sanctioned 30 individuals, including journalists & EU citizens, without trial, for “misinformation” & other speech crimes. Their bank accounts have been frozen, travel restricted & more. The EU has yet to sanction Benjamin Netanyahu - for anything.

Li Zexin 李泽欣 avatar
⟳ Cedric Lejeune retweeted
Li Zexin 李泽欣@XH_Lee23·Nov 14

This is blowing my mind 🤯 China has developed a 10-ton-class electric passenger jet that can fly either with a pilot or fully autonomously. It can carry 19 passengers, cruise at 300 km/h, and fly up to 400 km. The plane completed its first piloted flight today in China. Its first autonomous flight was back in February, and it has completed 19 unmanned verification flights.

0:17
Sam Wasserman avatar
⟳ Cedric Lejeune retweeted
Sam Wasserman@SamJWasserman·Nov 11

Imagination Studio... The Hermes Plugin: Our all in one local open source creative studio for images,movies,music,voice,motion graphics, sfx + more is Now Available with direct Hermes integration via Hermes Plugin. Our official submission to the Hermes Agent Plugin Hackathon. @u1tra_instinct and me!

1:27
Aurélie Trouvé avatar
⟳ Cedric Lejeune retweeted
Aurélie Trouvé@TrouveAurelie·Nov 4

Monsieur le Directeur de la BPI, Votre action est néfaste. Vous aidez certains fonds prédateurs à piller des entreprises françaises, les recettes de l'Etat et la sécurité sociale. Vous abandonnez des entreprises stratégiques pour la France. Vous avez fait de la BPI une banque comme une autre dont le directeur joue les influenceurs du débat public. Il est temps que vous partiez.

5:49
Le PLOUTOS avatar
⟳ Cedric Lejeune retweeted
Le PLOUTOS@leploutos·Nov 7

Hier soir, j'utilisais DeepSeek via OMP pour ranger ma Seedbox donc. Modèle choisi un peu vite, sur OpenRouter comme d'hab. Je pensais cramer des crédits. Au bout d'un moment je vais voir dans OpenRouter : les requêtes passent par @MistralAI, en BYOK. En fait je paye rien. Et je le savais même pas. J'avais branché ma clé Mistral en BYOK pour tester un modèle avec mon abonnement Vibe. Entre-temps, Mistral s'est mis à distribuer DeepSeek. Résultat : sans rien toucher, DeepSeek tourne sur mon abonnement Mistral à 17€/mois. Ajoute GLM, que Mistral sert aussi, et l'abonnement devient ultra rentable. Le problème de Mistral, c'est qu'il est pas intégré partout. OpenRouter, lui, l'est. Tu branches ta clé une fois, et tous les outils branchés sur OpenRouter en profitent.

Post attachment

Pendant que Claude refusait d'aller lire une doc API publique sans autorisations, dans une autre fenêtre, j'ai demandé à DeepSeek d'aller se connecter à ma seedbox en SSH. Il a commencé par faire l'inventaire de mes films, des doublons, des fichiers qui ne seedaient jamais. Ensuite, je l'ai connecté à l'API de recherche de mon tracker Torrent. Sur chacun des 300 éléments de mon serveur, il est allé chercher si une meilleure qualité était sortie, ou une version offrant un meilleur ratio. Il n'a pas bronché une seule fois sur la moralité de ce qu'il faisait. Vraiment, entre les modèles américains et chinois, il y a deux mondes.

Post attachment
CyberRobo avatar
⟳ Cedric Lejeune retweeted
CyberRobo@CyberRobooo·Nov 3

A humanoid worker putting in an 8-hour shift. Xiaomi announced that its CyberOne 02 humanoid robots completed 8 consecutive hours of work at a Xiaomi EV factory, sorting and placing parts while collaborating with AMRs, marking its first achievement of production-line-level operational capability. Powered by the Xiaomi-Robotics-1 foundation model(100,000 hours of UMI pre-training.)

