Our publications

• Behavioral Machine Learning generalization LLM neurips persona prompt stanford “Behavioral Machine Learning? Behind this question lies a whole stream of research aimed at studying the behavior of AI models when we ask them to imitate, closely or remotely, a human being. However, this approach is fraught with pitfalls, and…
• GraphRAG is the new black documents GraphRAG graphs knowledge RAG You are about to test the RAG to “ask questions of a document database”. Or rather: you’ve just tried these solutions and are a little disappointed? Here’s an opportunity to look back at the fundamental flaws in these approaches, and…
• Causality and AI? Between reasoning fantasies and scientific reality Causality Concept Extraction Graph LLM Diving into causality in AI means navigating between the dream of a reasoning machine capable of identifying cause and effect, and scientific reality, which highlights the limits of our statistical models. This dossier traces the rise of “causal representation learning”,…
• Freeform pixels EdgeAI feature visualization Imitation Learning interpretability physical modeling The subject of edge AI is still in the wilderness, with real difficulty in deploying neural networks correctly. The arrival of a self-sufficient, anonymity-preserving camera is an event not to be underestimated, especially as the approaching freeform Pixel is particularly…
• The RAG under control, the optimized black box, and the 3D modeling of the future black box evaluation gaussian splatting NeRF optimization RAG If you use a RAG and are desperately wondering how to evaluate the quality of its answers in an objective way, you’ll find here a recent and promising work to already measure things better. If you want to find solutions…
• Platonic" representations, very high-resolution images, world models & Mambas distribution Embeddings high resolution mamba system status A better understanding of what neural networks learn is fundamental to our field of work, and here we have a relevant (if somewhat ambitious) publication showing that these representations are similar across architectures and modalities. Beyond this, new work is…
• IAGen as a “universal” simulator, new architectures and better understanding distribution models GAHB Kolmogorov-Arnold UniSim Let’s be honest and warn the reader: two-thirds of this article is more technical than usual. That said, in this article we discuss a fundamental new architecture that could tomorrow revolutionize many Deep Learning approaches, as we move towards a…
• Tame your LLM architectures hallucinations limits LLM robustness safety LLMs are gaining ground everywhere, with incredible promises of new high-performance and, brace yourself, “intelligent” tools. Research is progressing more slowly than these promises, and regularly gives us a clearer and more precise view of things. Here, we outline the…
• Let's dance the Mamba DinoV2 Efficiency LLM mamba Sequence Mamba announces a new, efficient and versatile architecture family that is making its mark on the artificial intelligence landscape. Bonus: a better understanding of image embeddings from DinoV2, and a new way to bypass Large Language Models.
• Imitation learning: AI in robotics becomes credible and accessible ALOHA Deep Reinforcement Learning Diffusion Policies Imitation Learning Robotics If you work in robotic control, you have no right to ignore the ongoing revolution in imitation. And beyond robotic control, any optimization problem modeling an agent that has to make decisions can be inspired by these approaches.
• AlphaGeometry: what lessons can we learn from DeepMind's latest exploit? Deepmind Exploring the solution space IA Symbolic LLM synthetic data Deepmind has caused quite a stir with this AI that can solve complex problems in geometry. This approach offers us a number of theoretical and practical lessons for addressing other problems with Deep Learning: how to combat hallucinations, the value…