AI Updates on 2025-10-31

Kimi introduces CLI Technical Preview and Kimi For Coding with shell-like UI, Zsh integration, MCP support, and Agent Client Protocol compatibility @Kimi_Moonshot
OpenAI launches agent mode for ChatGPT, allowing it to take actions, research, plan, and complete tasks while users browse, now available for Plus, Pro, and Business users @OpenAI
OpenAI introduces Sora characters feature and launches ability to purchase additional generations beyond the free daily limit due to unexpectedly high demand from power users @billpeeb

OpenAI begins hiring junior software engineers, calling them "super juniors" due to their significant impact, with Head of ChatGPT Engineering noting they bring fresh perspectives and new ways of working @GergelyOrosz
Getty Images signs multi-year licensing agreement with Perplexity, causing Getty shares to jump 25% and legitimizing some of Perplexity's previous use of Getty's stock photos @AndrewCurran_
AI-generated song by Xania Monet (created using Suno) becomes first AI song to enter a Billboard radio chart, with creator signing a $3 million record deal @AndrewCurran_
Amazon cloud revenue grows 20% amid strong AI demand, with AWS continuing to see robust demand for cloud infrastructure services in the AI era @TechCrunch
Both Cursor and Windsurf new models are speculated to be built on Chinese base models, with Cursor Composer showing Chinese reasoning traces and Windsurf potentially using customized GLM 4.6 model @deedydas
China has overtaken the US in cumulative open-source AI model downloads, highlighting the competitive landscape in AI development @a16z
Linear reports that 60% of enterprises have added agents to their workspaces since launching their agent platform, demonstrating rapid enterprise adoption @karrisaarinen

Stanford HAI warns that the tide of openness in AI is receding, threatening the foundation of scientific progress, and calls for universities to reclaim AI research for public good @StanfordHAI
Yann LeCun argues that concentrating AI within a handful of companies poses a significant threat to democracy, emphasizing that open source platforms are essential for countries to maintain sovereignty and build culturally-appropriate AI @youtubejocoding
Microsoft's AI Diffusion Report reveals clear global divides in AI adoption, highlighting the need to expand access, build skills, and make AI work for every language and community @BradSmi
Ethan Mollick calls for more specific efforts to make AI benefits work for more people and mitigate obvious harms, noting that many interventions could yield significant benefits today rather than waiting for long-term solutions @emollick

Stanford Health develops ChatEHR, an AI chatbot platform for healthcare that integrates real-time data, strict privacy controls, and complex EHR systems, potentially serving as a model for health systems @StanfordHAI
Google launches Pomelli, an AI marketing tool designed to help small and medium businesses connect with their audiences faster @GoogleAI
Perplexity Finance now includes politician holdings of public stocks, expanding the platform's financial data capabilities @AravSrinivas
Google adds Gemini CLI extension for Jules agent, accelerating creative coding workflows @GoogleAI
NotebookLM Chat receives improvements including enabling the full 1M token context window for enhanced document analysis @GoogleAI

Hugging Face releases comprehensive 214-page "Smol Training Playbook" covering pretraining and post-training recipes, hyperparameter exploration, and practical model training guidance @Thom_Wolf
Research suggests switching from BF16 to FP16 provides fundamental solution for RL fine-tuning by offering 8 times more precision, reducing policy divergence between training and inference engines @natolambert
MIT researchers develop method enabling artists to design realistic simulations of elastic objects for animated movies and video games @MIT
Microsoft researchers receive Best Paper Award at ESEM 2025 for exploring challenges of cross-disciplinary collaboration between software engineers and domain experts in AI, health, and science @MSFTResearch
François Chollet emphasizes that human intelligence involves constant invention, noting that even babies must invent crawling from scratch with minimal data, challenging assumptions about AI intelligence requirements @fchollet
Yann LeCun argues that the term "AGI" makes no sense because human intelligence isn't general but specialized, advocating instead for building "World Models" that understand the physical world through abstract representations @youtubejocoding
Marc Andreessen discusses the US-China AI race, predicting the next phase will be fought in robotics rather than software, emphasizing the need for embodied intelligence beyond current disembodied AI systems @a16z