AI & Machine Learning
A language model called Mr. Chatterbox has been trained entirely on out-of-copyright text from the British Library, producing a model rooted in Victorian-era language and ethics. Simon Willison covers the release and provides an LLM plugin for running it locally.
Ollama has added preview support for Apple's MLX framework on Apple Silicon, aiming to improve local inference performance on Macs.
Google Research has open-sourced TimesFM, a 200-million-parameter foundation model designed specifically for time-series forecasting with a 16k token context window.
Research examining how AI agents can exhibit chaotic or unintended behaviors, exploring the safety implications of autonomous agent systems.
Financial Times · www.ft.com
The Financial Times argues that as AI disrupts the economy and the population ages, the US tax system's disproportionate burden on young workers versus capital needs to be rebalanced.
Security & Supply Chain
The popular Axios npm package was compromised with malicious versions that install a remote access trojan, marking the second major supply chain attack in a week after the litellm PyPI incident.
An essay arguing that the current state of vulnerability research is deeply troubled, examining systemic issues in how security vulnerabilities are discovered, disclosed, and handled.
An investigation reveals that US government-published apps contain Huawei spyware components and invasive surveillance features, raising questions about privacy double standards when the government bans foreign apps for similar concerns.
Google Research outlines a responsible disclosure framework for quantum computing vulnerabilities that could threaten cryptocurrency systems.