We propose RayRoPE, a positional encoding scheme with SE(3) invariance and geometric adaptability for multi-view transformers.
Google is preparing to expand Gemini Live, Gemini's real-time voice mode, and Skills, its user-customization feature, to the web and desktop environment.
Chinese AI startup Z.ai is boosting its market share by open-sourcing its models and is on the verge of hitting $1 billion in annual revenue.
Alibaba has open-sourced SAIL, its software stack for its own AI chips, in an effort to break the dominance of NVIDIA's CUDA ecosystem.
Moonshot AI has released Kimi Code CLI, an AI coding agent that can edit code and execute commands in the terminal.
A new neural network learning mechanism has been proposed that enables training while following Dale's principle, without the Weight Transport problem.
Claude Fable 5 showed superior performance and consistency over GPT-5.6 Sol on an NP-Hard optimization problem.
In connection with its trade secret infringement lawsuit against OpenAI, Apple sent legal letters ordering the preservation of evidence to dozens of former Apple employees who moved to OpenAI.
Alibaba plans to release Qwen3.8, a massive model with 2.4 trillion parameters, as open-weight.
NVIDIA has released Nemotron 3 Embed, a collection of open embedding models that maximizes retrieval performance.
Summarizes 10 recent AI papers on the themes of agent advancement, security reinforcement, and multi-modality verification.
Exposure to AI advice was found to impair users' ability to suspend judgment, lowering accuracy while only boosting confidence.
HuggingFace experienced a security breach incident carried out by an AI agent, and in the course of responding, pointed out guardrail issues with commercial models.
The ICM methodology has been released, managing agent task order and context through folder structures and markdown files.
BeeLlama v0.4.0, a fork of llama.cpp with improved KV cache quantization performance, has been released.
Research on ATSInfer, which greatly improves LLM inference performance on consumer devices through tensor-level scheduling, has been announced.
Decart has launched Lucy 2.5, a real-time AI model that can remove objects and replace backgrounds in video using natural language prompts.
Even with low reconstruction error, downstream task performance can drop, and the optimal compression code varies by consumer.
A study has been published that monitors AI agents' tool-use decisions through Mechanistic Interpretability.
In the AI Safety Index released by the Future of Life Institute, Anthropic recorded the highest score with a C+.
ASCIITermDraw-Bench, which evaluates VLMs' ability to generate and edit text-based diagrams, has been released.
Alibaba has announced the upcoming release of the 2.4T-parameter Qwen3.8 model along with an open-weight release.
iFixAi, an open-source tool that diagnoses operational misalignment and security blind spots in AI agents across 45 checks, has been released.
Ollama, an open model execution platform, has raised a Series A round of $88 million.
AI Briefing