TriAttention Boosts LLM Throughput 2.5x With Advanced KV Cache Compression
Researchers from MIT, NVIDIA, and Zhejiang University introduce TriAttention, a method that optimizes KV cache compression to match full attention performance.

Latest AI model news covering frontier model launches, benchmark shifts, reasoning upgrades, context-window changes, and competition across foundation model labs.
Researchers from MIT, NVIDIA, and Zhejiang University introduce TriAttention, a method that optimizes KV cache compression to match full attention performance.

Anthropic temporarily suspended OpenClaw creator Peter Steinberger following a controversial policy change that excludes third-party tools from subscriptions.

Google's Gemini app now generates interactive 3D models and dynamic charts, allowing users to simulate complex concepts and variables in real time.

Alibaba's Tongyi Lab introduces VimRAG, a multimodal RAG framework using memory graphs to navigate massive visual contexts and improve agent reasoning.

Anthropic is limiting access to its powerful Mythos AI, which can identify software vulnerabilities, to prevent cybercriminals from exploiting the technology.

Google AI Research unveils PaperOrchestra, a multi-agent framework that converts raw experimental logs into submission-ready, LaTeX-formatted research papers.

Meta has launched Muse Spark, its first major AI model under chief AI officer Alexandr Wang, marking a shift toward large-scale generative AI deployment.

AutoAgent is an open-source library that allows AI to autonomously engineer and optimize its own agent harness, achieving top-tier benchmark performance.

Meta AI introduces EUPE, a family of compact vision encoders under 100M parameters that rival specialist models in image understanding and dense prediction.

Netflix and INSAIT release VOID, an open-source AI model that removes objects from video while maintaining physical consistency and scene dynamics.

Follow the AI News RSS feed or sign up for the free weekly newsletter. Each story opens with key takeaways, so you can catch up on a week in a few minutes.
Use the topic filter to narrow the archive, then open a story for its full explanation and source links.