Gemma 4 has introduced Multi-Token Prediction (MTP) to enhance inference speed. New checkpoints optimized with Quantization-Aware Training (QAT) have been released to improve efficiency for mobile and laptop use.

Microsoft's new AI project, Scout, aims to increase user engagement to the point of addiction, according to a leaked internal strategy document. CEO Satya Nadella publicly denied the claims made in the document, expressing disbelief over its contents.

New York lawmakers have passed a bill imposing a one-year moratorium on permits for new large-scale data centers, which will be the first such freeze in the nation if signed by Governor Kathy Hochul. The ban is a response to rising electricity costs attributed to the energy-intensive demands of data centers supporting the artificial intelligence sector.

A distributional analysis of rsync releases examines whether Claude-assisted versions have a higher incidence of bugs. The analysis addresses concerns of bias by detailing the methodology used to gather and evaluate the data.

Programmers are utilizing Claude to maintain handoff documents for larger projects, detailing plans, completed tasks, and relevant information. This practice allows subsequent instances of Claude to quickly get up to speed by reading the documentation.

The European Commission is proposing the European Technological Sovereignty Package to enhance Europe's control over its tech ecosystems. This initiative is accompanied by an EU Open Source Strategy aimed at promoting open-source software and technologies.

Ladybird will no longer accept public pull requests, with code changes now solely introduced by project maintainers. This shift aims to establish a tighter development process, enhance security, and limit contributions to a smaller group of individuals as the project moves toward its first alpha release.

Fine-tuning an LLM can enable it to generate documentation in a style reminiscent of 1995. Local deployment of specialized LLMs is anticipated to become more common among tech writers by 2030, although current powerful connected models dominate.

The Pentagon is utilizing an AI-driven content mill called La Tilde to disseminate propaganda to internet users in Latin America. La Tilde, which publishes articles in both Spanish and English, aims to present itself as a modern media brand for Latin American audiences.

Magenta RealTime 2 (MRT2) is an open model and real-time inference engine for building and playing AI musical instruments on laptops, specifically requiring Apple Silicon. MRT2 allows live interaction through MIDI, audio, and text inputs, enabling low-latency on-device inference.