hotai.newsVisible Chinese AI discussions yesterday covered OpenAI product experiences, Claude and development workflows, domestic models and open-source ecosystems, and practical agent usage. This issue includes 15 posts from individual accounts, sorted by interaction volume at the time of verification, with a maximum of two posts per author. Summaries use short excerpts from the original posts; see the original posts for details. Candidates come from public trending lists and do not represent the complete ranking of the entire X platform; interaction numbers are snapshots taken during collection.
Excerpt from original post: Excellent article! Recently, several "decision models" have emerged, such as TypeSafe's Jev, Perplexity's Decider, Cloudflare's Clef, and OpenAI's Decisions API, which output...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.From the original post: Turks complain that Chinese people know everything, playing historical memes with expertise, leading some to suspect Chinese netizens are AI bots or internet armies. From the Chinese perspective, it seems they wonder why Turks know nothing, making them easily fooled by fabricated history. This makes me want to go to Turkey...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.Excerpt from original post: Oh no, Chinese models might soon be unable to compete with GPT... Yesterday, @OpenAI released an internal report on a security incident involving its Agent. After carefully reviewing the publicly shared Chain-of-Thought (CoT) data, I noticed it barely resembles normal English anymore. A large number of English words are being...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification. · Verification needed: Personal viewpoints or experiences mentioned in the post have not been independently confirmed.Excerpt from original post: My experience is that with cutting-edge models like Claude, if you don't specify otherwise, it defaults to this approach. It has already been trained sufficiently, so there's no need for overly deliberate prompting. On the contrary, because AI tends to use shadcn/ui, everything ends up looking the same, causing visual fatigue for viewers...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.Excerpt from original post: I think most people haven't realized that opening DeepInfer by @deepseek_ai yesterday was truly groundbreaking. Yesterday, DeepSeek directly opened its core inference framework, DeepInfer, which supports its large-scale production services...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.From the original post: The DeepSeek announcement is equally unbelievable. In the V4.1 Flash deployment documentation, the official specifically warns that deploying across data centers might encounter fiber optic cables being severed by construction work. Reading this makes me suspect the previous outage was indeed caused by a severed fiber cable.
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.Excerpt from original post: Sweeping the globe! Just last night, leaders from multiple countries, government cybersecurity departments, and AI companies issued urgent statements regarding a critical vulnerability in the PHP programming language. Nations including Finland, the Netherlands, and France have already taken down their government websites. The cause is that PHP recently exposed two extremely high-severity vulnerabilities...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.Excerpt from original post: I've been completely subdued by Grok Bot; I say nothing anymore, accepting my knees in submission. Here are my $200 plan details. This guy is a beast when it comes to work—so fast, so smooth. A project I expected would take several days to refactor, I let him handle...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.From the original post: A screenshot from an internal PPT at Southeast University School of Medicine recently circulated throughout the medical community... The overall doctor attrition rate reached 43.6%! In the Medical Imaging specialty, 21 were recruited, 15 left, leaving only 6—a 71% attrition rate. The Medical Laboratory specialty is nearly wiped out. Even in areas where recruitment was historically strongest...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.From the original post: Combining Pi with DeepSeek can essentially create any vertically specialized personal agent. The remaining task is to identify where true value exists for these specialized agents. In essence, much of the value lies in areas that are essential human needs but which large models refuse to answer.
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.From the original post: What is this permanent caching? OpenAI has gone too far this time. The API's Prompt Cache can now be retained for up to 12 hours. But Codex and Work are even more extreme: cache for logged-in users can be retained permanently...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.Excerpt from original post: I spent two days automating my Agent tasks to create an engaging pipeline visualization. The reason is that in my daily routine, there are dozens of automated tasks running—scheduled tasks within Claude and Codex, as well as webhook middleware triggered by custom...
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.Excerpt from original post: Amazing! The official report states a speed of 139 km/h. AI analyzes the video based on surveillance footage; at less than 50 km/h, please @grok analyze the video to determine the actual vehicle speed 😄
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.From the original post: iPhone Honor of Kings, let Codex play... This could be the first of its kind? Video below. Although GPT-6 alone performs like a fool in real-time action games (turn-based games should be fine), it verifies that AI is truly general-purpose.
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.From the original post: Personal assistants are impossible to succeed with in China. Supermarkets, banks, credit cards, airline miles... No sector will open its interfaces for you. AI belongs to Americans.
Reason for selection: Yesterday's Chinese-language personal AI discussions, ranked comprehensively based on likes, shares, replies, and saves at the time of verification.