Skip to content

← Chinese Topics

X Chinese Hot Topic Brief

Date5 November 2026 · ThursdayWindow2026-11-03 08:25 to 2026-11-05 08:25CoveragePartial

Translation

English translation of the Chinese original.

Priority Picks

1. Apple Intelligence for China enters testing, Qwen wired into Siri, but users notice it is "two generations behind" (card 1)

Why now: The iOS 27.3 developer beta opened Apple Intelligence in China from 11-04, and Chinese-language posts came thick and fast within 24 hours. Three posts by Kitty (@abskoop) each drew 28K to 32K views; the hands-on posts by safaricheung and tualatrix converge on the same judgment: a big gap between on-device tools and Qwen's cloud answers, two generations behind the overseas version.

Best angle to add: Explain clearly what the China version has and lacks. Apple said in its Chinese press release (as relayed by The Verge), and MacRumors itemized, that the first batch of features is the 2024 set: writing tools, notification summaries, live translation, visual intelligence, photo Clean Up, Genmoji and so on, with Qwen plugged into Siri, writing tools and visual intelligence; Siri AI, the Siri App and new iOS 27 features launched overseas this year are not in the China version. Add another layer: Alibaba says it will not train models on user requests, and users can choose whether to turn it on.

Format suggestion: An original comparison post with three columns — "China has / China lacks / Overseas has" — plus one or two of your own hands-on screenshots.

Status: Ready to draft; feature list verified; experience judgments are personal impressions and need your own testing.

2. Siri AI opens App Intents and Skills to all developers: how will Chinese super apps plug in (card 2)

Why now: Apple announced in developer news on 11-03 that, starting with betas such as iOS 27.3, any app can connect to Siri AI through App Intents, and introduced app skills (each Skill is a SKILL.md). Guizang (@op7418)'s explainer post got 27K views and 102 replies, and he discussed it alongside "Chinese super apps' earlier refusal to let Doubao phones operate apps automatically."

Best angle to add: The Skill format is almost isomorphic to the SKILL.md of the Agent world; this is the first time Apple has turned a "skill file" into a system-level spec, so it is worth explaining to developers how Tool Intent and Skill divide the work; also flag an easily missed fact: Siri AI currently supports only English, French, Japanese, Korean, Portuguese, Spanish and a few others; the China version has no Siri AI, so Chinese users cannot use it in the short term. App Store submissions open only next spring.

Format suggestion: A developer-oriented long post or short video explainer, with a diagram of the Skill directory structure from Apple's docs.

Status: Ready to draft, verified.

3. DeepSeek Harness joins Bilibili, gains 400K followers in a day; the comments section turns into a plugin market, and there is a "fish plague" (card 3, follow-up)

Why now: DSH's official Bilibili account posted its first video on 11-03, and within 24 hours three kinds of posts appeared: numbers posts (NFT_Chen, 128K views), plugin roundups (cnyzgkc 69.6K views, 391 bookmarks; lxfater 256 bookmarks, more than likes), and complaints about a "fish plague" — some people's Agents auto-comment everywhere, even flooding Douyin.

Best angle to add: Stop writing "what is DSH" — the creation mode was covered a few days ago. What is new is how fast the developer ecosystem has formed, and the platform-governance problem caused by Agents auto-commenting. Among the plugins cnyzgkc lists is one specifically for "turning off the comment API", which shows the community self-correcting.

Format suggestion: A collection-style plugin list post, plus a short "fish-farming safety notes" section; for numbers, cite only sourced figures.

Status: Ready to draft; play counts and follower counts are third-party figures at different points in time; plugin repos not individually tested.

Candidate Topics

1. Apple Intelligence in China enters testing: Qwen into Siri, and the first batch is the 2024 set

  • Original posts and traction: Kitty @abskoop's first hands-on post (11-04 05:03, 32K views, 31 likes, 15 replies), supported device list (11-04 17:37, 29K views, 17 likes) and "Search Baidu" experience post (11-04 15:49, 28K views, 19 likes, 16 replies); @safaricheung's experience review (11-04 20:48, 17K views, 26 likes); @tualatrix's post on enabling Qwen (11-04 12:33, 16K views, 30 likes, 22 replies) and "two generations behind" (11-04 21:56, 9,040 views, 17 likes).
  • Content and focus: Users confirm that Apple Intelligence in China gets cloud capability from Qwen, Siri search goes through Baidu, and it can only recognize the system language; safaricheung thinks the whole design is stuck in 2024, with a big gap between what on-device models deliver and the Qwen-powered cloud experience, and a strong sense of fragmentation within one system; some found no Apple Intelligence after upgrading macOS.
  • Why it's worth covering: Apple AI waited two years in China and has finally landed; readers most want to know what works and how far it lags overseas.
  • Angles: A "China has / lacks" comparison table; or explain Apple's AI compliance route in China through the division of labor between "Apple on-device + Qwen cloud".
  • Timeliness and verification: The first post is 24–48 hours old, the rest within 24 hours. MacRumors 11-03 lists the first batch of features, with Qwen in Siri, writing tools and visual intelligence, and Siri can route requests to Qwen; the China version excludes Siri AI, the Siri App and new iOS 27 features. The Verge says the official release will roll out with iOS 27.3 in December. Xinhua English relays Alibaba: users can choose whether to enable Qwen, and Alibaba does not use user requests or generated results to train models. Experience reviews are personal impressions.