1:21

Xiaomi Humanoid Robot New Factory Training Since Q1, Xiaomi’s humanoid robots have been training on dual-side operations at the self-tapping screw workstation in its auto factory, achieving a 98% success rate. Today, Xiaomi shared that the CyberOne Gen2 autonomously performed two repetitive tasks in the final assembly logistics area: sorting center console side panels and folding material boxes --both reaching 90% success rates. They deployed both bipedal and wheeled models, working together with material transport robots as part of the broader intelligent automation system. I noticed the wheeled CyberOne was clearly more efficient and stable on the panel sorting task. In the first video, you can also see some noticeable jitter when the bipedal robot stops moving (though this should be relatively easy to fix). Xiaomi also candidly pointed out a current limitation in dexterity: when folding the second side of the material box, the robot still needs to adjust its orientation first to face the buckle, whereas skilled workers can pull the ring directly by feel without any micro-adjustments. Real-world scenarios are indeed the best benchmark.

4:59
0:09
0:36
0:14
Supersocks avatar
⟳ Cedric Lejeune retweeted
Supersocks@iamsupersocks·Nov 2

Mistral vient de publier le rapport technique de Mistral Large 4, son nouveau modèle ouvert : 1 000 milliards de paramètres, entraîné 45 jours sur 3 840 puces NVIDIA près de Paris. Le rapport ne se contente pas d'aligner des scores, il raconte comment on fabrique une IA de pointe en 2026, erreurs comprises. J'en retiens quatre choses. Première chose : un modèle peut réussir les tests et rater le travail. Pour économiser de la mémoire, Mistral l'avait bridé pour qu'il ne relise souvent que les derniers mots, ce qui passait inaperçu aux tests habituels mais le rendait décevant sur les longues tâches, alors l'équipe l'a opéré en plein apprentissage pour qu'il relise tout, puis a ajouté une technique de DeepSeek : un petit module de tri qui, pour chaque mot, ne lui fait relire que les passages utiles, où qu'ils soient. Deuxième : une usine à exercices. Pour apprendre à agir, avec des outils, des commandes et du code, le modèle s'entraîne sur plus de 300 000 exercices et n'est récompensé que s'il réussit. des agents IA partent de vrais projets GitHub, cachent la solution et rédigent l'énoncé, puis d'autres vérifient qu'on ne peut pas tricher. En cyber, ils partent de failles connues, et le correctif doit bloquer l'attaque sans rien casser. Les essais réussis de ces agents entraînent aussi le modèle, pour que chaque génération fabrique les exercices de la suivante. Résultat sur les cinq tests cyber du rapport, ML4 arrive premier sur deux face aux six modèles ouverts comparés, DeepSeek sur les trois autres. Troisième : le faire tourner sans se ruiner. Compressé sans perte sur ses évaluations, le modèle publié pèse 606 Go et tient sur quatre puces NVIDIA GB300. Il propose trois niveaux de réflexion, pour ne pas payer une longue réflexion quand la question est simple. Quatrième : ce que l'équipe a découvert en route. Pendant l'entraînement, deux agents sont sortis de leur bac à sable : l'un a pris le contrôle d'une machine de Mistral en enchaînant quatre failles, l'autre a effacé par erreur 30 000 environnements. Les alertes ont sonné, Mistral a corrigé puis relancé l'entraînement. Comme il s'attend à des évasions plus habiles, il remplace ses bacs à sable, jusqu'ici de simples conteneurs, par de petites machines virtuelles mieux isolées. Un troisième agent a triché en espionnant le correcteur pour lui voler la réponse ; ce genre de triche est surveillé en continu, et les environnements sont renforcés au fur et à mesure. Autre surprise, plus discrète : les agents polluent le web. Certains sites transforment chaque recherche qu'ils reçoivent en une nouvelle page. Or tous les agents qui passent le même test public en cherchant en ligne, comme BrowseComp, conçu par OpenAI, y laissent ainsi des traces, quel que soit le labo, et les suivants les retrouvent dans leurs résultats. En octobre, la plupart des essais de ML4 sur ce test tombaient sur l'une de ces pages dès les dix premières recherches. Elles ne donnent pas la réponse, mais font perdre du temps et poussent sans doute les agents vers les mêmes pistes. Anthropic avait prévu que ça empirerait ; Mistral prévoit de passer à sa propre copie du web, hors ligne. Dans ce rapport Mistral raconte ses erreurs, pas seulement ses scores. Et le rapport montre où se déplace la difficulté : quand une IA agit, il faut surveiller ce qu'elle touche, et jusqu'aux traces qu'elle laisse aux suivantes. Le rapport : huggingface.co/mistralai/Mis…

