Knowledge catalogue
Samantha Subin / CNBC: Qualcomm closed down 11.46% on Tuesday as chip stocks pull back from record AI-driven rally; Intel closed down 6.82%, Sandisk dropped 6%, and Micron 3.61% — Chip stocks dropped
arXiv:2605.10500v1 Announce Type: new Abstract: Agent skills today are static artifact: authored once -- by human curation or one-shot generation from parametric knowledge -- and then consumed unchang
Semafor: Sources: Jensen Huang was left out of President Trump's China trip to avoid unwanted scrutiny and awkward conversations about the sale of Nvidia chips to China — THE SCOOP — The Trump adminis
arXiv:2605.06356v2 Announce Type: replace Abstract: High-resolution image-to-video (I2V) generation aims to synthesize realistic temporal dynamics while preserving fine-grained appearance details of t
The benchmarks show the gap. NVLS all-reduce latency drops from 586.1µs on H200 to 313.3µs on GB200. In MoE prefill at EP=4, combine falls from 730.1µs to 438.5µs. For decode, GB200 sustains much high
The EDA Primer covers the semiconductor design and manufacturing workflow, explaining how Electronic Design Automation tools transform Register Transfer Level (RTL) code into physical silicon through
arXiv:2603.29002v2 Announce Type: replace-cross Abstract: Modern large language models (LLMs) increasingly depends on efficient long-context processing and generation mechanisms, including sparse atte
We published new research on how we serve post-trained Qwen3 235B models on NVIDIA GB200 NVL72 Blackwell racks. GB200 is a major step up over Hopper for high-throughput inference on large MoE models,
arXiv:2605.07719v1 Announce Type: cross Abstract: Long-context inference increasingly operates over CPU-resident KV caches, either because decoding-time KV states exceed GPU memory capacity or because
arXiv:2605.07782v1 Announce Type: new Abstract: LLMs can generate hardware descriptions from natural language specifications, but the resulting Verilog often contains width mismatches, combinational l
arXiv:2605.07194v1 Announce Type: cross Abstract: Dataset distillation compresses a large training set into a small synthetic set that preserves downstream training utility. While most existing method
arXiv:2601.15127v3 Announce Type: replace-cross Abstract: Deploying federated learning across heterogeneous IoT device fleets requires tailored neural network architectures for each device class, yet
arXiv:2605.07662v1 Announce Type: new Abstract: Low-precision number formats are widely used in modern machine learning systems due to their efficiency. Accurate direction representation is key to the
arXiv:2605.07222v1 Announce Type: new Abstract: How few parameters do we really need to forecast a periodic time series? An hourly electricity series, reshaped as a 24-row matrix with one column per d
arXiv:2605.07985v1 Announce Type: cross Abstract: Selecting the optimal LLM inference configuration requires evaluation across hardware, serving engines, attention backends, and model architectures, s
This post likely discusses lesser-known Taiwanese companies in the kitchen or consumer appliance sector that play significant roles in powering AI infrastructure, drawing a parallel to how established
arXiv:2505.13741v2 Announce Type: replace Abstract: Direct training of Spiking Neural Networks (SNNs) on neuromorphic hardware can greatly reduce energy costs compared to GPU-based training. However,
arXiv:2510.07625v2 Announce Type: replace Abstract: While Model Predictive Control (MPC) delivers strong performance across robotics applications, solving the underlying (batches of) nonlinear traject
arXiv:2605.05615v2 Announce Type: replace Abstract: Large language models (LLMs) impose rapidly growing energy demands, creating an emerging energy and carbon crisis driven by large-scale inference. S
arXiv:2511.02805v2 Announce Type: replace-cross Abstract: LLM-based search agents often concatenate the full interaction history into the context, producing long and noisy inputs, and increasing compu
CNBC: Nvidia embraces its role as an AI investor in 2026, committing 40B+ to equity investments including a 30B stake in OpenAI, 3.2B in Corning, and 2.1B in IREN — Nvidia stepped on the gas last year
arXiv:2605.06929v1 Announce Type: cross Abstract: Designing photonic integrated circuits requires accurate electromagnetic field simulations, which remain computationally expensive even for simple dev
arXiv:2605.06675v1 Announce Type: cross Abstract: Large language models cache all previously computed key-value (KV) pairs during generation, and this KV cache grows linearly with sequence length, mak
arXiv:2605.06298v2 Announce Type: replace-cross Abstract: Training world models on vast quantities of unlabelled videos is a critical step toward fully autonomous intelligence. However, the prevailing
arXiv:2605.07897v1 Announce Type: cross Abstract: Online streaming video understanding requires models to process continuous visual inputs and respond to user queries in real time, where the unbounded
arXiv:2603.02883v3 Announce Type: replace Abstract: Diffusion Transformers (DiTs) achieve state-of-the-art video generation quality, but their substantial memory and computational footprints hinder ed
🧵 Slime: The Most Elegant & Comfortable RL Training Framework Ever A deep dive into why Slime redefines LLM RL training with clean architecture & production-grade engineering ✨ Insights from Zhihu con
arXiv:2602.06283v2 Announce Type: replace Abstract: Exploiting sparsity during long-context inference is key to scaling large language models, as attention dominates the cost of autoregressive decodin
arXiv:2605.06763v1 Announce Type: new Abstract: Sparse attention improves LLM inference efficiency by selecting a subset of key-value entries, but at the cost of potential accuracy degradation. In par
arXiv:2603.23198v2 Announce Type: replace-cross Abstract: Scaling autoregressive large language models (LLMs) has driven unprecedented progress but comes with vast computational costs. In this work, w
