The Stack — April 02, 2026

Today's technology news highlights significant developments across AI/ML, industry movements, infrastructure, and the developer world.AI and Machine LearningOllama has improved its runtime system for local machine learning models on Macs, integrating Apple's MLX framework to enhance performance on Apple Silicon Macs. This update, in preview, supports a single model and requires substantial hardware resources, reflecting a trend towards local model experimentation due to cloud service limitations and costs. Meanwhile, the introduction of 1-bit Bonsai LLMs offers a commercially viable large language model with 1-bit weights, improving efficiency for robotics and edge computing applications. Meta's BOxCrete model for sustainable concrete mix design is also notable, utilizing Bayesian optimization to support local production and sustainability, with open-source data available on GitHub.IndustrySalesforce has introduced AI-focused upgrades for Slack, adding 30 new capabilities such as reusable AI-skills for Slackbot, which can now draft emails, schedule meetings, and transcribe meetings. These enhancements position Slack as a versatile business tool. In funding news, Cognichip has raised $60 million to develop AI-designed chips, aiming to expedite and reduce costs in chip design through domain-specific datasets. Additionally, Oracle is reportedly laying off thousands of employees as part of efforts to manage costs in scaling AI infrastructure.Infrastructure and CybersecurityA significant cybersecurity incident involves AI recruiting startup Mercor, which experienced a supply chain attack linked to the open-source LiteLLM project. The hacking group Lapsus$ claimed responsibility for the breach, highlighting ongoing vulnerabilities in open-source projects. This incident underscores the critical need for robust cybersecurity measures in the tech industry.Developer WorldIn software development, Ollama has expanded its integration with Visual Studio Code, supporting local model experimentation. EmDash, a new CMS, emerges as a modern alternative to WordPress, focusing on plugin security and open-source development. TurboQuant's advancements in machine learning model compression for Apple Silicon devices further demonstrate the potential of combining hardware and software optimization, with implementations in Swift and Metal.These developments reflect ongoing innovation in AI efficiency, open-source software, and the application of advanced programming techniques to solve real-world problems, alongside the challenges of cybersecurity and infrastructure scaling.