#mistral — Public Fediverse posts

#ShaiHulud attack ships signed malicious #TanStack, #Mistral #npm packages

https://www.bleepingcomputer.com/news/security/shai-hulud-attack-ships-signed-malicious-tanstack-mistral-npm-packages/

#cybersecurity

#ShaiHulud attack ships signed malicious #TanStack, #Mistral #npm packages

https://www.bleepingcomputer.com/news/security/shai-hulud-attack-ships-signed-malicious-tanstack-mistral-npm-packages/

#cybersecurity

#ShaiHulud attack ships signed malicious #TanStack, #Mistral #npm packages

https://www.bleepingcomputer.com/news/security/shai-hulud-attack-ships-signed-malicious-tanstack-mistral-npm-packages/

#cybersecurity

#ShaiHulud attack ships signed malicious #TanStack, #Mistral #npm packages

https://www.bleepingcomputer.com/news/security/shai-hulud-attack-ships-signed-malicious-tanstack-mistral-npm-packages/

#cybersecurity

#ShaiHulud attack ships signed malicious #TanStack, #Mistral #npm packages

https://www.bleepingcomputer.com/news/security/shai-hulud-attack-ships-signed-malicious-tanstack-mistral-npm-packages/

#cybersecurity

@ErikJonker Correct, it is even a bit cheaper per month compared to Claude or Gemini but it misses text-to-video atm. I need to validate how #Mistral performs to support coding.

#TODO

https://www.europesays.com/news/28077/ SAP Unveils the Autonomous Enterprise | SAP Sapphire #Accenture #Anthropic #AutonomousEnterprise #aws #Cohere #Conduct #GoogleCloud #Headlines #IndustryAI #JouleAgents #JouleAssistants #JouleStudio #JouleWork #LiveKit #Microsoft #mistral #n8n #News #Nvidia #palantir #Parloa #RISEWithSAP #SAPAutonomousSuite #SAPBTP #SAPBusinessAIPlatform #SAPBusinessDataCloud(SAPBDC) #SAPGROW #TopStories

How it's going #leanstral #lean #mistral

Look how it is offering to prove Fermat's last theorem!

Habe heute entdeckt, dass in der Fakultät ein bisher ungenutzter Server mit 2x NVIDIA V100 32GB stand. Und nach einem Treiberupdate lief darauf auch das #leanstral Modell mit vernünftiger Geschwindigkeit.

Habe jetzt die 4-Bit Quantisierung genommen und dann passt es mehr oder weniger in den VRAM. Die Kiste hat 1TB normalen RAM, also kein Problem hier.

Bin gespannt, was das kann!

https://huggingface.co/jackcloudman/Leanstral-2603-GGUF

#lean #mistral

Marseille. Le trafic des trains SNCF stoppé, le toit de ce bâtiment menace de s’envoler

Par Fabien Binacchi Publié le 12 mai 2026 à 14h22 Pas de bagage abandonné, ni de problème technique.…
#Marseille #FR #France #Actu #News #Europe #EU #actu #Actualités #europe #Mistral(vent) #Provence-Alpes-Côted'Azur #Républiquefrançaise #SNCF #Société #Train
https://www.europesays.com/fr/927919/

https://www.europesays.com/fr/927919/ Marseille. Le trafic des trains SNCF stoppé, le toit de ce bâtiment menace de s’envoler #actu #Actualités #EU #europe #FR #France #Marseille #Mistral(vent) #News #ProvenceAlpesCôteD'Azur #RépubliqueFrançaise #SNCF #Société #Train

And more than code generation, folks at Mistral have achieved some amazing results for things like speech to text, video analysis and even small conversational models running entirely on webGPU - this works even on recent iPhones! https://huggingface.co/mistralai/spaces

#Mistral #webGPU

Le Chat, Mistral

Has anyone else experienced issues with the Le Chat page loading (not necessarily right now, but in general)? I suspect they have a problematic CDN and/or poorly written JavaScript. The page often simply hangs, loading indefinitely without displaying anything. It might even be a redirect loop, because at times, it feels like it disrupts other open tabs until I close the Mistral page. https://chat.mistral.ai/chat
#lechat #mistral #ai #problems #website

