China Desk — 23 Aug 2026, 16:36 UTC

Ubtech brings a real customer production line to the robot show

# QbitAI 1d

Ubtech recreated customer factory tasks at China's robot conference, using autonomous humanoids for loading, palletizing and sorting to argue that reliability matters more than stage demos.

The displayed workflows covered box handling, machine loading and mixed-product sorting, with the company reporting sub-millimeter positioning and peak sorting throughput near 1,100 items per hour.
The robots reportedly handled the full loop from visual identification through grasping, movement, placement, inspection and retry without human intervention.
This is a different engineering target from a dancing robot: industrial units need sustained loads, contact-force control, local model inference, calibration tolerance and recovery from failed grasps.
The company says its systems are already in small-scale customer delivery and are being expanded according to field performance, though the supplied account does not provide independent uptime or failure-rate data.
Ubtech's hardware distinction is consequential: work-oriented machines are larger, use more durable joints and need higher-end edge compute for perception, planning and force feedback.
The strategic claim is that the winning robotics company will own the deployment loop from customer process knowledge to data return and model updates, not merely the most impressive single benchmark.

Open at QbitAI Ask Claude ChatGPT

Why China's gig workers heard welfare talk as an insult

# Pekingnology 4h

A Peking University economist's claim that flexibility is a form of welfare triggered a backlash from Chinese gig workers who heard it as praise for insecurity, exposing a widening elite-worker language gap.

The economist was discussing a standard labor-economics tradeoff: flexible workers may gain control over time while giving up employer-linked benefits and social protection.
Chinese social media largely stripped away that qualification and heard the phrase as saying unstable work, weak benefits and unpredictable income were themselves desirable.
The backlash uses a familiar anti-elite frame, comparing the remark to stories of rulers who cannot understand ordinary people's material hardship.
The episode shows how precarious employment has become a political sensitivity in China: even analytically defensible language can fail when it appears to aestheticize insecurity.
It also reveals a communication problem for policy experts. A framework that treats autonomy and protection as tradeable benefits can sound evasive when workers want protection first.
The missing question is institutional rather than semantic: whether flexible workers can actually obtain portable pensions, insurance and bargaining power. The supplied account does not show that such protections are improving.

Open at Pekingnology Ask Claude ChatGPT

China's rural reform had no master plan, a former official argues

# Pekingnology 23h

A former Chinese rural-policy official argues that the decisive reforms of the 1980s emerged from local experimentation rather than a central blueprint, offering a provocative lesson about how China once made policy.

The account credits farmers, grassroots cadres and local governments with initiating changes such as household contracting, while central leaders were often divided over how to respond.
In this telling, the center's success was not designing every institution but allowing competing approaches to survive long enough for workable ones to emerge.
The reforms transformed both the rural economy and grassroots governance, helping enable private production, hired labor and movement from villages into wider markets.
The political context mattered: a relatively pluralistic policy environment gave local actors room to experiment and gave higher authorities options before they committed nationally.
The argument challenges the common image of Chinese reform as a perfectly sequenced top-down plan and instead describes a feedback system in which practice changed the boundaries of the possible.
Its contemporary implication is sharp: effective policy may require preserving controlled room for local variation, not merely issuing more precise central instructions. The article does not claim that today's political environment offers the same room.

Open at Pekingnology Ask Claude ChatGPT

Chinese media says Qing history is being used against national unity

# SCMP 6h

Chinese state-linked commentary attacks a US-born interpretation of Qing history as theoretical ammunition for separatism, showing how historical scholarship is being pulled into ethnic-policy politics.

The dispute concerns interpretations of the Qing dynasty and whether they weaken the official narrative of a unified Chinese nation, according to the report.
Chinese coverage links the academic argument to online debates, inflammatory rhetoric and claims that historical narratives can encourage territorial division.
This is part of a broader campaign against what Beijing calls historical nihilism: interpretations that challenge or undermine the state's preferred account of national development.
The political target is not only foreign scholarship. Domestic discussion of ethnicity and history is being framed as a security and sovereignty issue, narrowing the acceptable range of interpretation.
For researchers and companies operating in China, the lesson is that historical and cultural topics can become politically sensitive when they touch ethnic boundaries or national legitimacy.

