Benchmark lève 2 milliards de dollars pour son premier fonds de croissance
Benchmark raises $2 billion for its first growth fund, marking a new chapter for the venture capital firm.
Benchmark raises $2 billion for its first growth fund, marking a new chapter for the venture capital firm.
ChatGPT introduces a memory system to retain user preferences and context across conversations, making the assistant more relevant and helpful.
Study shows 60% of UK media articles on AI fail to disclose authors' defence sector connections. The lack of transparency raises concerns about conflicts of interest in coverage of military AI applications.
GitHub repository with implementations of various Transformer attention mechanisms. Originally developed for Small Language Model experiments and benchmarking, applicable to Computer Vision, Vision Encoders, RL, and other domains. Open to community contributions.
Meta launches Business Agent, an AI agent to automate sales and customer support on its platforms. The tool processes over one billion daily conversations between businesses and consumers.
Gemma 4 12B heretic model tested for code generation: single-prompt retro game creation (45k tokens total). Consistent 18.44-18.93 t/s throughput, 4,372-token code generation in 4 minutes. 91.7-96.4% cache reuse on llama.cpp with Ryzen 9 9950X + RX 6800.
PEEL is a methodological framework combining deterministic analysis (Voyant Tools) and Claude interpretation to assess LLM epistemic accountability in research. Applied to AI-generated summaries, it reveals systematic distortions invisible without non-AI measurement and proposes three implications: instrument AI tools, distinguish fluency from fidelity, design epistemic authority.
Position paper arguing deployed RL systems should adopt continual learning instead of train-then-fix paradigm. Authors identify four sources of post-deployment non-stationarity requiring never-ending learning and analyze real-world continual RL examples.
Neetyabhas presents a multi-agent simulation framework for optimizing public health policies under uncertainty. The model integrates 1,000 individual agents (masking, vaccination, shopping decisions) and policymakers using hierarchical RL agents (DQN, DDPG, TD3). Results show masking and vaccination significantly reduce epidemic peak height and duration.
Headroom compresses tool outputs, logs, files, and RAG chunks before sending to LLM, reducing token consumption by 60-95% without quality loss. Available as library, proxy, or MCP server.
Vercel enables building and deploying Shopify storefronts directly from its platform. Shopify integration available via Vercel Marketplace, v0, or CLI. Free test store creation, deployment without leaving workflow. Connecting existing stores coming soon.
Hugging Face redesigns its CLI to optimize it as an agent. The command-line interface becomes agent-friendly with structured commands and parsable responses, enabling autonomous systems to interact directly with the Hub.
Google offers opt-out toggle in Search Console for AI Overviews and AI Mode, reaching 3.5 billion monthly users. Move prompted by UK Competition and Markets Authority (CMA) over publisher disadvantage concerns.
Google releases Gemma 4 12B, a multimodal model processing text and images on consumer PCs without dedicated servers. The 12B parameter model makes advanced AI locally accessible.
Google patches 124 Android security vulnerabilities, including one already exploited by attackers in the wild. Major security update addressing critical risks.
Pull request on llama.cpp optimizing MTP (Multi-Token Prediction) for Qwen 3.5 by using post-norm hidden state. Performance improvement for multi-token prediction.
Real-time monitoring tool for AI coding agents. Displays Claude Code & Codex CLI sessions, tokens, context window, rate limits, and ports. htop-like interface.
BAML is an engineering framework for structured prompt management. Compatible with Python, TypeScript, Ruby, Java, C#, Rust, Go. Aims to professionalize prompt engineering with robust engineering tools.
Goose is an open-source, extensible AI agent that goes beyond code suggestions. It can install, execute, edit, and test with any LLM.
HexStrike AI MCP Agents is an MCP server enabling AI agents (Claude, GPT, Copilot) to autonomously run 150+ cybersecurity tools for automated pentesting, vulnerability discovery, and security research.
A merged PR in llama.cpp reveals implementation of a new « Gemma 4 Unified » model type. The llama.cpp team reportedly got early access. Code comments mention a « transformer-less vision tower », hinting at a novel Google architecture.
Developer builds Tauri v2 desktop chat shell for local LLMs. Compatible with Ollama, llama.cpp, and OpenAI-compatible endpoints. MIT license, ~12 MB binary.
Perplexity launches an orchestrator that automatically routes AI tasks between local models running on-device and powerful cloud models, optimizing execution based on task requirements.
OpenAI launches GPT-Rosalind with enhanced capabilities in biological reasoning, medicinal chemistry, genomics analysis, and experimental workflow for life sciences research.
Hugging Face explores applying DPO (Direct Preference Optimization) beyond chatbots, including for vision and reasoning model optimization. The article details how this alignment technique can improve performance on complex tasks without requiring an explicit reward model.
TorchDAE is a PyTorch library for solving Differential Algebraic Equations (DAEs) with GPU acceleration. It implements Generalized-Alpha integration, Dummy Derivatives index reduction, and adjoint sensitivity methods for DAEs. Target applications: system identification, scientific machine learning, and physics-informed modeling.
Nous Research releases Hermes Desktop, an open-source AI agent app under MIT license available across all platforms. The application provides a free alternative to proprietary assistants.
Microsoft announces seven in-house AI models at Build 2026, including its first reasoning model. The company also introduces a new tuning method and an autonomous background agent.
OpenAI outlines a federal framework for U.S. governance of frontier AI, addressing safety, resilience, and national security. The blueprint details oversight mechanisms and supervision protocols for frontier models.
Microsoft unveils Scout, an autonomous agent integrated into Microsoft 365 that can organize, coordinate, and execute tasks continuously to automate enterprise work.
Philosophical paper examining whether AI chatbot outputs (e.g., Anthropic's Claude) produce meaningful language. Authors argue that standard human language theory already applies to LLMs without requiring anthropomorphic assumptions about intentions or mental states.
Systematic review of hybrid architectures for wind power interval forecasting. Approaches combining deep learning, modal decomposition (VMD, EEMD), and statistical methods improve accuracy. Dominant strategy: two independent models (LSTM, ELM) for lower/upper bounds. Challenges: lack of standardized metrics, computational complexity, limited real-world validation.
University of Toronto researchers demonstrated that an AI worm could target any online device by exploiting language models. The study shows how LLMs can serve as vectors for cross-device malware propagation.
Hugging Face integrates MCP (Model Context Protocol) tools into Reachy Mini, a humanoid robot. This integration enables the robot to access external tools via MCP protocol, expanding its interaction and autonomy capabilities.
Helsing, a German startup, unveils the RX-1 on June 1, 2026. The European military robot relies on American inference silicon, exposing Europe's technological dependence in defense robotics despite sovereignty ambitions.
Uber capped employee AI spending after exhausting its budget in four months. The company now enforces access limits on AI tools to control costs.
Release of micropython-wasm 0.1a1 with fixes for limitations discovered while building datasette-agent-micropython. Enables Python execution in WebAssembly with sandboxing.
mistral.rs is an optimized LLM inference framework prioritizing speed and flexibility. Open-source project enabling efficient execution of language models.
Google Workspace CLI: unified command-line tool for Drive, Gmail, Calendar, Sheets, Docs, Chat, Admin. Dynamically generated from Google Discovery Service. Includes AI agent capabilities.
Nanoclaw is a lightweight OpenClaw alternative running in containers for security. Integrates WhatsApp, Telegram, Slack, Discord, Gmail and other messaging apps. Includes memory, scheduled jobs, runs on Anthropic's Agents SDK.