This weekend, I curated and translated several AI research papers, reviews, or annual summaries: 1️⃣ Andrej Karpathy's *'2025 LLM Year in Review'* > He shares his thoughts after accidentally coining the concept of 'Vibe Coding.' https://((((((randomarea.com))))))/year-in-review-2025/ 2️⃣ Anthropic's paper *'Bloom: An Open-Source Tool for Automated Behavior Evaluation'* > With Bloom, you only need to define a 'behavior seed,' and Agents can automatically evolve thousands of dynamic test scenarios, enabling precise quantification of hidden alignment risks in large models at an extremely low cost. https://((((((randomarea.com))))))/bloom-auto-evals/ 3️⃣ Google's *'The Safety of Distributed AGI'* > This paper argues that AGI is highly likely to emerge from multi-agent collaboration (aka 'Patchwork AGI') and introduces the innovative 'Virtual Market Economy' mechanism for governance, offering a novel system-level solution to address the risks of distributed intelligence. https://((((((randomarea.com))))))/distributional-agi-safety/ 4️⃣ OpenAI's *'Evaluating Chain-of-Thought (CoT) Monitoring'* > This study reveals a truth about AI safety: for humans to maintain control, AI must learn to 'think slowly.' https://((((((randomarea.com))))))/monitoring-monitorability/ 5️⃣ NeurIPS 2025 Best Paper https://((((((randomarea.com))))))/announcing-the-neurips-2025-best-paper-awards/ Full list: https://((((((randomarea.com))))))