Open at SCMP Ask Claude ChatGPT

China delays Chang'e-7 lunar south-pole mission

# SCMP 4h

China postponed Chang'e-7 after a comprehensive assessment, delaying a complex mission to search for ice near the lunar south pole with an orbiter, lander, rover and hopper.

The mission was expected to launch from Wenchang as early as Monday, but China's human spaceflight authority announced a postponement without a new date in the supplied report.
Chang'e-7 is designed as a multi-vehicle campaign rather than a single lander, combining orbital mapping with surface exploration and a hopping vehicle for difficult terrain.
The target is strategically important because water ice near permanently shadowed polar regions could support future lunar science and resource planning.
The delay is not itself evidence of a technical failure; the authority described it as the result of a comprehensive assessment, leaving the cause unspecified.
China's space program has increasingly presented lunar exploration as a sustained architecture, so schedule discipline and the ability to integrate several vehicles may matter more than any one launch date.
A revised launch window and explanation of what was assessed will determine whether this is routine program management or a sign of a deeper integration problem.

Open at SCMP Ask Claude ChatGPT

Alibaba launches a $10.2 billion share placement for AI investment

# Techmeme 46m

Alibaba is raising about $10.2 billion through a Hong Kong share placement to fund AI development, showing how China's biggest internet firms are financing an infrastructure-heavy AI push.

Alibaba launched an HK$80 billion placement, equivalent to about $10.2 billion, according to the supplied Reuters report; the offering is intended for AI-related development.
The size of the raise makes AI capital expenditure a corporate-finance story rather than merely a model-release story: Alibaba is willing to dilute shareholders to expand capacity and capability.
For Chinese platform companies, AI spending is increasingly being financed at the parent-company level, tying model ambitions to balance-sheet access and public-market confidence.
The placement was reported at a 3.6 percent discount to the prior close, suggesting Alibaba is trading some near-term pricing power for speed and funding certainty.
What matters next is allocation: evidence of new compute, model products, or commercially productive infrastructure would make this more consequential than a broad AI investment pledge.

Open at Techmeme Ask Claude ChatGPT

Chinese developers hunt the maker of the anonymous Ox Alpha model

# AI QbitAI 6h

Chinese developers are comparing tokenizer, video-token and API fingerprints that point toward Z.ai's GLM family, while Google remains another live theory and no maker has confirmed the model.

Also: V2EX

Ox Alpha appeared anonymously on OpenRouter with a context window of about 1.05 million tokens, multimodal input and a maximum output of about 131,000 tokens, according to the report.
Community tests found token-count offsets resembling GLM systems and video-token behavior similar to a GLM multimodal service; these are forensic clues, not proof of shared weights or ownership.
The API error format and parameter handling also looked familiar to some testers, while a claimed residual phrase linking the model to Z.ai is weaker evidence based on screenshots and secondhand reports.
Google remains a competing hypothesis because the model's long context, video input and agent-oriented profile resemble the company's Gemini direction, and employees made ambiguous social-media remarks.
The episode shows a growing form of model provenance research: developers infer backend lineage from preprocessing, token budgets and error surfaces that vendors normally treat as implementation details.

Chinese forum users speculate about the maker of Ox Alpha V2EX

Open at QbitAI Ask Claude ChatGPT

A Beijing robot cafe turns embodied AI into a live service operation

# AI QbitAI 3h

At Beijing's robot conference, Boundless Dynamics ran a real cafe where robots navigated customers, delivered drinks, collected waste and coordinated with human staff instead of following a fixed demo script.

