Infrastructure & Hardware

B
Showing 91–120 of 210
  • Lobste.rs (AI tagged) · EN Infrastructure & Hardware
    Qwenputer: a fully functioning, deterministic RISC-V CPU inside Qwen3.5
    Qwenputer: a deterministic RISC-V CPU running inside Qwen3.5
    A blog post shared on Lobsters presents Qwenputer, which per its title is a fully functioning, deterministic RISC-V CPU implemented inside the Qwen3.5 language model. The article body was not available, so implementation details and performance are unconfirmed.
    Read original (Lobste.rs (AI tagged)) ↗
  • arXiv cs.CL (Computation and Language) · EN Infrastructure & Hardware
    Integrating Heterogeneous Echocardiography AI Outputs in Language Space: Learnable Late Fusion with a Supervised LLM
    Read original (arXiv cs.CL (Computation and Language)) ↗
  • arXiv cs.LG (Machine Learning) · EN Multimodal
    Smooth Action Chunking Via Movement-Primitive Representations for Imitation Learning
    Computer Vision Inference Retrieval-Augmented Generation (RAG)
    Read original (arXiv cs.LG (Machine Learning)) ↗
  • arXiv cs.LG (Machine Learning) · EN Infrastructure & Hardware
    Softmax Routing Dynamics in Mixture-of-Experts Models: Collapse and Recovery
    Mixture of Experts (MoE) Transformer
    Read original (arXiv cs.LG (Machine Learning)) ↗
  • arXiv cs.LG (Machine Learning) · EN Infrastructure & Hardware
    Accuracy is not compatibility: two Evo 2 checkpoints hinge on one FP8 matrix multiplication
    Read original (arXiv cs.LG (Machine Learning)) ↗
  • arXiv cs.CL (Computation and Language) · EN New Model Releases
    How Specific Do I Need to Be? Effects of User Query Underspecification on Deep Research Agents
    AI Agents Reinforcement Learning Software Engineering
    Read original (arXiv cs.CL (Computation and Language)) ↗
  • arXiv cs.LG (Machine Learning) · EN Infrastructure & Hardware
    Dobb: Shard-Local Preconditioning in Block-Diagonal Eigenbases
    GPT
    Read original (arXiv cs.LG (Machine Learning)) ↗
  • arXiv cs.LG (Machine Learning) · EN Multimodal
    Semantic Mismapping in Generalized Category Discovery: Evaluation and Mitigation
    Deep Learning Retrieval-Augmented Generation (RAG) Reinforcement Learning
    Read original (arXiv cs.LG (Machine Learning)) ↗
  • arXiv cs.CL (Computation and Language) · EN Training & Fine-tuning
    XLaw-JP-27B: A Lightweight Open Model Specialized for Japanese Law
    Software Engineering
    Read original (arXiv cs.CL (Computation and Language)) ↗
  • arXiv cs.CL (Computation and Language) · EN Infrastructure & Hardware
    Reasoning or Pattern Exploitation? Mechanistic Insights from RL-Trained Language Models
    Reinforcement Learning
    Read original (arXiv cs.CL (Computation and Language)) ↗
  • arXiv cs.CL (Computation and Language) · EN Infrastructure & Hardware
    Reading-Level Verification of Kanji Words in Japanese Financial Speech Synthesis
    Deep Learning Neural Network Retrieval-Augmented Generation (RAG) Speech Processing
    Read original (arXiv cs.CL (Computation and Language)) ↗
  • arXiv cs.CL (Computation and Language) · EN New Model Releases
    A Novel Approach to Evaluating Evaluation Metrics for Multi-Output Structured Prediction
    Retrieval-Augmented Generation (RAG)
    Read original (arXiv cs.CL (Computation and Language)) ↗
  • ITmedia AI+ · JA Infrastructure & Hardware
    AI半導体市場「メガサイクル局面だ」　アドバンテスト好決算、CEOが語る「CPU」「光電融合」への期待とワクワク
    Advantest raises FY2026 sales forecast to 1.863 trillion yen on AI demand
    Semiconductor test equipment maker Advantest raised its fiscal 2026 full-year sales forecast to 1.863 trillion yen, citing AI-driven chip demand. The article asks what is underpinning the semiconductor market's entry into a 'mega cycle', and reports the CEO's expectations.
    Read original (ITmedia AI+) ↗
  • arXiv cs.LG (Machine Learning) · EN Training & Fine-tuning
    PALATE: Pretrained Autoregressive LATent Embeddings
    Deep Learning Embeddings GPT
    Read original (arXiv cs.LG (Machine Learning)) ↗
  • arXiv cs.LG (Machine Learning) · EN Infrastructure & Hardware
    PhysicsNeMo-Mesh: GPU-Native Meshes and Simulation Cases for Physics AI
    Machine Learning Reinforcement Learning
    Read original (arXiv cs.LG (Machine Learning)) ↗
  • Cohere Blog · EN Funding & M&A
    Who Gets to Build the Frontier for AI
    Cohere argues the democratic world needs more frontier AI builders
    Neural Network NVIDIA Reinforcement Learning
    Cohere's co-founders argue AI will become part of the operating infrastructure of economies, public services and national defence. They say who builds it, under what law and accountability shapes every adopting country's choices, so more builders are needed.
