IA générative, LLMs, recherche en machine learning et impacts sociétaux.
1455 articles
S’abonner (RSS)Optimizing GPU kernels is critical for efficient modern machine learning systems yet remains challenging due to the complex interplay of design factors and rapid hardware evolution. Existing automated approaches typically treat Large Language Models (LLMs) merely as stochastic code generators within heuristic-guided evolutionary loops. These methods often struggle with complex kernels requiring coordinated, multi-step structural transformations, as they lack explicit planning capabilities and frequently discard promising strategies due to inefficient or incorrect intermediate implementations. To address this, we propose Search via Co-Evolving World Model and build K-Search based on this method. By replacing static search heuristics with a co-evolving world model, our framework leverages LLMs' prior domain knowledge to guide the search, actively exploring the optimization space. This approach explicitly decouples high-level algorithmic planning from low-level program instantiation, enabling the system to navigate non-monotonic optimization paths while remaining resilient to temporary implementation defects. We evaluate K-Search on diverse, complex kernels from FlashInfer, including GQA, MLA, and MoE kernels. Our results show that K-Search significantly outperforms state-of-the-art evolutionary search methods, achieving an average 2.10x improvement and up to a 14.3x gain on complex MoE kernels. On the GPUMode TriMul task, K-Search achieves state-of-the-art performance on H100, reaching 1030us and surpassing both prior evolution and human-designed solutions.
Data Agent Kit brings data engineering and science skills, tools, and plugins integrate into VS Code, Claude Code, Codex, and the Antigravity CLI.
Administer Gemini features on your Looker (Google Cloud core) instance
Un sitemap fournit des informations qui aident Google à explorer plus intelligemment votre site. Découvrez comment un sitemap fonctionne et déterminez si vous en avez besoin.
Agentic AI deployment is rarely a simple matter of selecting a foundation model and connecting it to a user interface. Enterprise agents operate inside business processes that already contain policies, data quality issues, exception paths, audit requirements, identity controls and legacy systems. A sales operations agent, for example, may need to read CRM records, interpret account notes, generate a renewal recommendation, check discount eligibility, route an approval and update the opportunity.
Businesses are moving beyond traditional AI agents toward something more transformative: agentic AI. Find out why.
The IT industry stands on the brink of one of its most transformative eras, triggering major consolidations in many long-standing sectors and the emergence of major new markets. IDC’s latest research shows that the infusion of autonomy, adaptability, and decision-making into products and services with agentic AI is redefining the very foundation of technology design, […]
Discover Forrester’s AEGIS, a six-domain framework designed to help CISOs secure, govern, and manage agentic AI systems.
This report features detailed assessments, including profile overviews, key solutions, investments, and case studies of 27 providers that focus on AI and
“Never stop asking questions and seeking answers. Curiosity fuels progress”
Initially created for local containers, Lima has since grown into a universal sandboxing tool for developers. In this talk, I will introduce Lima! What would you do if an AI coding assistant hallucinated and attempted to delete your local workspace? As AI agents become integrated into our daily workflows, running them locally on your host machine is becoming a security risk. You'll also learn about Lima's new support for GPU acceleration, MCP servers, and protective sync modes, and how you can safely run your AI agents entirely in a VM or connect to the VMs from the host. Come see how Lima is helping to make local container development and AI experimentation fast, easy, and safe!
Virtual Event - Kubeflow Community Showcase 2026 is a half-day virtual event highlighting real-world use cases and innovations. Learn how the community applies Kubeflow to Generative AI, MLOps, and LLMOps across cloud, hybrid, and edge environments through demos, practical solutions, and success stories.
Summary and key findings1Password surveyed 500 American IT and security professionals and 500 developers to learn how organizations are adopting and governing AI for their most technical workers. We found that companies are racing to embed AI agents throughout their systems, but have done so with...
Microsoft a présenté le 27 juillet 2026 une architecture de sécurité entièrement bâtie sur l'IA agentique, avec une préversion publique annoncée pour le 3 août 2026. La cybersécurité est devenue l'un des terrains de bataille les plus visibles de la course à l'IA, et chaque acteur y va désormais de son programme
CORTEX