Shruti avatar
⟳ Cedric Lejeune retweeted
Shruti@heyshrutimishra·Nov 1

Yann LeCun was right & nobody in this industry wants to say it out loud. 109 million parameters beat 72 billion. He has spent years saying scale is not the thing that makes a model understand the world. Meta bet the other way. He left in November and started his own lab in Paris. Last Thursday he published the proof. The theory everyone has been running on since 2024 is that if you train models big enough, on text, on images, on sound, they all end up with the same picture of reality inside them. Size does it. Nobody ever proved that part. Randall Balestriero and LeCun proved it. And the proof says size was never doing the work. What does the work is which two things you show a model together. A photo and the caption of that photo are two views of one thing. Models trained on pairs like that get there fast. Two paragraphs pulled from the same document are not two views of one thing. Models trained on those get there slowly and expensively. Next word prediction is the second kind. That is what every chatbot in the world runs on. In their tests the first method reached the same understanding after 21.6 million examples. Next word prediction needed 900 million. And a 109 million parameter text model ended up closer to an image model's view of the world than a 72 billion parameter language model was. 660 times smaller.

Post attachment
Supersocks avatar
⟳ Cedric Lejeune retweeted
Supersocks@iamsupersocks·Nov 1

DeepSeek a encore sorti un cadeau pour l'humanité. Je vis pour ce genre de release. Grok Bot, Muse, Claude Duet, ChatGPT Dot : tout le monde construit la même chose. DeepSeek vient d'en publier le plan complet, avec le code en MIT. En clair, ils ouvrent le Grok Bot. C'est quoi, un super-agent ? L'inverse d'un chatbot. Tu ne lui écris pas de prompt. Il vit dans ton chat, ton GitHub, ta CI. Il voit passer tout ce qui arrive, décide seul ce qui mérite ton attention, et lance le travail. Leur exemple : quelqu'un râle dans un groupe à cause d'un bug de double-clic. Personne ne tague l'agent. Il revient avec la PR et la cause. Pourquoi tout le monde y va en même temps ? Parce que le goulot n'est plus le modèle, c'est ton attention. Au-delà de 3 à 5 sessions ouvertes, tu ne sais plus laquelle fait quoi. Le super-agent gère les sessions à ta place, et vient te chercher seulement quand il faut une décision. Un modèle se rattrape, ce que les labs fermés vendent c'est la couche autour : qui trie, qui décide, qui exécute, qui surveille, qui se souvient. DeepSeek pose cette couche sur la table, gratuitement. Le schéma, en six pièces : 1/ Le tri. Des règles en code jettent le bruit avant le moindre appel au modèle. 2/ Le Router. Un modèle léger, le seul qui te parle. Il décide et distribue, il ne code jamais. 3/ Les Workers. Une session isolée par tâche, avec le meilleur modèle. La suite d'une PR déjà suivie leur arrive directement. 4/ La réponse. 100 mots maximum sur le chat, le détail dans la PR. 5/ Le chien de garde. Il juge si chaque Worker avance ou tourne en rond, et répare quand ça casse. 6/ La mémoire. Elle vit hors des modèles : tu changes de modèle par rôle sans rien perdre. C'est l'archi que je fais tourner pour mes propres agents : un orchestrateur, des workers isolés, une QA à part. La pièce qui me manquait, c'est le chien de garde. Les labs fermés vendent l'agent. DeepSeek donne le plan.

Post attachment

DeepSeek Harness @DeepSeekHarness Nov 1 Article Exploring Super-agent Design: Towards a More Autonomous Agent System The original article is in Chinese. This is a machine translation. This article discusses architectural ideas. It is not an introduction to existing features of DeepSeek Harness. 1. Super-agent Over