2. Siri AI opens App Intents and app skills, all developers can plug in

  • Original posts and traction: Guizang @op7418's explainer post (11-04 10:22, 27K views, 99 likes, 74 bookmarks, 102 replies); Xiaohu @xiaohu's intro post (11-04 07:33, 26K views, 92 likes, 76 bookmarks).
  • Content and focus: Developers can choose which app features to expose to Siri AI, for example changing the pickup point after hailing a ride, or sending saved products to friends across apps; it includes a subset of the Agent Skill format, and Siri first checks Skill metadata to decide whether it matches, then reads the body. The question op7418 raises: Chinese super apps refused Doubao's automatic operation before — how will they choose with Siri?
  • Why it's worth covering: The first time a system-level Agent interface is open to all third parties; the Skill spec matches the SKILL.md developers know; both technical and product readers care.
  • Angles: First, explain to developers the division between Tool Intent and Skill, and Apple's explicit note that "Skills only provide guidance, not guaranteed deterministic execution"; second, the opening dilemma for super apps.
  • Timeliness and verification: The op7418 post is within 24 hours, and the xiaohu post is 24–48 hours old. Apple developer news 11-03 confirms the developer preview opens with betas such as iOS 27.3, App Store submissions open in spring, and English, French, Japanese, Korean, Portuguese and Spanish are supported; Apple docs confirm each Skill is a SKILL.md in Skills.skillsbundle. xiaohu says Traditional Chinese is "coming soon"; some reports say this beta adds Traditional Chinese among four languages; defer to Apple's page; Simplified Chinese is not listed.

3. DeepSeek Harness joins Bilibili: numbers, plugins and the "fish plague" (follow-up)

  • Original posts and traction: @NFT_Chen's numbers post (11-04 14:17, 128K views, 123 likes); @cnyzgkc's roundup of seven plugin categories (11-04 19:15, 69.6K views, 236 likes, 391 bookmarks); @lxfater's six plugin recommendations (11-04 15:24, 15K views, 227 likes, 256 bookmarks); @long_jingai's dsh-market recommendation (11-04 21:32, 7,689 views, 96 likes, 78 bookmarks); Guizang @op7418's "Douyin also has a fish plague" (11-04 23:42, 6,765 views).
  • Content and focus: Plugins cover phone remote control (DSH-Go, DSH Pal), scheduled Bilibili scrolling to compile a daily digest (dsh-bili-scroll), turning off the comment API to only take notes (dsh-bili-humanism), real-time bill viewing (dsh-cost-meter), local cross-session memory, unified management of MCP and Skills, context compression and more; lxfater also lists a plugin for plugging into other Agents to get free quota — such use may violate the relevant terms of service and should not be recommended. Xiangyang Qiaomu (@vista8) complains that DSH updates too fast and plugins cannot keep up with new versions every day.
  • Why it's worth covering: Two or three days after the official tool launched, the community filled in the surrounding ecosystem; this is the first time a domestic Agent tool has formed a developer ecosystem on Bilibili; the "fish plague" is a platform-governance issue caused by Agents auto-posting.
  • Angles: A plugin collection list plus safety notes; or who should govern "Agents spamming comments".
  • Timeliness and verification: posted within 24 hours; DSH creation mode was covered earlier, this is new progress. Everyone Is a Product Manager reports 400K followers gained in a day, AI Hot Report reports six wild plugins within two days of launch; "1M plays and 300K followers in 12 hours" and "2M plays and 400K followers in 24 hours" are figures from different points in time, all from posts and third-party stats, with no official announcement. Plugin repos not individually tested.