Le Chat, Mistral

Has anyone else experienced issues with the Le Chat page loading (not necessarily right now, but in general)? I suspect they have a problematic CDN and/or poorly written JavaScript. The page often simply hangs, loading indefinitely without displaying anything. It might even be a redirect loop, because at times, it feels like it disrupts other open tabs until I close the Mistral page. https://chat.mistral.ai/chat
#lechat #mistral #ai #problems #website

Le Chat, Mistral

Has anyone else experienced issues with the Le Chat page loading (not necessarily right now, but in general)? I suspect they have a problematic CDN and/or poorly written JavaScript. The page often simply hangs, loading indefinitely without displaying anything. It might even be a redirect loop, because at times, it feels like it disrupts other open tabs until I close the Mistral page. https://chat.mistral.ai/chat
#lechat #mistral #ai #problems #website

Le Chat, Mistral

Has anyone else experienced issues with the Le Chat page loading (not necessarily right now, but in general)? I suspect they have a problematic CDN and/or poorly written JavaScript. The page often simply hangs, loading indefinitely without displaying anything. It might even be a redirect loop, because at times, it feels like it disrupts other open tabs until I close the Mistral page. https://chat.mistral.ai/chat
#lechat #mistral #ai #problems #website

Le Chat, Mistral

Has anyone else experienced issues with the Le Chat page loading (not necessarily right now, but in general)? I suspect they have a problematic CDN and/or poorly written JavaScript. The page often simply hangs, loading indefinitely without displaying anything. It might even be a redirect loop, because at times, it feels like it disrupts other open tabs until I close the Mistral page. https://chat.mistral.ai/chat
#lechat #mistral #ai #problems #website

#MistralAI has fucking beautiful docs 😍

#ai #mistral

#MistralAI has fucking beautiful docs 😍

#ai #mistral

@toschi Ich habe meinen Account und die App gelöscht.

Ich benutze Le Chat von Mistral, das auch nicht amerikanisch ist.

https://mistral.ai/products/le-chat

#LeChat #Mistral #MistralAI #KI

@toschi Ich habe meinen Account und die App gelöscht.

Ich benutze Le Chat von Mistral, das auch nicht amerikanisch ist.

https://mistral.ai/products/le-chat

#LeChat #Mistral #MistralAI #KI

https://winbuzzer.com/2026/05/02/mistral-medium-3-5-unified-flagship-chat-reasoning-code-xcxwbn/

Mistral Medium 3.5 Folds Chat, Reasoning, and Code Into One 128B AI Model

#AI #MistralMedium35 #Mistral #LeChat #AIReasoningModels #AIAgents #AgenticAI #AICoding #AIBenchmarks #OpenSourceAI #EnterpriseAI

Mistral apresenta o novo Medium 3.5 que unifica chat e código num único modelo
🔗 https://tugatech.com.pt/t82652-mistral-apresenta-o-novo-medium-3-5-que-unifica-chat-e-codigo-num-unico-modelo

#chat #mistral #modelo 

« Pour l'instant, force est de constater que les garde-fous manquent »

L'intelligence artificielle française Mistral épinglée pour avoir relayé de la désinformation russe, chinoise ou iranienne

https://www.franceinfo.fr/internet/intelligence-artificielle/la-start-up-francaise-d-intelligence-artificielle-mistral-relaie-de-la-desinformation-russe-chinoise-ou-iranienne_7969925.html

#mistral #mistralAI

« Pour l'instant, force est de constater que les garde-fous manquent »

L'intelligence artificielle française Mistral épinglée pour avoir relayé de la désinformation russe, chinoise ou iranienne

https://www.franceinfo.fr/internet/intelligence-artificielle/la-start-up-francaise-d-intelligence-artificielle-mistral-relaie-de-la-desinformation-russe-chinoise-ou-iranienne_7969925.html

