China's AI Models Gain Traction Amid US Restrictions

@levelsio found Kimi K3 can complete tasks (e.g., Windows XP emulation) that Claude Code refuses due to safety restrictions, driving users to Chinese models. — via 1 2
@levelsio predicts browsers will be replaced by AI agents within two years, as most people have not yet deeply used AI. — via 1
Arvid Kahl notes all frontier AI models have likely absorbed all Stack Overflow data, raising training data exhaustion concerns. — via 1
Jon Yongfook argues AI will improve slower in design due to lack of high-quality training data (unlike open-source code). — via 1
@levelsio shares his monthly income ($23k) and urges creators to diversify revenue sources as AI threatens traditional content. — via 1

@levelsio extensively tested Kimi K3 and found it capable of emulating Windows XP—a task Claude Code refused due to safety restrictions. He argues that overzealous safety measures in US models waste time and weaken capabilities, while Chinese models attract users by offering unrestricted functionality. — via 1 2
@levelsio further claims that US export controls on AI models are ineffective, as users simply switch to Chinese alternatives, and that the US should let everyone use American models to maintain surveillance. — via 1
France and Germany plan to jointly develop a roadmap to address trade imbalances with China, aiming to present it by autumn. Proposed measures include accelerating EU trade investigations and establishing emergency protection mechanisms, which could impact tech trade. — via 1

@levelsio predicts the AI bubble hasn't burst because most people haven't deeply integrated AI into their workflows. He envisions browsers being replaced by AI agents within two years, with webapps becoming the database layer for agents. — via 1
Arvid Kahl highlights the key benefit of agentic work: dramatically faster iteration cycles. However, he warns that humans must deliberately slow down to learn consciously, ensuring they are truly processing rather than just trying. — via 1

Arvid Kahl asserts that every frontier AI model has likely ingested the entirety of Stack Overflow's conversations, questions, and answers, signaling a potential plateau in code-related training data. — via 1
Jon Yongfook provides a counterpoint for design: AI progress will be slower than in programming because high-quality design data is scattered across books, design systems, private emails, and designers' minds, lacking a centralized open-source corpus. — via 1

@levelsio breaks down his monthly income (~$23,000 from X, merch, and book sales). He explains that due to AI's threat to replace content, he is shifting toward authentic, non-scalable content to diversify revenue. — via 1
Arvid Kahl defends ads and sponsorships as critical for small and medium creators, criticizing those who consume content for free while undermining ad revenue—effectively subsidizing large corporations. — via 1
Tw93 shares four Aha moments from his product Mole: low AirPods battery auto-reminder, cleaning 86GB of unused build files, always-on screen for remote control, and iPhone battery display—all driven by user feedback. — via 1

Discover