The service robot had to replan around moving visitors, distinguish personal objects from waste and manipulate soft paper, turning a cafe into a compact test of perception, planning and physical interaction.
The workflow spans delivery, table clearing, waste disposal and self-disinfection, so the demonstration tests long-horizon task sequencing rather than one polished motion.
The company says the model has already been piloted in a high-end cafe before appearing at the conference, giving the display more operational credibility than a fenced-off stage performance.
The technical challenge is not simply navigation: it combines dynamic obstacle avoidance, semantic object selection, soft-object handling and coordination with people and other agents.
Boundless Dynamics says it is pursuing a full-stack route from models to robot bodies and components, suggesting that Chinese robotics companies increasingly view sustained deployment data as a competitive moat.

Open at QbitAI Ask Claude ChatGPT

A young Chinese team bets on an end-to-end bipedal robot brain

# AI QbitAI 7h

A newly formed Chinese startup led by doctoral researchers is pursuing a pure end-to-end model for bipedal robots, rejecting the conventional split between task planning and motion control.

The founders argue that separating a robot's high-level brain from its low-level motor system creates information bottlenecks, cascading errors and repeated interface tuning for every new task.
Their proposed model aims to learn perception, physical reasoning and action together, a more aggressive architecture than the modular systems many researchers still view as easier to deploy.
The team focuses on bipedal bodies because human environments, tools and vehicles are designed around human morphology, while bipedal control must manage whole-body dynamics rather than only arm trajectories.
A 29-degree-of-freedom biped requires simultaneous control of balance, contact, inertia and manipulation, making the shift from kinematics to dynamics the central technical hurdle.

Open at QbitAI Ask Claude ChatGPT

The investor who backed Unitree is now chairing another robot startup

# Analysis QbitAI 5h

An early Unitree backer turned a 2 million yuan, roughly $280,000 angel investment into a reported 280 million yuan, about $39 million, return and now chairs Galaxy General.

The investor supplied Unitree's first reported angel funding in 2016, taking 15 percent for 2 million yuan, or roughly $280,000, according to the article.
The article estimates total proceeds of about 280 million yuan, roughly $39 million, after later transfers, a return of more than 140 times; those figures come from the source's reconstruction rather than a fresh filing.
His move from early investor to chair of Galaxy General illustrates how China's robotics ecosystem recycles operators, capital and technical judgment across successive startups.
His previous robotics company, ROOBO, struggled with hardware businesses and later narrowed its focus, giving him experience with the manufacturing and cash-flow risks that many current robotics investors underestimate.

Open at QbitAI Ask Claude ChatGPT

Hong Kong professors are building an embodied-AI startup ecosystem

# Analysis QbitAI 5h

A survey finds at least 16 Hong Kong university professors involved in embodied-AI startups, spanning models, sensors, hands, data, autonomous driving and complete machines.

The professors generally avoid building only a complete humanoid and instead specialize in a technical layer such as world models, tactile sensing, manipulation, data or industrial vehicles.
The pattern links university research to venture creation while preserving a division of labor across the robotics stack, giving Hong Kong a role as an upstream technical node rather than a mass-manufacturing center.
The article presents the professor-student model associated with Li Zexiang and DJI as a local template for turning laboratories into companies.
The ecosystem's strength is breadth and access to research talent; its risk is that many companies still need mainland supply chains, deployment sites and capital to move beyond technical specialization.

Open at QbitAI Ask Claude ChatGPT

A Chinese humanoid robot plays autonomous tennis against a pro

# AI Leiphone 5h

Galaxy General demonstrated autonomous humanoid tennis with live movement, shot selection, doubles coordination and self-recovery, making the company's integrated brain-and-control architecture visible in a real contest.

Also: QbitAI, GeekPark

The robot handled serves, forehands, backhands, court movement and net play against a former Grand Slam player, according to the report; it also worked with a human doubles partner.
AstraBrain combines a high-level decision system, a real-time motion-control system and a bridge between them, rather than treating tactical choice and body control as isolated modules.
The architecture matters because a fast rally requires millisecond-level perception and continuous coordination of footwork, balance, torso rotation and racket contact.
The company trained its tennis system from fragmented ordinary-player motion data and used latent-space constraints to keep actions human-like while adapting to the incoming ball.
Virtual opponents and a large embodied dataset provide scalable practice before real-robot testing, making the demonstration a statement about data generation as much as about control.
The result is still a staged sporting environment, not proof of general household competence; its importance will rise if the same stack transfers to unfamiliar tasks and bodies.

