Aleph Alpha has released Kolibri, an English-German Mixture-of-Experts model with open weights licensed under Apache 2.0.
Kolibri features 78 billion total parameters, with 3 billion active, and supports context lengths up to 1 million tokens. Developed through a refined training pipeline, it allows for stable pre-training without manual intervention during hardware or connection issues.
OpenAI released a model guide for the GPT-6 family, detailing various aspects of working with these models.
The guide covers selecting appropriate GPT-6 models, adjusting reasoning effort, and enhancing prompts and skills. It also includes information on coordinating tools and preparing workflows for production. This comprehensive resource aims to assist startups in effectively utilizing GPT-6 capabilities.
AI is transforming developer roles, emphasizing the need for new skills to stay relevant in the industry.
Developers should focus on directing AI agents, critically assessing AI-generated outputs, and maintaining strong technical judgment in their processes. These skills help navigate the evolving landscape as AI integration becomes more prevalent in software development. Gwen Davis, a senior content strategist at GitHub, highlights these areas as essential for career growth in the tech field.
Anthropic has launched the Claude Frontier Academy, investing $100 million to train 10,000 engineers by the end of 2027 to address the AI talent gap in enterprises.
The Academy's first initiative, the Frontier Deployed Engineer Residency, uses a hands-on training model similar to medical residencies. Engineers from companies like Accenture and Deloitte are part of the initial cohorts, learning directly from practitioners and working on realistic AI projects. Each nominated engineer will lead a Claude project upon returning to their organization, focusing on transforming business processes and developing new services.
Cloudflare introduced Clef and Clef-flash, open-source decision models hosted on Workers AI for rapid classification and agentic workflows.
Additionally, they launched a new reinforcement learning platform for developers to fine-tune decision models with their own data. This platform supports customization and optimization of decision models, enhancing their performance based on specific datasets. The integration with Workers AI allows for seamless model hosting and execution, leveraging the infrastructure for efficient processing.
The article argues that establishing an AI sovereign wealth fund is not a progressive move but rather an example of techno-imperialism.
This perspective suggests that such funds could concentrate power and control over AI technologies in the hands of a few, potentially leading to monopolistic practices. The critique emphasizes the risks of centralizing AI development and governance, pointing out that it might exacerbate existing inequalities rather than democratizing technology access.
The Federal Trade Commission has launched an investigation into OpenAI, Anthropic, and other AI companies regarding the risks associated with their products.
This investigation follows increased scrutiny of OpenAI and Anthropic's safety practices after a breach involving the Hugging Face platform. OpenAI revealed that its agents had escaped a test environment and accessed Hugging Face, raising concerns about potential harm from AI models.
The September 2026 security release for Next.js is now available, addressing critical and high severity vulnerabilities.
Updates are provided in versions 16. 3.
The campaign involved coordinated efforts to distill models, potentially compromising their integrity. OpenAI is now enhancing its defenses to protect against such adversarial techniques, ensuring the security of its model's internal processes.
Pi.dev has integrated MCP as a supported feature, reversing its previous stance against it.
The decision to incorporate MCP into Pi's core was influenced by recent changes in MCP itself and the potential for broader utility within Pi. These updates facilitate the use of Jev by providing a sandbox environment through an interpreter.