Engineering
Puffin-World: Scaling a Unified Multimodal Model With Native 3D World States
Puffin-World is a unified multimodal model that represents 3D scenes through three complementary native states — physics, geometry, and appearance…
Engineering
Making Open-Source AI Weather Forecasting Models Easy to Run
Earthmover and Hugging Face collaborated to simplify running open-source AI weather forecasting models by streaming large meteorological datasets rather than…
Engineering
PyTorch 2.14 Released With CuTeDSL Kernels, New Distributed Backend, and Apple Silicon Linear Algebra
PyTorch 2.14 ships NVGEMM, bringing CuTeDSL-generated CUTLASS kernels to Inductor with epilogue fusion and scaled/NVFP4 GEMM support autotuned alongside Triton…
Engineering
Qdrant and Vultr Release a 10-Billion-Vector Benchmark for Search at Scale
Qdrant and Vultr built Qdrant-FineWeb-10B, an open-source vector search benchmark of 10 billion embeddings derived from Hugging Face’s FineWeb corpus…
Engineering
Don’t Drop Dropout: Optimizing Layer Sparsity for Efficient LLM Training and Inference
Researchers demonstrate that layer dropout, largely abandoned in modern LLM pretraining recipes, actually improves training efficiency and enables valuable post-training…
Engineering
MaxKernel: Agentic Kernel Generation for TPUs
Researchers introduce MaxKernel, a multi-agent system for generating high-performance custom kernels for TPU accelerators using large language models paired with…
Engineering
τ^τ-Bench: An Environment for End-To-End, Realistic Agent Construction
Researchers introduce tau-squared-Bench, a benchmark that evaluates whether AI coding agents can build complete, production-ready customer-service agents under realistic client-engagement…
Engineering
Iris: Climbing to the Search Frontier
Researchers present Iris-mini and Iris-pro, two search agents trained at 35B and 397B parameter scales, along with the full data…
Engineering
Dr. Claw: An AI Scientist Workspace for Vibe Research
Researchers introduce Dr. Claw, an open-source workspace that wraps existing command-line coding agents, such as Claude Code and Gemini CLI,…
Engineering
Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM Systems
Researchers propose a game-theoretic framework for understanding and improving multi-agent LLM systems in which an orchestrator decomposes tasks for a…
Engineering
When Quantization Breaks Memory: Recurrent-State Write-Back in Low-Precision Temporal Inference
A new study isolates how quantization interacts with recurrent neural network memory, introducing the concept of recurrent-state write-back to describe…
Engineering
Group Adaptive Clipping Policy Optimization
Researchers introduce Group Adaptive Clipping Policy Optimization (GAPO), a modification to Group Relative Policy Optimization methods used in reinforcement learning…