Open at Leiphone Ask Claude ChatGPT

Boundless Dynamics argues that world models must learn from real work

# AI QbitAI 9h

At China's robot conference, Boundless Dynamics promoted a latent-space world model and reinforcement-learning approach that treats industrial and commercial deployments as the training loop.

The company argues that a useful world model should predict how actions change the physical world, not merely generate a visually plausible next frame.
Its proposed route extracts high-level physical and causal structure in latent space, then uses reinforcement learning to connect predictions with action value and feedback.
The company frames factories as places to learn reliable skills and commercial environments as places to learn generalization, an unusually explicit division between controlled practice and open-world testing.

Open at QbitAI Ask Claude ChatGPT

African firms adopt Chinese AI while keeping US options open

# Analysis SCMP 4h

African companies are choosing Chinese AI models where price or flexibility is better, showing that customers may treat the US-China AI rivalry as a procurement problem rather than a loyalty test.

The supplied account uses Nigerian insurtech Curacel as an example of a company expanding its model mix beyond Silicon Valley providers.
Chinese models are attractive in this framing not because customers have chosen Beijing politically, but because they can offer useful performance, lower cost or deployment flexibility.
That creates a different form of overseas competition from a hardware embargo: model access, inference economics, support and openness may matter more than national branding.
For Chinese AI companies, Africa can function as both an export market and a real-world feedback environment, although the supplied report does not quantify deployments or revenue.
The pattern also limits the effectiveness of pressure that asks third countries to choose one AI ecosystem outright; pragmatic buyers can multi-home across providers.

Open at SCMP Ask Claude ChatGPT

Chinese scientists build programmable plant immune receptors

# AI SCMP 14h

Chinese researchers created an AI-guided platform for designing plant immune receptors that can detect specific pathogen proteins and trigger defenses against emerging crop diseases.

The platform is intended to engineer synthetic receptors that recognize proteins from bacteria, viruses and fungi, then activate an immune response inside the plant.
AI-guided protein design is the key architectural move: instead of waiting for naturally occurring resistance genes, researchers can search for receptor designs matched to a chosen pathogen target.
A programmable receptor could shorten the response cycle for crop epidemics, especially when a commercial variety lacks a useful natural defense.
The approach is closer to an adaptable biological detection layer than a conventional pesticide, but the supplied account does not establish field performance, ecological safety or regulatory readiness.
Its broader significance is that China is applying generative and design-oriented AI to agricultural biotechnology, where the bottleneck is not only prediction but reliable expression and deployment in living systems.

Open at SCMP Ask Claude ChatGPT

China's exam reforms raise fears of a weaker historical memory

# Policy SCMP 12h

A Chinese academic warns that students dropping history as an exam subject could leave the public more vulnerable to pseudo-history, as Beijing intensifies its campaign against historical nihilism.

The concern is tied to secondary-school subject choices: if history is treated as less important for university entrance, fewer students may acquire the background needed to evaluate historical claims.
The warning arrives as the state is actively policing narratives that conflict with its official historical line, making history education both a civic subject and a tool of political legitimacy.
That creates a tension between protecting historical literacy and narrowing acceptable interpretation; students may become more vulnerable to misinformation while being exposed to a more tightly managed official narrative.
The debate over bizarre pseudo-historical claims matters because it shows how online misinformation and state narrative control can reinforce each other as competing explanations of the past.
For technology companies and platforms, historical content is not a neutral moderation category in China: questions of ethnicity, sovereignty and national continuity can trigger political scrutiny.

Open at SCMP Ask Claude ChatGPT

Beijing's trees may account for half of ozone-forming chemistry

# Signal Solidot 2h

A study by researchers at Jinan University finds that plant emissions, especially isoprene from willows and poplars, contribute disproportionately to Beijing's summer ozone formation.