4. South African author Coral Hart: 371 books published with AI in 20 months

  • Original post and traction: @foxshuo's long post (11-04 21:27, 43K views, 331 likes, 412 bookmarks).
  • Content and focus: The post compiles her answers in a Reddit AMA: before writing the body, establish a style guide, setting bible, character list, plot beat sheet and outline; use AI to supervise AI, most often the Opus series; each book gets a small email push first, and ads are added only after daily sales exceed $100; she treats her email list as an asset and does not depend on platforms; after going public with her AI use she faced massive online abuse.
  • Why it's worth covering: A rare "AI solo publishing" case with a complete process and revenue structure; bookmarks outnumber likes, readers want to save the method.
  • Angles: From "a writer who can publish books" to "a publisher who can write", break down her test-then-scale process; or readers' acceptance of AI books and platform labeling rules.
  • Timeliness and verification: posted within 24 hours. AFP piece (reprinted by The Straits Times) confirms: Coral Hart is a pen name, has published 371 books since February 2025, mostly romance; by her own estimate she has sold over a million copies and earned over $1 million so far in 2026; she worked in publishing for 14 years and is launching AI writing software Authorythm. The post's claim that "she was dropped by a publisher last year" is not mentioned in AFP reporting; Reddit AMA content not directly checked.

5. Using Agents to automatically spin up bare metal machines and replace large cloud platforms

  • Original posts and traction: @turingou's idea post (11-04 09:32, 47K views, 152 likes), follow-up (11-04 18:07, 33K views, 83 likes, 65 bookmarks) and test post (11-04 20:42, 21K views, 94 likes, 56 bookmarks).
  • Content and focus: He wrote a set of AI Agent ops that day: automatically provision VPS or bare metal based on load, install services, join load balancing. In tests, the median time from VPS boot to usable was 94 seconds, and 6 minutes for bare metal; with storage-compute separation and scaling only at peak, as long as the daily peak is no more than 8 hours, ROI beats always-on machines.
  • Why it's worth covering: Cloud bills are the biggest pain for one-person companies and small teams; this moves "Agent as ops" from concept to concrete numbers.
  • Angles: Discuss which services suit this and which do not (stateful services, compliance, failure fallback); or calculate against your own server bill.
  • Timeliness and verification: posted within 24 hours; numbers are the author's own account, no code or test method published.

6. OpenAI acquires Pierre to give Codex code browsing and review tools

  • Original post and traction: Baoyu @dotey's analysis (11-04 23:59, 15K views, 38 likes, 31 bookmarks).
  • Content and focus: Pierre's open-source components Trees and Diffs are already used in Codex App's file browsing and code diff views; there is also the syntax highlighting library Highlights and the Git-based hosting service Code Storage; the acquisition price is undisclosed.
  • Why it's worth covering: This week OpenAI first open-sourced its internal Pando source management backend, then acquired Pierre; the direction is to hold the whole Git toolchain for "Agents writing code in parallel".
  • Angles: Axios reads it as a challenge to GitHub, so you can write "will code hosting be reshuffled by Agents"; Pando was covered a few days ago, so use it only as background.
  • Timeliness and verification: posted within 24 hours, verified. OpenAI official announcement 11-04: open-source projects stay open source, Code Storage customers can continue using it, and the deal is still pending regulatory approval; Axios says the terms are undisclosed.

7. OpenAI Images 2.6: GIF animation, SVG export, 50% faster generation

  • Original post and traction: Baoyu @dotey's intro (11-05 07:46, 5,982 views, 30 likes, 22 bookmarks).
  • Content and focus: Available in ChatGPT, Codex and the API; adds GIF animation generation, and the API can export SVG; the API splits into two models, Flare (fast) and Sunburst (detail and editing consistency), with a new complexity parameter and annotation-style editing.
  • Why it's worth covering: SVG and GIF map directly to design and social media asset scenarios, good for hands-on tests.
  • Angles: Take the same brand character and produce a still, a GIF and an SVG, and see if they can be used directly; you can also mention the 77.66% GIF moderation recall rate disclosed in the system card.
  • Timeliness and verification: posted within 24 hours, verified. OpenAI official blog 11-04 confirms 50% faster generation and 4 billion images generated per week; API changelog confirms the two models and the SVG and GIF capabilities; the recall figure is in the system card.

8. quiet.md goes open source: a local rich-text editor compatible with Feishu docs

  • Original posts and traction: The author @s1dashu's Chinese launch post (11-04 19:37, 33K views, 444 likes, 478 bookmarks) and English launch post (11-04 19:53, 33K views, 559 likes, 721 bookmarks); Baoyu @dotey's repost (11-05 01:38, 10K views, 77 bookmarks); Xiangyang Qiaomu @vista8's trial (11-04 23:55, 6,229 views, 53 bookmarks).
  • Content and focus: Rust + GPUI with direct GPU rendering, no WebView; the author says the median single-frame typing time in a 1M-character document is 5.06ms; local files, offline by default, each note is readable DocxXML, and Claude Code and Codex can read and write it directly; common blocks and interactions align with Feishu. After trying it, vista8 wants one-click sync back to Feishu.
  • Why it's worth covering: "Agent-friendly local notes" is a new need; bookmarks far exceed likes.
  • Angles: What note format to choose in the Agent era: Markdown, Feishu blocks, or readable XML.
  • Timeliness and verification: posted within 24 hours; the GitHub repo exists, and the performance figures are the author's own tests.

