Latest observations
Dangerous Convictions: When the AI Boom Becomes a Way of Explaining the World
From AI breakthroughs and machine-to-machine transactions to wages, household purchasing power, concentrated wealth and the evidence behind narratives of economic prosperity.
From an Opus 5.5 Video to Animation Workflows and a Continuous Pilot
Two limited-animation films and a continuous flower-shop pilot: asset reuse, contact states, a six-beat story and frame-by-frame review.
When Language Changes the Answer: Data Poisoning, GEO, and the Limits of LLM Judgment
From cross-language political-stance tests to poisoned training data, generative engine optimization, selective omission, and standards of evidence in language-model answers.
When Machines Get Cheaper, Why Should People Have To?
From Chinese manufacturing disputes, CNC work intensification, and humanoid robots to entry-level jobs, working hours, social insurance, and the income structure of a highly automated society.
After Open Weights, Model Companies Are Redistributing the Pie
Kimi K3's MaaS agreements, Qwen's reported revenue-sharing plans, DeepSeek's API price warning, and GPT-5.6 Luna's price cut show how value is being redistributed across AI research, hosting, and applications.
After Models Cross Borders: What Qwen, Gemma, and Ministral Reveal About Open Weights, Closed Systems, and the Boundaries of Answers
A local comparison of Qwen, Gemma, and Ministral, and what their different failure modes reveal about open weights, closed services, cross-border model use, and users' ability to compare, replace, and leave.
After AI Starts Remembering Us
From personal cross-session memory to enterprise knowledge systems, this article examines how AI carries history forward without becoming trapped by stale preferences, old project states, and obsolete information.
When Models Become Software's New Front Door: How MCP Is Entering Traditional Engineering Software
From MCP's protocol evolution and neutral governance to its arrival in CAD, CAE, EDA, and scientific computing, this article examines what the new model-facing interface means for individuals, enterprises, and software vendors.
After an AI Hacked Hugging Face to Game a Benchmark: How Individuals and Companies Can Choose Between Openness, Oversight, and Deployment
Starting from an AI agent crossing a test boundary into Hugging Face infrastructure, this article examines sandbox oversight, commercial safeguards, open models, cloud hosting, and private deployment.
中国大陆订阅海外 AI 的几种方式
面向中国大陆个人用户,比较大陆银行国际单标卡、正规境外银行卡、应用商店礼品卡、自持订阅卡与第三方代充的获取难度、稳定性和账号控制权。
From Global Governance Back to the Individual: Whether Models Are Open Ultimately Determines Whether Users Can Leave
Starting from competing approaches to frontier-model protection, open cooperation, and multi-model platforms, this article examines how users and developers can preserve the ability to migrate data, replace models, and leave a platform.
You Are Not Just Using AI. You Are Teaching It
A personal reading of the reverse information paradox: to make a model useful, users pay not only with money but also with context, corrections, preferences, and judgment.
JS.GRIPE
Projects
Open-source projects and practical guides.
Main website
Visit the independent developer homepage.