The study estimates that plant emissions made up about 10 percent of Beijing's total emissions during May to July 2021, while contributing 52 percent of the chemical precursors that ultimately formed ozone.
About 35 percent of Beijing's trees reportedly emit isoprene, a volatile organic compound that reacts with hydroxyl radicals and nitrogen oxides in sunlight.
The result is a reminder that emissions inventories are not simply a split between cars, factories and clean vegetation; species selection can change the atmospheric chemistry of an urban greening program.
Human activity still produces more volatile organic compounds in total, but the plant compounds appear more effective at driving ozone formation under the studied conditions.
The policy implication is not to remove urban trees wholesale, but to treat species choice and seasonal chemistry as part of air-quality planning.

Open at Solidot Ask Claude ChatGPT

Qiyuan opens reservations for two personal robots

# AI QbitAI 9h

Shangwei New Materials has opened reservations for its Q1 and T1 personal robots, moving Chinese consumer robotics from showroom demonstrations toward deposits and real customer validation.

Also: Leiphone

The company reports 210 million yuan, roughly $30 million, in robot prepayments and says it has opened experience stores in seven major cities, though reservations are not the same as delivered units.
Q1 is a small force-controlled humanoid, while T1 can switch between wheeled-bipedal and quadruped forms; the two products test whether families will pay for novelty, companionship and experimentation before clear household utility exists.

Qiyuan's Q1 and T1 personal robots enter public reservations Leiphone

Open at QbitAI Ask Claude ChatGPT

Kaspersky finds an Android malware campaign targeting car head units

# Signal Solidot 1h

Kaspersky says malware spread through an Android car-head-unit updater and may link to the BADBOX ecosystem, exposing a new route from cheap-device abuse into vehicles.

The malware was distributed through the built-in update mechanism of an Android-based automotive head unit, according to Kaspersky's report.
Kaspersky linked the activity to the MoYu Group, which the supplied text associates with the BADBOX botnet; the company says the vendor was notified and fixed the issue.
The apparent objective was to build a residential-proxy botnet, using infrastructure similar to campaigns against low-cost TV boxes rather than targeting vehicle control systems directly.
The important architectural risk is the updater and embedded Android software supply chain: a compromised head unit can become a network foothold even when the vehicle itself is not being remotely driven.

Open at Solidot Ask Claude ChatGPT

A Chinese scholar says US AI supply-chain plans will backfire

# Policy SCMP 5h

A Chinese political scientist argues that US chip and supply-chain alliances will fail and that restricting access to Chinese models could backfire, revealing Beijing's preferred counter-narrative to containment.

The scholar treats initiatives such as chip and rare-earth alliances as attempts to organize supply chains against China, and predicts that they will not secure durable technological advantage.
He argues that cutting US firms off from low-cost Chinese models would raise costs and weaken innovation rather than contain China's influence; this is an attributed argument, not an established forecast.
His alternative is US-China cooperation to expand global demand, reflecting a Chinese policy discourse that presents growth and interdependence as antidotes to bloc competition.
The article matters less for the prediction itself than for the framing: Chinese elite commentary is positioning export controls as economically self-defeating and cooperation as the more rational path.

Open at SCMP Ask Claude ChatGPT

Washington brings an America First message to APEC talks in China

# Policy SCMP 10h

US officials are pushing reciprocal trade and barriers for American business at APEC meetings hosted by China, weeks before a planned visit by China's leader to the US.

The US State Department frames its APEC role as advancing an America First economic agenda while still presenting US leadership as beneficial to the wider region.
China's role as host gives the talks a dual setting: Washington is using a multilateral forum inside China to press its own trade priorities while Beijing prepares for a high-level bilateral visit.
The contrast suggests managed competition rather than diplomatic disengagement; both sides continue using regional institutions even as their economic language becomes more transactional.
The supplied report does not establish whether the talks produced concrete agreements, so the immediate significance is positioning and agenda-setting.

