talk-data.com

À mesure que les systèmes d’IA se développent et que les discussions autour de l’intelligence artificielle générale (AGI) s’intensifient, la question de l’open source devient un sujet central — et de plus en plus controversé. Est-ce un garde-fou essentiel pour garantir transparence, innovation et contrôle démocratique ? Ou bien un facteur de risque face à des technologies trop puissantes et imprévisibles ("mettre le bouton rouge dans la main de tous") ?

Nos invités : - Thomas Scialom (Chercheur en IA Générale chez Meta / Créateur de Llama-2) - Siméon Campos (Expert IA à l'OCDE / Fondateur de SaferAI) - Joséphine Boone (Journaliste Tech & IA chez Les Echos)

Le débat sera suivi d'un temps de Q/R et d'un cocktail dans les magnifiques locaux d'Ekimetrics.

Si vous êtes sur liste d'attente, vous pourrez tout de même visionner le débat en live. Lien vers le live (ouverture à partir de 19h30) : https://tinyurl.com/ya3dd837

Meta has been at the absolute edge of the open-source AI ecosystem, and with the recent release of Llama 3.1, they have officially created the largest open-source model to date. So, what's the secret behind the performance gains of Llama 3.1? What will the future of open-source AI look like? Thomas Scialom is a Senior Staff Research Scientist (LLMs) at Meta AI, and is one of the co-creators of the Llama family of models. Prior to joining Meta, Thomas worked as a Teacher, Lecturer, Speaker and Quant Trading Researcher.  In the episode, Adel and Thomas explore Llama 405B it’s new features and improved performance, the challenges in training LLMs, best practices for training LLMs, pre and post-training processes, the future of LLMs and AI, open vs closed-sources models, the GenAI landscape, scalability of AI models, current research and future trends and much more.  Links Mentioned in the Show: Meta - Introducing Llama 3.1: Our most capable models to dateDownload the Llama Models[Course] Working with Llama 3[Skill Track] Developing AI ApplicationsRelated Episode: Creating Custom LLMs with Vincent Granville, Founder, CEO & Chief Al Scientist at GenAltechLab.comRewatch sessions from RADAR: AI Edition New to DataCamp? Learn on the go using the DataCamp mobile appEmpower your business with world-class data and AI skills with DataCamp for business

Join us for "Unlocking Llama 3: An Interactive Q&A with Meta's AI Expert Thomas Scialom" a captivating session where you'll have the unique opportunity to engage directly with one of the technical leaders behind Meta's groundbreaking Llama models. This event promises an open forum to explore the intricate capabilities and future potential of generative AI technologies, offering insights straight from the forefront of AI innovation. Whether you're a developer, researcher, or simply an AI enthusiast, come equipped with your questions and ready to delve deep into the world of advanced machine learning.

To be sure we can address your questions, due to time limitations, please send them in advance here: https://forms.gle/gbpRw5XBFt9qJfxg7

This event is in-person only and will be followed by a networking apéro. We are looking forward to seeing you all in person!

EDIT : We are unfortunately closing inscription due to high demand, thank you for your understanding

Due to the Meetup's success, we'd like to inform you that the available places for the event are limited to the first 100 people to show up in order to ensure the safety and comfort of all attendees (we are closing registration at 250 people to anticipate no-shows) . Thank you for your understanding !

*** Thomas Scialom - Research Scientist (LLMs) - Meta AI Title: LLMs: past present and future Summary : Thomas will present the brief history of LLMs from GPTs to the lattest fronteer models, before diving into the science beyind RLHF, the technology powering ChatGPT and Llama-2. Finally, we will present his perspective on what could be the future of the field.

*** Francois Role - Université Paris Cité Title: Aligning Text and Image Representations Using Vision-Language Pretrained Models Summary : Data related to the same topic often comes in many modalities (audio, image, text, etc.) The ability to bridge the gap between these modalities is what makes possible applications such as multimodal information-retrieval, multimodal classification, automatic image captioning, etc. In this talk, we will present Vision-Language Pretrained Models (VLP models) that have been designed to jointly encode vision and langage, with a focus on bidirectional contrastive learning. We will first explain the loss function used in this context, starting from an intuitive example before presenting its formal definition. It will then be shown that training methods based on this kind of loss, while very useful, do not necessarily lead to an optimal alignment of the different modalities. We will therefore present a method for improving the quality of the text and image representations produced by the VLP models.