    Read original (Cohere Blog) ↗
  • arXiv cs.LG (Machine Learning) · EN Infrastructure & Hardware
    Closed Gates Leak: Decomposing Adjacent-Layer Cancellation in Gated Language Models
    Machine Learning Neural Network Retrieval-Augmented Generation (RAG) Transformer
    Read original (arXiv cs.LG (Machine Learning)) ↗
  • arXiv cs.CL (Computation and Language) · EN Infrastructure & Hardware
    Tracing the Influence of Retrieved Literature on Scientific Ideation in LLMs
    Machine Learning Retrieval-Augmented Generation (RAG)
    Read original (arXiv cs.CL (Computation and Language)) ↗
  • arXiv cs.CL (Computation and Language) · EN Multimodal
    Contextual Embeddings Beyond the Gold Chunk
    Computer Vision Embeddings Retrieval-Augmented Generation (RAG) Reinforcement Learning Software Engineering
    Read original (arXiv cs.CL (Computation and Language)) ↗
  • arXiv cs.AI (Artificial Intelligence) · EN Infrastructure & Hardware
    AtomTree: A Hierarchical Framework for State-Aware Embodied Instruction Following with LLMs
    AI Agents Neural Network
    Read original (arXiv cs.AI (Artificial Intelligence)) ↗
  • arXiv cs.AI (Artificial Intelligence) · EN New Model Releases
    From Grind to Grasp
    Reinforcement Learning
    Read original (arXiv cs.AI (Artificial Intelligence)) ↗
  • arXiv cs.AI (Artificial Intelligence) · EN Multimodal
    inkGround: Benchmarking Fine-grained Visual Grounding Capabilities of Vision-Language Models for Cursive Handwriting in the Wild
    Computer Vision Retrieval-Augmented Generation (RAG)
    Read original (arXiv cs.AI (Artificial Intelligence)) ↗
  • Data Center Dynamics · EN Infrastructure & Hardware
    Tata Communications to invest $450m into 10,000km of AI-ready fiber to support data center connectivity in India
    Tata Communications to invest $450m in AI-ready fiber across India
    Neural Network
    Tata Communications plans to spend $400-450 million on about 10,000km of long-haul AI-ready fiber in India. Phase one covers the Mumbai-Pune-Hyderabad-Chennai corridor, with later phases reaching east coast and western hubs and linking to cable landing stations.
    Read original (Data Center Dynamics) ↗
  • Data Center Dynamics · EN Infrastructure & Hardware
    KBRA assigns preliminary ratings to up to $525m in US Signal notes, secured against ten data centers
    KBRA gives preliminary A- rating to US Signal's $525m data center ABS
    US Signal plans to raise up to $525 million via three asset-backed notes secured by ten owned data centers, and KBRA assigned a preliminary A- rating. The pool leans toward older assets in secondary markets, so KBRA applied a stressed 12.0% terminal cap rate.
    Read original (Data Center Dynamics) ↗
  • Data Center Dynamics · EN Infrastructure & Hardware
    Vertiv shares jump 14 percent as data center earnings beat expectations
    Vertiv shares jump 14% as Q3 beat and orders double year over year
    Vertiv reported Q3 2026 net sales of $4.31 billion, up 61%, with adjusted EPS up 78% to $2.21 and orders doubling year on year. It raised its 2026 sales outlook to $14.45-14.85 billion, and the shares rose about 14%.
    Read original (Data Center Dynamics) ↗
  • Data Center Dynamics · EN Infrastructure & Hardware
    Oracle AI World takeaways: making Nvidia envious, and storage... sexy?
    DCD column: Oracle widens its chip and storage choices at AI World
    NVIDIA Retrieval-Augmented Generation (RAG) Reinforcement Learning
    At Oracle AI World, OCI said it will offer Google's TPUs and add chips from Cerebras, SambaNova and Positron, while partnering with Vast Data and WEKA on storage. Executives say wider choice helps in a supply-constrained market and as AI inference grows more specialized.
    Read original (Data Center Dynamics) ↗
  • Data Center Dynamics · EN Infrastructure & Hardware
    UK budget: Grid connection incentives for data centers under new “covenant”
    UK budget: data centers get grid priority under a new covenant
    Neural Network
    Chancellor John Healey announced a British Data Center Covenant requiring operators to invest locally, create jobs and protect water supplies. Compliant critical sites get priority grid connections and permission for on-site generation while they wait.
    Read original (Data Center Dynamics) ↗
  • Data Center Dynamics · EN Infrastructure & Hardware
    Former New Zealand pulp mill eyed for storage-focused data center
    Former NZ pulp mill site eyed for storage-focused data center
    Retrieval-Augmented Generation (RAG)
    Proposals filed with Ruapehu council would turn the former Karioi pulp mill into a wood pellet plant plus a Data Igloo cold-storage data center campus. Details are scarce, though the firm cites a multi-exabyte first campus.
    Read original (Data Center Dynamics) ↗
  • Data Center Dynamics · EN Infrastructure & Hardware
    Blockchain and crypto firm Blockmate Ventures enters neocloud market with purchase of Nvidia B200 cluster
    Blockmate Ventures enters neocloud market with Nvidia B200 purchase
    Neural Network NVIDIA
    Blockmate Ventures will buy eight Nvidia B200 GPUs for $635,000 with Scannit operating and selling the compute. Delivery is due in Q1 2027, and the company hopes to recover its capital in about two years, depending on utilization and pricing.
    Read original (Data Center Dynamics) ↗
  • Data Center Dynamics · EN Infrastructure & Hardware
    Genius Group looks to pivot to data centers, partners with FingerMotion and BlueFlare for project in Canada
    Edtech firm Genius Group pivots to data centers with Canada project
    Software Engineering
    NYSE-listed Genius Group launched GENIUSpark and is partnering with FingerMotion and BlueFlare on a 2.4MW data center in Hanna, Alberta. It holds 49%, with construction planned for December 2026 and first compute in Q2 2027.
    Read original (Data Center Dynamics) ↗