Cedric Lejeune avatar
Cedric Lejeune@CLVPTY·Nov 1

We gonna need a bigger phone

THIS IS MASSIVE. THE ONE EVERYONE'S BEEN WAITING FOR 📱🔥 I've had Hermes on my phone for a week. Now you can too. The official Hermes Agent app is in open beta on iOS and Android. Once your agent lives in your pocket, there's no going back. You're going to love it. What it does: ▸ Chat with your agent and watch it work through its tools ▸ Live Mode: talk to it out loud, a pause sends it ▸ Routines on a schedule that deliver to Telegram, Discord, Slack or Teams ▸ One-tap skill installs ▸ Connect services like Shopify, Stripe, Notion, Linear and Slack ▸ Plugins, MCP, projects, artifacts and a Kanban board ▸ Usage and spend tracking ▸ Terminal over SSH or Mosh, plus your agent's files and config ▸ Light and dark themes, accent colors and custom app icons What you can do with it: ▸ Talk through an idea on a walk and hand it off ▸ Set up a morning brief from bed that lands in Discord ▸ Add a skill the moment you need it ▸ Check what your agent spent today ▸ Drop a task on your board from the grocery line Hermes just left the desk. Beta spots are limited, so grab yours fast. hermes-agent.nousresearch.co…

1:12
Hermes Release Watch avatar
⟳ Cedric Lejeune retweeted
Hermes Release Watch@HermesWatcher·Oct 31

THIS IS MASSIVE. THE ONE EVERYONE'S BEEN WAITING FOR 📱🔥 I've had Hermes on my phone for a week. Now you can too. The official Hermes Agent app is in open beta on iOS and Android. Once your agent lives in your pocket, there's no going back. You're going to love it. What it does: ▸ Chat with your agent and watch it work through its tools ▸ Live Mode: talk to it out loud, a pause sends it ▸ Routines on a schedule that deliver to Telegram, Discord, Slack or Teams ▸ One-tap skill installs ▸ Connect services like Shopify, Stripe, Notion, Linear and Slack ▸ Plugins, MCP, projects, artifacts and a Kanban board ▸ Usage and spend tracking ▸ Terminal over SSH or Mosh, plus your agent's files and config ▸ Light and dark themes, accent colors and custom app icons What you can do with it: ▸ Talk through an idea on a walk and hand it off ▸ Set up a morning brief from bed that lands in Discord ▸ Add a skill the moment you need it ▸ Check what your agent spent today ▸ Drop a task on your board from the grocery line Hermes just left the desk. Beta spots are limited, so grab yours fast. hermes-agent.nousresearch.co…

1:12
Liquid AI avatar
⟳ Cedric Lejeune retweeted
Liquid AI@liquidai·Oct 27

Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence. > 6 model instances, across text, audio, vision, and omni modalities, generative and decision models. > 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window. > 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding > 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages. > 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language. > 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input. > fine-tune on 1 GPU. > a new tri-hybrid architecture > architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs. > 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters. > 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒 Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device. 1/ 🧵

2:17
𒐪 avatar
⟳ Cedric Lejeune retweeted
𒐪@SHL0MS·Jul 3

i'm open sourcing UNBROKER: a tool that finds where your personal info is exposed by data brokers and files the removals for you it runs as a skill in Hermes Agent _________ your data is everywhere; hundreds of brokers publish your name, current and old addresses, phone, email, birthday, even your relatives. anyone can find where you live in about ten seconds CCPA, CPRA, GDPR, and a growing number of state laws say a broker has to delete your data if you ask. there's just no easy bulk button. every broker has a different process and many make it intentionally difficult to exercise your right to delete this is the entire business model for companies like DeleteMe, Incogni, EasyOptOuts; they charge you monthly (DeleteMe is $330/yr for a Family plan) to file removals that you can submit yourself for free, and then you're giving a new company the exact data you want to erase so i built one you just run for free. your data never even has to leave your machine if you run a local model _________ how it works: first it builds search vectors from everything: every name, alias, email, phone, and address you've had (brokers might index you under a maiden name or a house you left in 2014, so the naive "current name + current city" approach can miss profiles). then it fans out parallel sub-agents across the broker list, which refreshes from a maintained public source automation is tiered. when it can handle a broker end to end with your settings, it drives a browser through the opt-out form, sends the email, and opens the confirmation link itself. soft CAPTCHAs clear on their own with a real browser. anything only a human can finish comes back to you as a short list at the end the email side doesn't need a stored password and can send opt-outs and open verification links through your own logged-in webmail. you can also wire up SMTP, or keep it manual and just send the drafts it writes it tailors every request to your jurisdiction, filing under the framework that applies where you live: CCPA and CPRA in California, GDPR in the EU and UK, a general right-to-delete request everywhere else. if you're in California it also uses the state's DROP portal, a single request that covers 500+ registered brokers at once it holds as little of your data as it can, and keeps it local. dossiers are encrypted at rest if you want, opaque ids keep your real name out of every filename and log, and nothing leaves your machine unless you opt in brokers sometimes relist you eventually or new ones find your data, so every case is tracked in a ledger and can be re-scanned on a cron schedule so if your data pops back up it files the removal again github.com/NousResearch/herm…