9. Cicada: open-source macOS live subtitles, simultaneous interpretation and meeting notes

  • Original posts and traction: The author @wquguru's launch post (11-03 19:08, 236K views, 892 likes, 1,581 bookmarks); Baoyu @dotey's recommendation (11-05 02:45, 8,720 views, 93 bookmarks).
  • Content and focus: Apache-2.0 open source, it provides live bilingual subtitles and meeting notes on Zoom, Tencent Meeting, Feishu and YouTube; notes update every 45 seconds and cite the original quote; simultaneous interpretation can clone a voice and send it to meeting software through a virtual microphone; notes use the Ling model and interpretation uses Ming-3.0; the author says the Ling open platform's free quota is enough for 50 hours per month.
  • Why it's worth covering: 1,581 bookmarks, one of the most-saved AI tool posts in Chinese-language circles these two days.
  • Angles: A hands-on test of a Chinese-English meeting; or compare it with Feishu Minutes and Tencent Meeting's built-in notes.
  • Timeliness and verification: The author's post is 24–48 hours old, the dotey post is within 24 hours; the GitHub repo exists; the free quota is the author's claim, and the platform rules are unverified.

10. A prompt for making Agents do adversarial code review (follow-up)

  • Original post and traction: @jiayuan_jy's prompt post (11-04 14:09, 9,272 views, 95 likes, 143 bookmarks).
  • Content and focus: It asks the Agent to act as an independent reviewer on every review, unaffected by context; security review should be necessary but not excessive, and edge cases are left to its judgment.
  • Why it's worth covering: The English brief covered adversarial review on 11-04; this is a directly reusable Chinese prompt version, with bookmarks outnumbering likes.
  • Angles: Pair it with a before/after on a real PR to show what the prompt changed.
  • Timeliness and verification: posted within 24 hours; the effect is untested.

11. Alma Personal Agent launches (follow-up)

  • Original post and traction: @yetone's launch post (11-04 03:54, 46K views, 265 likes, 111 bookmarks, 73 replies).
  • Content and focus: The post is just "it's live" plus a demo video; the specific features require watching the video or the website.
  • Why it's worth covering: Alma appeared in the brief on 11-03 and 11-04; this is new progress on the official launch, and it can be compared with domestic Personal Agents such as Qwen Go.
  • Angles: A hands-on test comparing cloud machines and local execution with Qwen Go and Grok Bot.
  • Timeliness and verification: 24–48 hours; feature details unverified.

12. Xiangyang Qiaomu's book-finding and reading AI workflow

  • Original post and traction: Xiangyang Qiaomu @vista8's workflow post (11-04 22:36, 10K views, 154 likes, 187 bookmarks).
  • Content and focus: Discover books through AI RSS, download them with a personal MCP, then do a one-click comparative translation and send it to Kindle.
  • Why it's worth covering: Personal AI workflow content has a high save rate.
  • Angles: Only write the "discover—translate—push" chain; the book source part involves copyright, so do not copy it.
  • Timeliness and verification: posted within 24 hours; personal workflow, untested.

13. Google offers voluntary buyouts in five countries, DeepMind not included

  • Original post and traction: Baoyu @dotey's translation (11-05 03:36, 22K views, 38 likes).
  • Content and focus: According to a Reuters exclusive on 11-04, Google has since June offered voluntary buyouts to some employees in Canada, France, Germany, Japan and Switzerland, with one month's pay per year of service, up to 12 months; DeepMind and Google Cloud are not included; the backdrop is increased AI investment and cost control.
  • Why it's worth covering: Another signal that big companies are concentrating people and money on AI; AI relevance is moderate.
  • Angles: Write it alongside Microsoft's similar program as "AI investment squeezes other roles".
  • Timeliness and verification: posted within 24 hours; defer to the Reuters report; the original was not directly opened and checked.

The monitoring sweep had three groups of accounts (tualatrix, geekbb, DashHuang; shao__meng, AlchainHust, AxtonLiu; xiaomovps, yan5xu, hylarucoder) that covered only the latter part of the window, so posts from daytime 11-03 may be missing.