All news

ArchiveLens reconstructs semantic content from historical maps with generative models.

Pre-history memory for humanoid loco-manipulation.

PROSPECT unifies streaming vision-language navigation with semantic-spatial fusion.

Shows much apparent rare-word TTS error is really ASR verifier error.

Learns flow maps that jump forward in time to make stochastic dynamical system predictions cheaper.

Foundation4D unified perception and reconstruction for autonomous vehicles.

Memory-enhanced test-time adaptation for meta-RL.

An agent-assisted workflow for building a digital twin of an AI data center from plans and equipment data.

Position on recursive self-improvement in physical AI.

Generalist dexterous bimanual tracking from human demos.

Explainable artifact detection in AI-generated videos.

Reward models for when LLM assistants should use memory.

Gforth 1.0 is out, the first 1.0 release of the GNU Forth system.

A user found Samsung TVs run an INT8 TensorFlow Lite model on 960x540 screenshots every few seconds to detect what game is being played.

Max-severity unauthenticated RCE in self-managed GitLab spans 12 years of releases.

A tutorial on embedding trained neural networks as constraints inside optimization models.

Pharo 14, the Smalltalk-style language and environment, is released.

CARSS combines online arrival forecasting with co-simulation to adapt job scheduling.

A simple Sudoku solver written in assembly, part of a series across languages.

Workload-aware monitoring for AI-RAN platforms sharing CPU and GPU.

Continual deep RL on dynamic graphs for UAV-assisted multi-user 360 VR video streaming.

More metadata can mislead LLM museum descriptions.

vLLM supports MiniMax-M3.1-Flash on day 0, including its NVFP4 KV cache, NVFP4 MoE weights and a draft model for speculative decoding.

vLLM runs Recursive Flint, a 40B MoE with 2.5B active parameters, on one GPU at about 500 tokens per second with 100K context.

Cut-topology FPGA logic architecture with better area efficiency.

Cost-aware abstention for LLM IT ticket routing.

Atlassian's survey finds developers saving 10+ hours a week with AI fell from 68% to 26% year over year.

Agentic system for 3D virtual film cinematography.

A preregistered study measures wire-level interrupt-to-decision latency of on-sensor machine learning cores versus host inference on a Jetson Orin Nano.

Goose is a memory-safe systems language with no heap allocator, GC or lifetime annotations, claiming benchmarks faster than C++.