Firestorm Labs, a defense technology startup, has raised $82 million in funding to deploy mobile drone factories inside shipping containers, enabling manufacturing capabilities to be brought directly to front-line operations. This represents a significant advancement in defense technology and autonomous systems deployment.
阅读原文 →
Shapes is a new application that enables human users and AI characters to participate in shared group chat environments. Drawing inspiration from Discord-style community platforms, the app facilitates multi-party AI-augmented conversations, creating new paradigms for human-AI collaboration in social and productivity contexts.
阅读原文 →
Scout AI, founded by Colby Adcock, has secured $100 million in funding to develop AI models specifically for military and defense applications. The company operates a training bootcamp where it is building AI agents capable of helping individual soldiers control fleets of autonomous vehicles, representing significant investment in defense AI capabilities.
阅读原文 →
Chinese regulators have suspended issuance of new autonomous vehicle permits following incidents involving Baidu's robotaxi operations that caused traffic disruptions. The restrictions prevent companies from adding new driverless cars to fleets, expanding into new cities, or initiating new test projects. Authorities are reviewing the sector to prevent similar episodes.
阅读原文 →
GitHub's security team patched a critical remote code execution vulnerability within just six hours of discovery. The bug, uncovered by Wiz Research using AI models, affected GitHub's internal git infrastructure and could have allowed attackers to access millions of public and private code repositories. The rapid response demonstrates AI's growing role in cybersecurity vulnerability discovery.
阅读原文 →
General Motors is rolling out Google's Gemini AI assistant to approximately four million vehicles across the United States. Model year 2022 and newer Cadillac, Chevrolet, Buick, and GMC vehicles with Google built-in will receive the AI upgrade via over-the-air updates. This represents one of the largest deployments of Gemini in the automotive industry.
阅读原文 →
vLLM v0.20.0 introduces significant enhancements including DeepSeek V4 support, CUDA 13.0 as default, PyTorch 2.11 upgrade, Python 3.14 support, HuggingFace Transformers v5 compatibility, and new large model support including Hunyuan v3 preview and Granite 4.1 Vision. Additional improvements include FlashAttention 4 as default MLA prefill backend, TurboQuant 2-bit KV cache compression, online quantization frontend, vLLM IR skeleton, and extensive performance optimizations.
阅读原文 →
The v0.20.1-rc0 release adds system_fingerprint field to OpenAI-compatible API responses, improving compatibility and conformance with OpenAI API specifications for LLM inference endpoints.
阅读原文 →
LangChain Tests version 1.1.7 introduces new content-block-centric streaming (v2) capabilities, bumps pytest to version 9.0.3, upgrades langsmith to 0.7.31, and updates minimum core versions for improved compatibility across the LangChain ecosystem.
阅读原文 →
Version 1.4.2 of LangChain's Anthropic integration restores cache_control functionality on non-direct subclasses, standardizes integration test invocation, and includes the new content-block-centric streaming feature introduced in langchain-core.
阅读原文 →
Ollama v0.22.0 introduces support for two significant new models: NVIDIA's Nemotron 3 Omni and Poolside's first open-weight coding model, Laguna XS.2. The release expands Ollama's model library with both general-purpose and coding-specialized AI models.
阅读原文 →
The v0.22.1-rc0 release features MLX sampler batching across multiple sequences, fixed multi-regex BPE offset handling in tokenizer, NVIDIA TensorRT Model Optimizer import support, improved desktop app startup handling to preserve active sessions, model support for batching, and new model additions.
阅读原文 →
This release enhances API compatibility by accepting "max" as a valid think parameter value and maps OpenAI responses reasoning effort to the think parameter, improving interoperability with OpenAI-compatible clients.
阅读原文 →
Version 0.21.2 enhances the reliability of the OpenClaw onboarding flow in `ollama launch`, fixes recommended models ordering, and now bundles Ollama's web search plugin directly in OpenClaw integration for improved search capabilities.
阅读原文 →
Ollama v0.21.1 adds support for installing and running the Kimi CLI through Ollama, enabling long horizon agentic execution tasks via multi-agent system capabilities. Additional improvements include MLX logprobs support, faster MLX sampling with fused top-P/top-K, GLM4 MoE Lite performance enhancements, and bug fixes for macOS app model picker and Gemma 4 structured outputs.
阅读原文 →