As Physical AI advances, ensuring safety and governance for robots and industrial equipment is emerging as a key challenge.
In the AutoBe benchmark, a structured harness narrowed the gap between frontier and local models.
llama.cpp has added support for MTP (Multi Token Prediction) as a beta feature.
LocalVQE, an ultra-compact audio model that removes real-time echo and noise using about 1 million parameters, has been released.
Pilot Shell, an open-source tool that extends Claude Code's capabilities and cuts token costs by up to 90%, has launched.
It's a tool that evaluates whether B2B APIs are suitable for use by AI agents and provides improvement prompts.
A Hangzhou court in China ruled that adopting AI alone cannot justify firing a worker.
AMD's next-generation APU, the Ryzen AI Max+ 495, supports up to 192GB of large-capacity unified memory.
Memtrace, a structural memory tool that helps AI coding agents track codebase changes in real time, has been released.
Codex CLI 0.128.0 adds `/goal`, a goal-based automatic iteration execution feature.
An Anthropic paper used Claude usage data to show the gap between AI capability and actual task coverage.
Analyzes memory and compute scaling in LLMs along with batching strategies for efficient inference.
A tool that replaces Claude Code CLI's backend with DeepSeek and others to cut agent costs by up to 17x.
An analysis of the architecture and key challenges required when AI agents run autonomously for days beyond a single session.
claudely, an open-source tool that lets you connect local LLMs while keeping Claude Code's ecosystem, has been released.
AI fluency has become a basic requirement in interviews, but AI-generated applications are complicating and delaying the hiring process.
OpenAI has opened up ChatGPT subscriptions to OpenClaw, taking the exact opposite strategy from Anthropic.
OpenAI o1 outperformed doctors in emergency room note-based diagnosis and treatment planning.
Musinsa integrated four retail tech systems into Megastore Seongsu, keeping them hidden behind the experience.
Moonshot AI's Kimi K2.6 won an AI coding challenge, edging out GPT-5.5 and Claude.
Running Mistral Medium 3.5 128B Q3_K_M locally on 3x3090 (72GB VRAM) measured 9.67t/s.
A Strix Halo 128GB system ran Mistral Medium 128B at 1.57 tokens per second.
PORTool improved both accuracy and tool-call efficiency together through step-wise importance estimation.
In mixed traffic, class-aware vLLM routing showed the best balance.