Global AI Community Virtuelle Francophone

*FR* Une bonne soirée "Meet & Greet" pour bien lancer le #MicrosoftAITour Paris qui aura lieu demain ! Merci à tous les participant(e)s de leur bonne humeur et leur partage ! Et un merci particulier à Henk Boelman d'avoir pu rendre possible cet événement. A demain pour le #MicrosoftAITour Paris ! *EN* A good starter before #MicrosoftAITour Paris (tomorrow), with our "Meet & Greet" event tonight! And special thanks to Henk Boelman for making it possible. See you tomorrow at the #MicrosoftAITour Paris! Nicolas & Paul for the Global AI Community Virtuelle francophone (which was not virtual and not really francophone for this event!). I apologize if I missed some names in the photo, don't hesitate to add tags!)

Opensource LLM platform by Microsoft! I’m still buzzing from all the latest AI developments I saw at Gitex, and now Microsoft! bitnet.cpp is their new open-source framework designed to run LLMs more efficiently. The special part is it uses 1-bit quantization - meaning it shrinks down the model's size by up to 32 times compared to traditional methods. This makes AI models way faster and lighter, even on regular CPUs (no GPUs required). Basically, instead of using 16-bit or 32-bit values for calculations, bitnet.cpp compresses everything into 1-bit, saving tons of memory and energy without losing much performance. It is great for running things like real-time text generation or AI-powered applications on regular hardware. I love the new possibilities it brings, especially for industries and environments that can’t afford high-end infrastructure. It makes powerful AI more accessible - no need for a supercomputer to work with large models anymore. Github https://lnkd.in/g54WnH4Z __________________ I share my learning journey here. Join me and let's grow together. For more on AI and learning materials, please check my previous posts. Alex Wang #llms #artificialintelligence #innovation #llms #generativeai #opensource

[GAIC-vf] Le NLP au service de l'assurance IA chez Covéa - le NLP au service de l'assurance par Phileas CONDEMINE Stack technique, modèles testés et retenus, librairies Python utilisées, étapes clefs des projets et métriques de performance Evaluation: les LLM sont-ils juges et parties ? par Paul PETON Présentation des techniques de LLM as a judge

Meet & Greet ! Nous sommes heureux de pouvoir vous proposer une soirée en préambule à l'AI Tour Paris de #Microsoft du 22 octobre prochain. La soirée aura donc lieu le lundi 21, à partir de 18h30, au Doppio Malto La Défense. Nous accueillerons plusieurs speakers intervenus au cours de l'année ainsi que Henk Boelman qui organise la communauté mondiale. Pas de slides, pas de démos, uniquement un temps convivial d'échanges informels et de réseau :) Attention, le nombre de places est (très !) limité, donc veillez à bien réserver uniquement si vous êtes certain.e de venir. Nous engageons en effet des dépenses vous permettant de profiter d'un buffet offert. https://lnkd.in/ePu64WcN.. cc ♾ Nicolas Robert, Randa MEZRAG, Henk Boelman

#Microsoft #Copilot is available in #WhatsApp ! https://lnkd.in/eNN3QtTi

Today we launched GPT-4o-Realtime-Preview in #AzureOpenAI Service. This takes voice capabilities to the next level, enabling real-time, natural voice interactions that go beyond traditional text-based #AI.    Customers that have already been using it have seen:    ✅ Faster responses: Voice responses significantly faster than many traditional text-to-speech engines, leading to reduced latency and smoother interactions ✅ Natural conversations: Minimized robotic tone often associated with AI-generated speech, making conversations sound more engaging ✅ Multilingual support: The support of multiple languages, allowing for natural, multilingual conversations    This opens up endless possibilities for developers and AI experiences, from virtual assistants to real-time customer support. AND it's integrated with #Copilot, as part of the new #CopilotVoice product announced today!    Check it out in action below and read more here: https://lnkd.in/gf2EyK_E

Today we announced general availability of fine-tuning for GPT-4o and GPT-4o mini on #AzureOpenAIService. Fine-tuning customization is faster, cheaper and more reliable than training models from scratch.   A great example in the blog below is Lionbridge's fine-tuned #GPT models that outperformed both baseline GPT and Neural Machine Translation engines in languages like Spanish, German, and Japanese.   In addition to fine-tuning we also introduced:   ✅ New Models: Including Phi-3.5-MoE, Phi-3.5-vision, Meta’s Llama 3.2, SDAIA’s ALLaM-2-7B, and upcoming updated Command R and Command R+ from Cohere ✅ Enterprise Features: Upcoming Azure OpenAI Data Zones for enhanced data privacy and security, and full network isolation support in Azure AI Studio ✅ Responsible AI Enhancements: New capabilities in Azure AI Content Safety, including Correction for hallucination issues and Protected Material Detection for Code   Read more here and get started today: https://lnkd.in/gFw5y2aV   #AzureAI #Finetuning #AIInnovation 

[GAIC-fr] Approche Text-to-SQL basée sur Azure AI Search et Vanna.ai Découvrez comment nous développons une approche innovante Text-to-SQL pour des bases de données complexes en utilisant plusieurs appels LLM, Azure AI Search et Vanna.ai.

🔥Watermarks in preview in Azure OpenAI Service Microsoft is proud to announce the rollout of a new built-in feature in Azure OpenAI Service. ‘Watermarks’ add invisible watermarks to all images generated using DALL·E, the company’s flagship generative AI image generator. This watermarking technology is designed to provide an additional layer of transparency and disclosure to AI-generated content. https://lnkd.in/eXS8PRT2 #Azure #AI #AzureML #DataScience #Microsoft #AIGen #openai #AzureopenAI

#Risk and #safety #evaluations for #indirect #prompt #injection #attacks are now available in #public #preview, accessible through #Azure #AI #Studio UI and SDK experiences. Indirect prompt injection attacks (also known as cross-domain prompt injection attacks or #XPIA) are an emerging attack vector where a threat actor poisons a model’s grounding data source, such as a public website, email, or internal document, to pass hidden, malicious instructions to a model and circumvent safety guardrails. With the Azure AI Evaluation SDK, users can now simulate indirect prompt injection attacks on their generative AI model or application and measure how often their AI fails to detect and deflect the attacks (the defect rate) along subcategories of manipulated content, intrusion, and information gathering. Users can also drill into evaluation details to better-understand how their application typically responds to these attacks and the associated risks. #Security #ResponsibleAI #Microsoft #GenerativeAI https://lnkd.in/evN_vG-R