Applied Materials boosted its stake in fellow chip-equipment maker BE Semiconductor and vowed to step up their work together, citing the European company’s growing role in energy-efficient AI computing

NVIDIA INTRODUCES DSX POD AI FACTORIES TO ACCELERATE EUROPE’S INFRASTRUCTURE DEPLOYMENT; DSX POD UNITS REDUCE DEPLOYMENT TIME TO AS LITTLE AS 14 DAYS; EUROPEAN ECOSYSTEM BUILDS WORLD’S MOST EFFICIENT AI FACTORIES WITH NVIDIA DSX

On Day 3 of the Defence Industry delegation visit to Tokyo, Japan, ATLA and the Embassy of India facilitated visit to EdgeCortix, a Japan-based semiconductor company specialising in hardware-software Co-design for energy-efficient AI edge computing. The delegation met with the leadership and technical team at EdgeCortix and engaged in discussions on the company’s cutting-edge AI and semiconductor design technologies, followed by live demonstrations of its solutions. EdgeCortix’s SAKURA-II AI accelerator and MERA software framework support advanced AI workloads, including Generative AI, vision and multimodal applications, with use cases across defence, aerospace, robotics and other demanding environments. The visit provided the SIDM delegation with valuable insights into Japan’s growing semiconductor design and AI ecosystem, and explored potential avenues for technology collaboration and industry partnerships between Indian and Japanese companies. SIDM remains committed to strengthening India–Japan defence and technology industry partnerships and fostering greater collaboration in emerging technologies.

Still calling an LLM for every single memory operation your agent makes? This paper splits that work between a "fast brain" and a "slow brain" instead. Title: Jev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agents URL: Inspired by dual-process cognition (System One vs. System Two), Jev-Mem handles most memory operations with lightweight structured decisions and reserves the LLM for genuinely complex reasoning. Three highlights stand out. 🧠 Typed System-One control Frequent operations like typing, relation judgment, and query routing run through a lightweight interface that returns probabilities and labels instead of free-form text — the same control layer governs both memory construction and retrieval. 🕸️ A four-relation memory graph Memory is organized across semantic, temporal, causal, and entity relations, with retrieval budget allocated to whichever views matter most for a given query — avoiding wasted traversal. 📊 Accuracy and speed improve together On the LoCoMo benchmark, Jev-Mem scores 0.777 overall, an 11% gain over the best baseline, while building memory 6.6x faster (158 seconds) and answering queries 36.7% faster (0.93 seconds). What stands out to me is that separating control from reasoning improved accuracy and efficiency at the same time, not one at the expense of the other. #AIAgents# #MemoryArchitecture#

🏋️ Train smarter. From distributed training to optimization techniques and foundation models, the Training Track at #PyTorchCon# North America, Oct 20-21 in San Jose, is packed with practical sessions to help you build faster, more efficient AI systems. Learn more: 📅 Schedule: 🎟️ Register:

Exclusive: Netherlands-based AI chip startup @AxeleraAI has more than quadrupled its partner roster to over 60 partners since launching its global partner program in June of last year to generate demand for its energy-efficient AI accelerator chips.

📢 𝗝𝗨𝗦𝗧 𝗜𝗡: Multiverse Computing Launches Comet 60B in collaboration with $NVDA NVIDIA 👉 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: ➤ 𝗠𝘂𝗹𝘁𝗶𝘃𝗲𝗿𝘀𝗲 𝗖𝗼𝗺𝗽𝘂𝘁𝗶𝗻𝗴 launches 𝗖𝗼𝗺𝗲𝘁 𝟲𝟬𝗕, based on NVIDIA Nemotron 3.5 Super VL. ➤ Model size reduced 𝟱𝟬%, from 𝟭𝟮𝟬 𝗯𝗶𝗹𝗹𝗶𝗼𝗻 to 𝟲𝟬 𝗯𝗶𝗹𝗹𝗶𝗼𝗻 parameters. ➤ Retains 𝟭𝟮 𝗯𝗶𝗹𝗹𝗶𝗼𝗻 active parameters and multimodal capabilities. ➤ Accuracy remains within 𝟱% of NVIDIA's original model across evaluated tasks. ➤ Comet 60B outperforms the original in selected 𝗔𝗜 𝗯𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸𝘀. ➤ Supports 𝘁𝗲𝘅𝘁, 𝗶𝗺𝗮𝗴𝗲, 𝘃𝗶𝗱𝗲𝗼, and 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜 workloads. ➤ Built using 𝗖𝗼𝗺𝗽𝗮𝗰𝘁𝗶𝗳𝗔𝗜 compression and reinforcement learning techniques. ➤ Available as an 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲 model in 𝗕𝗙𝟭𝟲, 𝗙𝗣𝟴, and 𝗡𝗩𝗙𝗣𝟰 formats. ➤ Reduced model size enables lower 𝗚𝗣𝗨 𝗺𝗲𝗺𝗼𝗿𝘆 requirements and deployment costs. ➤ Multiverse previously launched 𝗣𝘂𝗹𝘀𝗮𝗿 𝟭𝟲𝗕 using NVIDIA Nemotron 3 Nano. 👉 𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗠𝗮𝘁𝘁𝗲𝗿𝘀: ➤ 𝗟𝗼𝘄𝗲𝗿 𝗔𝗜 𝗖𝗼𝘀𝘁𝘀: Smaller models could reduce GPU infrastructure and inference expenses. ➤ 𝗚𝗿𝗲𝗮𝘁𝗲𝗿 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆: Preserves near-original performance while requiring substantially less model storage. ➤ 𝗘𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗔𝗱𝗼𝗽𝘁𝗶𝗼𝗻: Makes advanced multimodal AI more accessible to smaller organizations. ➤ 𝗔𝗜 𝗜𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲: Could reduce memory requirements per deployment, improving hardware utilization. ➤ 𝗠𝗮𝗿𝗸𝗲𝘁 𝗜𝗺𝗽𝗮𝗰𝘁: Highlights growing competition around efficient AI models rather than parameter counts alone. 👉 𝗘𝘅𝗽𝗲𝗿𝘁 𝗦𝘁𝗮𝘁𝗲𝗺𝗲𝗻𝘁: 𝗘𝗻𝗿𝗶𝗾𝘂𝗲 𝗟𝗶𝘇𝗮𝘀𝗼, Co-founder and CEO of Multiverse Computing: “Agents need models that can read documents, understand images and follow video across long tasks, and they need to run where the data lives,” said Enrique Lizaso, co-founder and CEO of Multiverse Computing. “Nemotron 3.5 Super VL is an excellent multimodal model. With Comet 60B, we have cut its size in half while keeping accuracy within 5% of the original, so companies can run it on less hardware and at lower cost. That is exactly what our compression technology is built for.”

📢 Android App: Google Login is Live! We’ve added Google login support to make accessing your account even more convenient and secure, alongside our existing login options. Simply tap “Sign in with Google” to log in quickly with just one tap, with no extra information required. Seamlessly switch between devices and enjoy a smoother, more efficient AI experience wherever you go. 📲 Scan the QR code below to download or update the App and try the new login experience!