Articles from X
A critical thread analyzing the Qwen-AgentWorld paper, which proposes language world models for general agents. The critique raises concerns about simulator fidelity, benchmark design, and cost, rating it 4.5/10 on a bullshitometer.
Fable 5, a new Anthropic model, has appeared in the Amazon Bedrock catalog, signaling an imminent release.
Share a curated AI evaluation (evals) resource library, including high-quality blogs, podcasts, papers, and projects, compiled by Xiangyi Li.
Tweet announces Gemma 4 31B multimodal model with high speed, calling it a step towards superintelligence.
Someone packaged Xiaohongshu post operation experience into a Codex Skill (XHS Visual Director) for automatically judging content, breaking it down, and generating images, replacing ad-hoc prompts to improve exposure efficiency.
OpenAI has publicly released the AGENTS.md development guide for the Codex project, covering internal specifications and governance details for the Rust codebase.
Covering Figma's CONFIG 2026 live from San Francisco, sharing updates in a thread.
Sentient Foundation launched a $42M open-source AGI funding program with two tracks: grants with no equity and investments for commercial open-source AI products, focusing on technical quality and ecosystem value.
Promotes a structured MIT deep learning course that covers foundations, generative models, agents, and sequence problems. The course aims to build practical understanding before advanced topics.
Youssef Hosni announced that his LLM Roadmap book, which became a bestseller on To Data & Beyond, is now open source and freely available, aiming to provide a clear learning path for LLMs and generative AI.
Starlink will provide satellite internet to support emergency communications across all 78 municipalities in Puerto Rico, as announced by Governor Jenniffer González Colón.
Dan Shipper interviews Edwin Chen, CEO of Surge AI, about AI progress, the potential for AGI, and the implications for human motivation and uniqueness. They discuss AI's ability to solve novel math problems, the pitfalls of optimizing for engagement, and why AI still struggles with writing.
Moxie Marlinspike's Confer is developing end-to-end encryption for AI chats, with posts on connectors, encrypted memory, private inference, and more.
Runlayer raises $30M from Felicis and Khosla Ventures to help companies adopt AI with security and control.
Introduce a tool that by changing 'Github' to 'ARGithub' in any repo URL, deploys an agent to orient itself on the codebase, resolve setup issues, and run experiments.
A tweet recommends using vLLM instead of Ollama for local AI, citing better GPU utilization, higher efficiency, and up to 2x faster performance in tests. vLLM is a fast, open-source library for LLM inference and serving that supports many models and hardware backends.
LangChain is sponsoring VB Transform 2026, a conference on agentic AI orchestration, taking place July 14-15 at Hotel Nia in Menlo Park. Use discount code VBLangChain20 for registration.
Binance CEO Changpeng Zhao comments on CEA Industries ending a proxy fight with YZi Labs and appointing Ella Zhang to its board of directors.
Recommends four open-source paper writing skill packs suitable for machine learning/computer vision/NLP and other fields, focusing on structure standardization, polishing and review, complete research workflow, and Chinese collaboration, supporting AI assistants such as Codex, Claude Code, and Gemini.
Genspark launches Genspark Design, an AI design tool powered by Claude Opus 4.7 that can create UI prototypes, posters, videos, HTML animations, and convert designs into code, aiming to be a full creative production tool.