talk-data.com

📍Doctrine office, 190 Rue Championnet, 75018 Paris

📆 November 5th, 7:30 p.m.

👥 Boris Toledano, COO / Co-founder @ LinkUp ➡️ Beyond Web Search: Real-Time Web Intelligence for AI-Native Agents Summary: AI agents evolve. They need more than legacy search engines optimized for human clicks and ad revenue. The next frontier is AI-native web search infrastructure: systems designed to retrieve and integrate web intelligence directly into LLM-based applications. At Linkup, we're building this next generation of search by indexing the web for agents, enabling RAG across the entire web. Our API-first platform makes it simple to integrate live, structured web data into your AI stack, empowering agents and applications with continuously updated context, higher accuracy, and richer reasoning capabilities than ever before.

👥 Julien Khlaut, Machine Learning Researcher & PhD student @Raidium ➡️ Training LLM to pass the residency exam?" Summary: In the expanding field of LLMs, medical knowledge remains a significant challenge due to the specialized nature of the domain.

At Raidium, we are integrating an LLM into our AI-native PACS viewer, which is required to answer questions on all anatomy and for a large variety of pathologies while complying with privacy rules. We developed an open-source model achieving human-level performance on the French ECN (residency exam) by training LLMs on medical textbook knowledge and AI-generated questions. This allows us to self-host and keep the patient's data private.

📍8 rue Cambacérès, 75008 Paris 📆 October 9th, 7:00 p.m. ⚠️ Limited spots available. Be sure to reserve your place in advance!

👥 Alexandre Défossez - Chief Exploration Officer @ Kyutai ➡️ Moshi: a speech-text foundation model for real-time dialogue. Summary: We will discuss Moshi, our recently released model. Moshi is capable of full-duplex dialogue, e.g. it can both speak and listen at any time, offering the most natural speech interaction to date. Besides, Moshi is also multimodal, in particular it is able to leverage its inner text monologue to improve the quality of its generation. We will cover the design choices behind Moshi in particular the efficient joint sequence modeling permitted by RQ-Transformer, and the use of large scale synthetic instruct data.

👥 Louis Lacombe, Valentin Laurent, Thibault Cordier - Data Scientist @ Quantmetry - Part of Capgemini Invent ➡️ Enhancing NLP Model Reliability with MAPIE: Conformal Prediction for Uncertainty Quantification Summary: This talk introduces MAPIE, an open-source Python library designed to quantify uncertainties and control risks in machine learning models, with a focus on NLP applications. We will begin by discussing the importance of uncertainty quantification based on conformal prediction framework that ensures guarantees with few assumptions. Then, we will present MAPIE, showcasing how to compute conformal prediction sets for NLP tasks like text classification. Finally, we will explore practical use cases, highlighting the capabilities of MAPIE and providing attendees with a comprehensive overview of its potential applications.