Open at SCMP Ask Claude ChatGPT

Chinese archaeologists begin a joint field project in Peru

# Culture SCMP 1h

Chinese archaeologists are starting their first joint South American field project in Peru, pairing ancient-site research with Beijing's broader effort to deepen cultural ties in the region.

The project links China's Institute of Archaeology with Peru's Caral Archaeological Zone and the National University of San Marcos.
It will examine sites thousands of years old, making archaeology part of China's outward-facing cultural diplomacy rather than only a domestic heritage program.
The partnership gives Chinese researchers field access and institutional ties in South America while offering Peru another channel for international archaeological cooperation.
Its importance is cumulative: repeated scientific and heritage partnerships can create durable relationships even when they are less visible than infrastructure or trade agreements.

Open at SCMP Ask Claude ChatGPT

An orange lichen may reveal hidden rare-earth deposits

# Signal SCMP 2h

A Chinese research team says a brightly colored lichen can indicate previously unknown rare-earth deposits, potentially enabling cheaper field surveys with drones and visual screening.

The researchers identify the orange-yellow crust on exposed rock as a visible indicator associated with rare-earth mineralization, according to the report.
A biological field marker could narrow the search area before expensive drilling, especially when combined with unmanned aerial vehicles and image analysis.
The finding is an exploration tool, not evidence that every orange lichen marks a deposit; its value will depend on false-positive rates across different geology and climates.
The work fits China's broader interest in rare-earth supply security, but the supplied account does not establish that the method is ready for commercial prospecting.

Open at SCMP Ask Claude ChatGPT

Wang Yi meets Jordan's foreign minister ahead of the king's China visit

# Policy Jiemian 1h

China's foreign minister met Jordan's Ayman Safadi as King Abdullah prepares to visit Beijing, with both sides stressing political trust and expanded cooperation.

Wang Yi framed the visit as an expression of the relationship between the Chinese and Jordanian leaderships and praised Jordan's friendly policy toward China.
The Chinese account emphasizes Jordan's adherence to the one-China principle and its willingness to deepen cooperation across sectors.
Safadi praised China's role in Middle East affairs, reinforcing Beijing's preferred image of itself as a responsible regional actor rather than only an economic partner.

Open at Jiemian Ask Claude ChatGPT

China's AI rise was built through university networks, a WSJ analysis says

# Analysis Techmeme 4h

A Wall Street Journal account highlighted former Tsinghua teacher-student links behind Z.ai and Moonshot AI, framing China's AI progress as an institutional pipeline rather than a sudden breakthrough.

The Techmeme item points to a university ecosystem that trained researchers who later founded or led major Chinese AI companies.
That framing matters because it shifts attention from imitation or isolated startup talent toward long-running academic networks, lab culture and movement between universities and industry.
The article portrays Chinese teams as combining technical ingenuity with aggressive adaptation of ideas pioneered elsewhere, while arguing that they understand commercialization.
As an aggregator item, this is evidence of what US technology media is emphasizing about China, not an independent source for every biographical or company claim.

Open at Techmeme Ask Claude ChatGPT

India rescues two Chinese sailors from a sunken cargo ship

# Policy SCMP 18h

India rescued two Chinese nationals after a Panama-flagged cargo vessel sank off its eastern coast, while search operations continued for other crew members.

The vessel reportedly sank within eight minutes after developing a list, leaving rescuers to search for survivors in difficult conditions.
The Chinese dimension is the consular and human cost: two Chinese seafarers were among the 24-person crew, making the incident relevant to China's overseas commercial workforce.
The supplied account gives no indication that the ship or rescue involved Chinese state or corporate assets, so its broader strategic significance remains limited.

Open at SCMP Ask Claude ChatGPT

A smart ring brand tries to sell emotion instead of metrics

# Culture Leiphone 10h

Chinese brand Shakespeare launched a $210-to-$340 smart ring built around jewelry and AI-generated emotional interpretation, seeking a niche beyond Oura-style health tracking.

