ibm

回顾性文章介绍 IBM 在1985年推出的 Token Ring 网络技术，它曾与以太网竞争，凭借无冲突和高负载下更稳健的优势一度表现优秀，但因成本更高、升级速度落后于以太网，最终在1980-90年代的“网络标准之争”中败下阵来，如今已成为被遗忘的技术古董。

IBM 推出 Granite Switch 4.1（8B 与 30B），通过 aLoRA 适配器在同一端点上组合聊天、决策与语音转写能力，几乎不额外占用 VRAM。新增 Decision 与 Guardian aLoRA，并嵌入 OpenASR 榜单最快的 Granite-Speech-5.0-Turbo-CTC 模型。

PyTorch Foundation 博客介绍了 IBM 如何利用 PyTorch 现有的设备、流、内存管理和编译抽象，将 Spyre 数据流 AI 加速器集成成为原生 PyTorch 设备，从而在无需为每种硬件单独开发框架的情况下支持更广泛的 AI 硬件。

1963 年 IBM 总裁小沃森的内部备忘录（又称"保洁员备忘录"）质问 IBM 如何将"行业领导地位"输给了一个"仅 34 人、包括保洁员"的公司——控制数据公司（CDC），凸显了 CDC 6600 对 IBM 主机霸主地位的冲击。

Red Hat is reportedly being gradually absorbed and dissolved by IBM, with employees leaving, badges replaced by IBM branding, and open-source culture eroding across its subsidiaries like Nordcloud and the quantum teams.

IBM's open-source agent CUGA integrates Tavily for web search, enabling reliable access to live information in air-gapped and sovereign environments, and streamlining enterprise applications.

NASA and IBM have released an open-source AI model focused on geospatial analysis for lunar foundations, aiming to advance research and applications in space exploration.

IBM and NASA have open-sourced a lunar foundation model that integrates multi-sensor data to map the Moon, aiding in tasks like crater mapping, volcanic history investigation, and ice detection.

IBM released the Granite Time Series PatchTST-FM-r2 model, a 385M-parameter foundation model for zero-shot time-series forecasting with top performance on the GIFT-Eval benchmark and a commercial-friendly Apache 2.0 license.

IBM Bob is an AI-powered development partner that helps developers build software faster with natural language coding, analytics for tracking contributions, and seamless integration with enterprise tools like Red Hat and Instana.

IBM released the Quantum Nighthawk r2 processor featuring 120 qubits and independent high-speed qubit reset, enabling over 100,000 circuits per second with up to 25x higher throughput than IBM Quantum Heron. The system is available on the IBM Quantum Platform and advances IBM's roadmap toward fault-tolerant quantum computing.

Former MIT researchers now at IBM discuss their work on AI and quantum applications through the MIT-IBM Computing Research Lab, emphasizing the transition from academic research to real-world deployment.

IBM and Confluent have partnered to offer time series foundation models for real-time intelligence on streaming data, now available in early access on Confluent Cloud.

IBM unveils a next-generation dual-architecture processor for IBM Z and LinuxONE that natively supports both Arm and IBM Z instructions, featuring 2nm technology and AI accelerators to enhance enterprise computing options.

IBM has launched Granite 4.2, a family of open-weight large language models with variants up to 30B parameters, featuring a 128,000-token context window and agentic reinforcement learning for enhanced capabilities.

The Harvest (IBM 7950) was a Cold War supercomputer built by IBM for the NSA, pioneering cryptanalysis with a modular coprocessor and automated tape library for high-speed data processing.

IBM and USTA are introducing AI-powered features like Live Updates, Serve Quality, and Match Chat to the US Open app for the 2026 tournament, leveraging live data streams and AI agents.

The article discusses the gap between described and established governance in AI agents, referencing a paper by Paulo Cavallo, and highlights how companies like Microsoft, IBM, and Lyzr are developing control-plane capabilities to enforce policies at runtime.

IBM launches Docling, an open-source Python library designed to convert documents into structured data.

A new IBM paper introduces BenchDrift, an auditing framework that measures how LLM performance varies with question rephrasing, revealing significant accuracy gaps and highlighting benchmark reliability issues.