arXiv:2509.05276v4 Announce Type: replace-cross Abstract: Mainstream Transformer-based large language models face major efficiency bottlenecks: training computation scales quadratically with sequence
This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. A few months before he was awarded the Nobel Prize in economic
arXiv:2605.07655v1 Announce Type: cross Abstract: Searching a multi-biometric database of a billion records for a country-level identity system requires pushing the limits of all aspects of a biometri
arXiv:2508.02001v2 Announce Type: replace-cross Abstract: Pervasive encryption makes large-scale labeling infeasible for traffic analysis, while security operations demand edge analysis to avert servi
arXiv:2605.06927v1 Announce Type: cross Abstract: Object detection on heterogeneous edge devices must satisfy strict energy, latency, and memory constraints while still providing reliable perception f
arXiv:2605.07378v1 Announce Type: new Abstract: Zero-shot proxies, also known as training-free metrics, are widely adopted to reduce the computational overhead in neural network evaluation for scenari
This article likely examines the relationship between Arm Holdings (the British semiconductor design company), the UK government, and Apple, possibly covering topics such as Apple's use of Arm-based c
I was talking to a room of senior accountants a couple months ago and 10% had OpenClaw installations. Of course there are far more non-users and firms lag behind their people, but there is a sort of S
The market is trying to price a transition it hasn’t fully internalized. It sees Nvidia Corp.’s market cap with a five-handle and assumes the valuation is too high to grow further. We believe that’s t
“You are entering the world at an extraordinary moment,” NVIDIA founder and CEO Jensen Huang told graduates as he delivered the keynote address at Carnegie Mellon University’s 128th commencement cerem
Yann LeCun closed 1.03B for AMI Labs on March 10. Three days later, this paper dropped from his NYU collaborators. 15M parameters. Single GPU. A few hours of training. LeWorldModel is the first JEPA t
Neocloud company IREN Ltd. has secured a 2.1 billion commitment from the chipmaker Nvidia Corp. as part of a new data center partnership aimed at artificial intelligence workloads. The partners plan t
Learn how to deploy any Hugging Face model in one session using Goose and Together's Dedicated Container Inference. Skip the setup complexity — one prompt gets your model running in a production-grade
Great collab with @SakanaAILabs on an #ICML26 paper about sparse transformer kernels + formats optimized for modern NVIDIA GPU execution. • TwELL sparse packing • Fused CUDA kernels • 20%+ inference/t
Grammar-constrained decoding modifies language model generation by applying grammar constraints at each step to block structurally invalid tokens , ensuring syntactically correct Bash command generati
OPINION: Ever since private equity bought the @Veritasium YouTube channel, the content just hasn't been as good. Used to be a GREAT channel, maybe the best Science channel on YouTube, and now? SAD! Ve
Tickets are now available for the 4th annual AI Film Festival: June 11 at Alice Tully Hall at Lincoln Center in NYC, and June 18 at The Broad Stage in LA. It’s the biggest and most important celebrati
We’ve rolled out a significant update to Google Kubernetes Engine (GKE) that solves one of the most annoying problems in cloud infrastructure: cold start latency. GKE now has up to 4x faster node star
arXiv:2605.04074v1 Announce Type: new Abstract: AI data centers experience rapid fluctuations in power demand due to the heterogeneity of computational tasks that they have to support. For example, th
arXiv:2605.04595v1 Announce Type: new Abstract: The rapid adoption of large language models (LLMs) has created significant challenges for efficient inference at scale. Unlike traditional workloads, LL
This article discusses optimization techniques for maximizing efficiency on NVIDIA's GB200 NVL72 system using Slurm block scheduling, a job scheduling approach designed to improve resource utilization
arXiv:2605.04711v1 Announce Type: cross Abstract: Optimizer states occupy massive GPU memory in large-scale model training. However, gradients in different network blocks exhibit distinct behaviors, s
arXiv:2605.04357v1 Announce Type: cross Abstract: The usage of large language models (LLMs) has grown increasingly fragmented, with no single model dominating. Meanwhile, cloud providers offer a wide
arXiv:2605.05023v1 Announce Type: new Abstract: Efficient CUDA implementations of attention mechanisms are critical to modern deep learning systems, yet supporting diverse and evolving attention varia
arXiv:2605.04198v1 Announce Type: new Abstract: Performance of deep learning models is strongly governed by architectural capacity, with width and depth as primary controls. However, in physical-scien
arXiv:2605.04997v1 Announce Type: new Abstract: DualTCN is the first deep-learning framework for inverting time-domain marine controlled-source electromagnetic (MCSEM) transient data. Moving away from
Hot take on Elon’s surprise decision to rent 30 megawatts of compute to Anthropic: 1. It’s a tacit concession that xAI is not all that close to AGI (despite what he suggested last year). 2. It’s more
.@huggingface's agentic robotics app store for Reachy Mini is a big step toward more accessible physical AI. 🙌 Excited to see NVIDIA Isaac GR00T N integrated with Hugging Face LeRobot, helping develop
This post discusses a request for filtering functionality on Google Maps that would allow users to search for restaurants by ethnic cuisine type or authenticity level, suggesting a desire for better t
Lambda Labs secured a $1 billion senior secured credit facility to fund the expansion of its AI infrastructure and meet growing demand for gigawatt-scale computational resources. This financing enable