China Is Training Developing Countries on AI for Free. The US Has No Equivalent Program.

Code faster without cutting corners.

Why spend hours figuring out the best way to use Claude when proven workflows already exist? Learn the shortcuts top AI developers rely on every day.

Unlock 100+ practical Claude Code techniques to write cleaner code, fix bugs faster, automate repetitive tasks, and boost your productivity.

1  Alibaba released Qwen3.8, claiming it rivals the best from OpenAI and Anthropic and trails only Claude Fable 5.

Qwen3.8 is now in live preview and arrives days after Moonshot AI shipped Kimi K3. That is two competitive Chinese frontier models in the same two-week window, with Meituan's LongCat-2.0 having shipped the week before. The cadence matters as much as any individual model claim: Chinese labs are not just releasing competitive models, they are sustaining a release pace that keeps pressure on US labs continuously rather than episodically.

2  Xi Jinping called for global AI cooperation at Shanghai's World AI Conference and offered 5,000 free training spots to developing countries.

At the World AI Conference in Shanghai, Xi argued that AI should not be dominated by any single nation and called on countries to build the technology together. The 5,000 training spots offered to developing nations are the concrete commitment behind the rhetoric. The US currently has no equivalent international program. When countries in the Global South look for AI education and infrastructure partnerships, China is increasingly the only offer on the table.

3  Decart's Lucy 2.5 edits video in real time with near-zero lag, opening a path from live AI effects in film to live broadcast and commerce.

Lucy 2.5 is Decart's most capable model yet, built to add or remove objects, restyle frames, and apply visual effects to live video at speeds that make real-time use practical rather than aspirational. Decart sees the technology powering e-commerce, streaming, and advertising. The more significant shift is directional: the polished visual work that previously required post-production now has a plausible path into a live broadcast.

Mac OS, vibecoded. Analyst Max Weinbach used Kimi K3 to build a working replica of the Mac operating system in a single session (4M views). The project has been appearing in discussions about what the new generation of Chinese frontier models can do in practice. Try the Mac replica here.

40 million views for a font. The Decoy Font, designed to confuse AI models by mixing sharp letterforms with blurred backgrounds, has gone viral at 40M views — double the reach of Ghost Font from two weeks ago. The speed at which these human-readable, AI-confusing typefaces are spreading says something about how many people are thinking about AI-proof design.

Gym analytics, overkill edition. An AI hobbyist reverse-engineered his gym's API using GPT-5.6 Sol to track crowd levels and find the quietest times to train. One reply in the thread jokes the project is doing way too much (1.5M views). The underlying technique — using an LLM to interpret and query an undocumented API — is genuinely useful beyond the gym use case.

42 AI skills, bookmarked 9,000 times. A post outlining 42 AI skills that cover nearly every business function picked up 9,000 bookmarks in the first four days. The volume of saves suggests people are treating it as a reference document rather than a read-once thread.

You're ahead of almost everyone. A graph from VC firm a16z showing how few households currently pay for AI has reached 1M views. The number is lower than most people in the AI space assume. If you use a paid AI tool regularly, you are in a small minority of the global population — which is either reassuring or alarming depending on your prior assumptions about

Find 10 business ideas for any audience in one prompt

This ChatGPT prompt takes any target audience and returns ten structured business ideas grounded in real problems that audience faces. For each idea it maps the problem, the limitations of current solutions, the proposed fix, the value proposition, and a specific monetization path. It is designed to surface ideas you can evaluate immediately rather than concepts you need to develop further before they are useful.

COPY AND PASTE THIS PROMPT

Replace [TARGET AUDIENCE] with the most specific description you can write. "Freelance designers" produces better output than "creative professionals." "Solo founders running SaaS products under $10K MRR" produces better output than "startup founders." The more the prompt can picture a specific person with specific constraints, the more actionable the ten ideas it returns. For a second pass, pick the two strongest ideas from the first run and ask the same prompt again with each one as the audience — it will surface the sub-problems within the problem you started with.

Top engineers at Anthropic say AI now writes 100% of their code.

Are you using AI to write yours?

These 100+ Claude Code hacks show you exactly how. Sign up for The Code and get:

Claim your free playbook