#mistral #mistralAI

Omg, #Mistral released Mistral Medium 3.5 yesterday, and it's so good! Already fully switched to it as my main model. Some benchmarks apparently suggest it's comparable to Sonnet 4.5. Let's support #European tech! 🇫🇷🇪🇺

https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5

#AI #MistralAI #coding #software #technology

Omg, #Mistral released Mistral Medium 3.5 yesterday, and it's so good! Already fully switched to it as my main model. Some benchmarks apparently suggest it's comparable to Sonnet 4.5. Let's support #European tech! 🇫🇷🇪🇺

https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5

#AI #MistralAI #coding #software #technology

Mistral AI veröffentlicht Mistral Medium 3.5 mit 128 Milliarden Parametern, 256.000 Token Kontext und asynchronen Remote Agents für Coding-Aufgaben.

Das Modell läuft auf vier GPUs und ist On-Premise hostbar. Open Weights stehen unter einer modifizierten MIT-Lizenz bereit. In Benchmarks wie SWE-Bench Verified (77,6%) übertrifft es Claude Sonnet 4.5.

#Mistral #LLM #OpenWeights #MistralAI #AIGeneratedImage

https://www.all-ai.de/news/news26top/mistral-medium-3-5-flagschiff

Mistral AI veröffentlicht Mistral Medium 3.5 mit 128 Milliarden Parametern, 256.000 Token Kontext und asynchronen Remote Agents für Coding-Aufgaben.

Das Modell läuft auf vier GPUs und ist On-Premise hostbar. Open Weights stehen unter einer modifizierten MIT-Lizenz bereit. In Benchmarks wie SWE-Bench Verified (77,6%) übertrifft es Claude Sonnet 4.5.

#Mistral #LLM #OpenWeights #MistralAI #AIGeneratedImage

https://www.all-ai.de/news/news26top/mistral-medium-3-5-flagschiff

https://blog.gslin.org/archives/2026/04/30/13017/mistral-%e6%8e%a8%e5%87%ba%e4%ba%86%e8%b2%b4%e4%b8%8d%e5%b0%91%e7%9a%84-mistral-medium-3-5/

Mistral 推出了貴不少的 Mistral Medium 3.5

#35 #ai #language #large #llm #medium #mistral #model

https://blog.gslin.org/archives/2026/04/30/13017/mistral-%e6%8e%a8%e5%87%ba%e4%ba%86%e8%b2%b4%e4%b8%8d%e5%b0%91%e7%9a%84-mistral-medium-3-5/

Mistral 推出了貴不少的 Mistral Medium 3.5

#35 #ai #language #large #llm #medium #mistral #model

https://blog.gslin.org/archives/2026/04/30/13017/mistral-%e6%8e%a8%e5%87%ba%e4%ba%86%e8%b2%b4%e4%b8%8d%e5%b0%91%e7%9a%84-mistral-medium-3-5/

Mistral 推出了貴不少的 Mistral Medium 3.5

#35 #ai #language #large #llm #medium #mistral #model

Mistral Medium 3.5

https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5

#HackerNews #Mistral #Medium #3.5 #MistralAI #RemoteAgents #AItechnology #MachineLearning

Mistral Medium 3.5

https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5

#HackerNews #Mistral #Medium #3.5 #MistralAI #RemoteAgents #AItechnology #MachineLearning

@ennopark sorry, ich hatte #Mistral nicht gesagt, dass #Immobilien auch #Vermögenswerte sind.

Und natürlich muss die aktuelle Regelung zur #Grundsteuer Anwendung finden.

Da ist die #KI ganz alleine drauf gekommen.

Also @ennopark - ja, die KI kann das!

