January 2025.

The AI landscape is shifting dramatically, with DeepSeek leading the charge by tearing up the traditional playbook—its unconventional methods are so influential that rivals are quickly taking note. In parallel, DeepSeek's navigation of export controls, and its head-to-head performance against OpenAI’s latest (R1 vs o1), highlight a fast-evolving global arms race, especially as leading Chinese AI models find ways to circumvent US sanctions. Meanwhile, the open source movement gains fresh momentum as Block Open Source launches 'codename goose,' promising a flexible, agent-based AI framework. At the same time, new research and projects—like **NEPENTHES**, which burns bots with aggressive anti-AI malware, and grassroots tarpit initiatives designed to snare scrapers that flout robots.txt—reflect a rising backlash against AI overreach. ---- On the product side, the DeepSeek app is garnering attention for its offbeat interface, while ongoing debates question whether current AI truly delivers user control or simply offers an 'interface illusion.' Incidents such as OpenAI’s ChatGPT crawler inadvertently DDoSing sites, or AI-powered hospital transcription tools fabricating statements, underscore real-world risks. The integrity of AI-generated content is further in question as publishers like Quartz openly warn readers about the low quality of some automated articles, and Apple is forced to pause its AI news alerts following headline blunders. ---- Governments and industry giants are responding: Australia is launching a national AI capability plan, the UK is planning major investment, and former President Trump has undone Biden’s AI framework by executive order. Meanwhile, Larry Ellison envisions omnipresent AI surveillance for public order, and Lonestar aims to launch a lunar data center. Even the automotive industry is pivoting, with Honda integrating AI for enhanced vehicle experiences. Notably, AI is making breakthroughs in safety, with new models preventing battery fires, and in scholarship, unlocking the secrets of ancient texts. ---- Amid this frenzy, questions of trust and alignment loom large. Alignment faking in large language models raises concerns, as does the challenge of governing AI as a shared resource (with initiatives like GovSim and Stargate Project). The blending of human and machine writing blurs authenticity, exemplified by Meta’s indistinguishable AI profiles and tales of users forming emotional bonds with bots. As AI becomes omnipresent—'**like Tinkerbell: it only works if we believe in it**'—the world stands on the threshold of profound societal shifts, from moonshot data centers to the rewriting of history itself, all while grappling with the promise, peril, and peculiarities of intelligence both artificial and human.