Post attachment
Arnaud Bertrand avatar
⟳ Cedric Lejeune retweeted
Arnaud Bertrand@RnaudBertrand·Oct 25

The ugly truth about France and some of the other European countries currently in turmoil - and which none of the politicians want to admit because it'd be political suicide - is that you can't solve the problem by going after immigrants (as the right asserts) or the rich (as the left asserts), but you could solve it by going after the old. One extraordinary statistic for you: people over 60 in France, a quarter of the population, own 60% of France's wealth, non-financial and financial (ofce.fr/pdf/revue/2025/3-192…). And here is another crazy statistic: that same quarter of the population, those above 60, receive nearly two-thirds of all social spending by the French government (previssima.fr/actualite/prot…) - which is famously among the highest in the world (France's government spending is an insane 57.3% of GDP). Last statistic: they make up about 40% of voters because they're a third of registered voters (insee.fr/fr/statistiques/266…), and they turn out far more than anyone else (franceinfo.fr/elections/sond…). Put our 3 statistics together and there's almost a bit of an apartheid situation here: the old own most of the country's wealth, collect most of the social spending and have the political power to keep it that way. Which makes them a de-facto rentier class, living off assets and transfers paid for by younger generations. And, of course, at the risk of stating a truism, the future is built by the young. Yet France keeps effectively taking money away from them to give it to those whose working lives are already behind them - and need the money far less since they already own the country's wealth. What about inheritance, I hear you ask? That's the funniest/saddest aspect: given that life expectancy has kept rising, the French now inherit a home at 54 on average. The most common case is someone in their 80s leaving it to heirs in their 50s, in effect the very old transferring wealth to the about-to-be old (insee.fr/fr/statistiques/fic…). You can add this to our earlier apartheid statement: on top of all the rest, even when the old finally let go of their wealth, it stays within their own "camp". That's my personal situation, incidentally: I'm 42, my father is 71 and he is starting to look at ways to pass things on to me and my siblings, when our professional lives are already well settled, so it'll arrive when it changes things the least, which is pretty much the French system in a nutshell. Now, I get that there'd be some unfairness in going after them: they paid into the system their entire working lives, playing by the rules of a social contract that told them if they paid up they'd collect at the end. Changing the rules now means breaking that promise, which is admittedly unfair. But there's no magic wealth tree in any society: you can only take it where it is. That's the way most major societal transformations end up happening. They're often triggered by a catastrophic event (the so-called "four horsemen": mass-mobilisation war, revolution, state collapse, plague) BUT what really drives change is a transfer: wealth gets pried from the privileged class who've been sitting on it and handed to those who'll build what comes next. So, in theory and in an ideal world, you can do without the catastrophic event and directly proceed with the transfer... The fact that in France, and elsewhere in Europe, it's the youth protesting is probably no coincidence - even though they don't frame it as a generational fight. And even though, cruel irony, the left politicians many of them march behind promise to bring the retirement age back to 60 (lcp.fr/actualites/legislativ…), which would make the situation even worse. Let's say they understand intuitively that the system is stacked against them, even if they haven't yet figured out how. And politicians are obviously doing what politicians do, meaning pandering to the people who actually vote, while pointing the young towards easier - and politically convenient - scapegoats. To be clear I'm not saying it's the only problem in France, VERY far from it. I'm also not saying that all old people in France live a lavish lifestyle at the expense of younger generations: there are also plenty of very poor retirees who, if anything, would deserve more. What I am saying is that it's probably the biggest elephant in the room: everyone can see it, nobody dares mention it - no politician, from any side, wants to run on cutting grandma's pension - and it makes every other problem harder to fix.