Mistral AI lança Connectors no Studio com acesso total via API e novos fluxos de controlo humano
🔗 https://tugatech.com.pt/t82010-mistral-ai-lanca-connectors-no-studio-com-acesso-total-via-api-e-novos-fluxos-de-controlo-humano

#ai #api #controlo #lan #mistral #studio 

European AI. A playbook to own it. By Mistral

https://europe.mistral.ai/

#HackerNews #EuropeanAI #Mistral #AIPlaybook #TechInnovation #AICommunity

Arthur #Mensch, patron de #Mistral #AI, héraut d’une intelligence artificielle ouverte et souveraine https://www.lemonde.fr/economie/article/2026/03/30/arthur-mensch-patron-de-mistral-ai-heraut-d-une-intelligence-artificielle-ouverte-et-souveraine_6675432_3234.html

Platforma Forge od Mistral AI ukazuje, kam se posouvá využití umělé inteligence ve firmách. Místo napojení na externí služby přináší možnost stavět vlastní AI modely nad interními daty a provozovat je pod plnou kontrolou – od trénování až po nasazení.

#Mistral #Forge : l’offensive de #Mistral #AI pour rendre l’;#IA d’ #entreprise réellement #souveraine

https://www.zdnet.fr/actualites/mistral-forge-loffensive-de-mistral-ai-pour-rendre-lia-dentreprise-reellement-souveraine-492045.htm

#Mistral #Forge : l’offensive de #Mistral #AI pour rendre l’;#IA d’ #entreprise réellement #souveraine www.zdnet.fr/actualites/m...

Mistral Forge : l’offensive de...

The internet(tm) claims that DevStral2 would be almost on par with Claude Sonnet 4.5.
Also, open weights model with 120B, i.e. reasonable on current om-prem hardware a la AMDs 395+.
Prices are also pretty ok if hosted, also: EU hostable.

Did anyone compare that thing to the above Claude model, ideally using Claude Code as control system?
I'd also be interested in on-prem reports, ideally also with one of the AMD 395+ UMA boxes.

#llm #claudecode #claude_sonnet_45 #sonnet45 #devstral #devstral2 #mistral #aicoding #uma #amd #amdstrixhalo

#Mistral #AI, le champion #français de l’ #IA #générative, accusé d’avoir pillé des #œuvres pour son #IA

En même temps le reste du môde l'a fait sans demander à personne.. et personne ́n'a rien dit.. difficile de rattraper un retard sans grandes #bdd #db, non?

https://www.01net.com/actualites/mistral-ai-le-champion-francais-de-lia-generative-accuse-davoir-pille-des-oeuvres-pour-son-ia.html

@GavinCampbell same for me. Once I tried Claude Code → OpenCode → Rush → OpenCode, I stayed with OpenCode. It's very difficult to go back to Claude after getting used to OpenCode's advantages, especially with regard to using local models.

Also, like many people in Europe, I prefer to move off anything USA-based ASAP. I've been testing #Mistral (very generous free tier) and I'm impressed. For general local tasks I use OLMo 3 via #Ollama.

Highly recommend Mistral. In various contexts it seemed to me to be far more useful and accurate than #Claude, which is making a lot of stupid mistakes lately.

#Anthropic #Palantir #NerdReich

@matthiasgiger Wenn du bei connector.php in Z. 45 den Code im Bild 1 ergänzt, kann das Modell auch Fragen für die Questionbank generieren. Bild 2 zeigt Fragetypen und Quellen, wobei Mistral keine Dateien annimmt, so dass "Kursinhalte" kaum funktionieren und "inhalte" sich auf reine Texteingabe beschränken werden.
#isb #moodle #mistral #ki #ai #europe

🎯 Zero accuracy loss - preserves what matters: errors, anomalies, high-scoring items & query-relevant content using BM25/embedding similarity

✅ Full provider support: #OpenAI, #Anthropic, #Google, #Cohere, #Mistral & #LiteLLM with exact token counting

📈 Performance: Search results (1000 items) 45K→4.5K tokens (90%), Log analysis 22K→3.3K tokens (85%), API responses 15K→2.25K tokens (85%)

For my personal old Macbook Air M1 16Gb I found an LLM model that fits and performs reasonably well for my simple use cases. It's Mistral's `ministral-3:8b` — it gives an average 10 tokens per second. I use it in conjunction with Raycast AI chat app #llms #ai #genai #localai #localllm #raycast #mistral

La #start-up #française #Mistral se lance dans la course au #vibe-coding et publie #Devstral 2 sous une #licence #MIT modifiée? une nouvelle génération de son modèle d' #IA #AI conçu pour le #codage #dev #tech #FrenchTech #frenchtech #france intelligence-artificielle.developpez.com/actu/378346/...

