Chinese original
→科研数据集一年暴增 3.9 倍：当知识生产的基础设施开始“为 Agent 而不是为论文排版”
从 Hugging Face 的 Project ProtoStar、MIT 的物理世界模型到麦格劳希尔接入 MCP，知识生产机构正在把核心交付物从“给人读的论文与电子书”，重构成“给模型和 Agent 调用的数据基础设施”。
People and AI
How working with AI changes our sense of ability, creativity, and human agency.
AI and Education: Which Historical Transformations Offer Useful Parallels?
Writing, printing, calculators, school systems, and MOOCs offer ways to understand how AI may change learning—and where historical analogies reach their limits.
ICLR Submissions Passed 60,000—Faster Than I Expected
A comparison of ICLR submission volumes raises questions about AI-driven research growth and the rules that might match its pace.
OpenAI Math: After 722 Papers
From a few major mathematics problems to hundreds of manuscripts at once: how long will humanity need to digest the results, and what if next month brings a hundred times as many?
How Should We Think About OpenAI’s AI Text Watermarking?
How textGrain embeds a signal, what false positives mean, and why detecting AI involvement does not settle a work’s correctness or value.
We Tested Jev on 1,711 Gaokao Questions
A model that makes judgments instead of generating text answered 90.8% of 1,711 exam questions correctly. Seventeen essays offered a different test.
Reading MIT's AI Education Report with ChatGPT (Part Four)
As AI keeps changing, how can a university keep observing, experimenting, measuring, and revising its judgments?
Reading MIT's AI Education Report with ChatGPT (Part Three)
If AI can provide information and help at any time, why come to a university to live and learn with other people?
OpenAI's Mathematics Breakthrough and Another Kind of Scaling
What a mathematical research effort built on massive parallel exploration suggests about how AI can tackle difficult work.
A Thought Experiment: What Should We Learn If AI Can Do Everything?
Why learn if AI can do every cognitive task? A thought experiment on education's purpose, followed by an argument for assessing and defending AI-assisted work.
Designing What an Agent Sees After a CLI Command
A CLI delivers an action to a system—and the context an agent needs for its next step.
Ten Days with GPT 6 Astra: A Few Impressions
After ten days with GPT 6 Astra, I keep thinking about disappearing credits, the limits of my own understanding, and whether tokens will become an everyday measure.
GPT Added One Number to My Grill-Me Exercise
A confidence score from 0 to 100 revealed something a correct answer alone could not: the learner's sense of how well they understood it.
Reading MIT's AI Education Report with ChatGPT (Part One)
When AI can complete assignments and solve problems, what is the value of students learning for themselves?
Reading MIT's AI Education Report with ChatGPT (Part Two)
If an excellent assignment no longer proves a student has learned, how should courses and assessment change?
PPT Master from 1.0.0 to 5.0.0: Designing Content Production with Agents
PPT Master offers a concrete case of how agentic content production inherits experience, selects capabilities, preserves decisions, and allocates authority among people, agents, and tools.
A Skill That Changed How I Learn English
I thought I knew words like check, charge, and over. A teacher showed me how to look beyond their Chinese translations—and I turned that way of learning into a Skill.