La start-up fran�aise Mistral ...

#Mistral AI launched #Devstral2, a new #codingmodel, and #MistralVibe, a command-line interface for #codeautomation. Devstral 2, available in two sizes, emphasises context awareness and production-grade workflows. Mistral is valued at €11.7 billion and partnered with Kilo Code and Cline to release Devstral 2. https://techcrunch.com/2025/12/09/mistral-ai-surfs-vibe-coding-tailwinds-with-new-coding-models/?eicker.news #tech #media #news

Cisco ostrzega: otwartym modelom AI brakuje cierpliwości. Długie rozmowy łamią ich zabezpieczenia

Najnowsza analiza przeprowadzona przez zespół Cisco AI Defense ujawnia niepokojącą lukę w bezpieczeństwie popularnych, otwartych modeli językowych (LLM).

Okazuje się, że aby „złamać” sztuczną inteligencję i zmusić ją do wygenerowania szkodliwych treści, nie są potrzebne zaawansowane umiejętności hakerskie, lecz jedynie cierpliwość. Badanie wykazało, że modele, które skutecznie blokują niebezpieczne zapytania w pojedynczych interakcjach, masowo kapitulują podczas dłuższych, wieloetapowych konwersacji.

Popularność modeli typu open-weight, które można pobrać i uruchomić na własnej infrastrukturze, rośnie lawinowo – do sierpnia 2025 roku pobrano je z platformy HuggingFace już około 400 milionów razy. Cisco przetestowało osiem wiodących rozwiązań tego typu, w tym modele od Alibaby, Google, Mety, Microsoftu, Mistrala czy OpenAI, stosując metodę „black box”, czyli bez wglądu w ich wewnętrzną architekturę.

Dialog, który usypia czujność

Kluczowym wnioskiem z raportu jest drastyczna różnica w skuteczności zabezpieczeń między atakami jednorazowymi a tzw. atakami multi-turn (wieloetapowymi). W krótkich wymianach zdań modele zazwyczaj radzą sobie dobrze, konsekwentnie odmawiając współpracy przy szkodliwych żądaniach. Jednak w trakcie trwania długiego dialogu, systemy te stopniowo tracą swoje „hamulce” bezpieczeństwa.

Skala tego zjawiska jest uderzająca, co doskonale obrazują dane dotyczące konkretnych modeli. Przykładowo, w przypadku modelu Mistral Large-2, skuteczność ataku przy pojedynczym zapytaniu wynosiła zaledwie 22 proc., ale przy zastosowaniu strategii wieloetapowej rozmowy wzrosła aż do szokujących 92,8 proc.. Podobną tendencję zaobserwowano w modelu Meta Llama 3.3, gdzie skuteczność ataku skoczyła z niespełna 17 proc. do 87 proc.. Najbardziej odpornym na manipulację okazał się model Google Gemma 3-1B-IT, choć i w jego przypadku skuteczność ataków w dłuższej rozmowie wyniosła niemal 26 proc.

Skąd biorą się te różnice?

Eksperci Cisco zauważają, że podatność na manipulację jest ściśle powiązana ze strategią strojenia modelu (alignment). Modele zaprojektowane z myślą o maksymalizacji możliwości i elastyczności są znacznie łatwiejsze do „zmanipulowania” w toku rozmowy niż te, w których priorytetem od początku było rygorystyczne bezpieczeństwo. Aby przeciwdziałać tym zagrożeniom, organizacje wdrażające AI powinny inwestować w rozwiązania takie jak trening przeciwstawny (Adversarial Training) oraz systemy monitorowania interakcji w czasie rzeczywistym.

Sieci Wi-Fi 7 w firmach i zarządzanie „na klik”. Cisco zbroi się na erę AgenticOps

#aiSi #cisco #cyberbezpieczenstwo #hacking #llama #llm #mistral #news #openSource