The ring combines silver and lab-grown diamonds with health sensors and an AI assistant that turns biometric readings into literary-style emotional commentary, according to the company.
The brand targets urban women aged 20 to 40 and says its ecosystem investment comes from Honor platforms, but the model depends on whether emotional interpretation drives repeat purchases rather than novelty.

Open at Leiphone Ask Claude ChatGPT

China warns citizens in Saudi Arabia about cash rules and scams

# Policy Jiemian 2h

China's embassy in Saudi Arabia warned citizens about a lower cash-declaration threshold and a rise in telecom and informal currency-exchange scams.

Saudi Arabia's revised rules lower the declaration threshold for cash and valuables to 40,000 riyals or equivalent foreign currency, according to the embassy notice.
The warning combines legal compliance with fraud prevention, reflecting how Chinese consular messaging increasingly treats private scams and informal finance as risks to citizens abroad.
The embassy also stresses that Chinese missions will not demand bank details by phone, a standard anti-impersonation message aimed at a growing overseas Chinese population.

Open at Jiemian Ask Claude ChatGPT

China tech roundup: robot games, DeepSeek pricing and Ox Alpha

# GeekPark 16h

A Chinese roundup combines the humanoid robot games with DeepSeek's weekend price cut and the anonymous Ox Alpha model, reflecting a news cycle dominated by AI deployment and spectacle.

The roundup says the robot games drew 666 teams and 2,056 robots across 51 events, while Tiangong Ultra won headline races; the figures are event publicity rather than an industrial benchmark.
DeepSeek's weekend API pricing was set at its low period rate, with the flagship output price falling from 27 yuan to 13.5 yuan per million tokens, roughly $3.80 to $1.90, while Ox Alpha drew attention for its long context and free access.

Open at GeekPark Ask Claude ChatGPT

A Chinese developer proposes using Santa to limit AI-agent data access

# Signal V2EX 1h

A V2EX post suggests using macOS Endpoint Security Framework controls to stop agents from reading sensitive files directly, an unverified but technically interesting supply-chain defense idea.

The poster argues that Santa can distinguish trusted processes from broad file permissions, potentially limiting what an agent can read even when it has extensive system access.
The proposal is a forum experiment, not a validated security result; its practical value depends on whether legitimate helper tools can mediate access without becoming an exploitable bypass.

Open at V2EX Ask Claude ChatGPT

Nanning raises its flood emergency response level

# Signal Jiemian 1h

Nanning raised its Yujiang flood response after water levels approached the warning mark, a routine example of local emergency administration in China.

The city raised the response from level four to level three after the water level reached 71.41 meters and forecasts put the urban river above the 73-meter warning level.
The item shows China's tiered emergency system converting hydrological forecasts into a formal government response, but it provides no evidence yet of wider damage or policy change.

Open at Jiemian Ask Claude ChatGPT

Stephen Chow's Kung Fu football keeps recycling its jokes

# Culture Initium 17h

An Initium review questions how much comedy remains in Stephen Chow's repeated use of familiar gags, but the supplied article body is incomplete.

The headline presents the film as a cultural argument about repetition, audience expectations and the durability of a familiar comic style.
The source payload contains website code rather than the review, so no further claims about the film, reception or critic's conclusion can be safely extracted.

Open at Initium Ask Claude ChatGPT

Chinese ChatGPT subscribers complain about being routed to a weaker model

# Culture V2EX 3h

A V2EX thread says some Chinese Plus subscribers were routed to a smaller ChatGPT model and calls for domestic competition, capturing user frustration rather than verified service behavior.

Commenters suspect routing, account risk controls or heavily shared proxy addresses, but the thread does not establish which explanation is correct.
The mood is less nationalist than competitive: the poster says a credible Chinese alternative would be welcome mainly to reduce dependence on a perceived foreign monopoly.

Open at V2EX Ask Claude ChatGPT

A Singapore migrant searches for the taste of Hong Kong

# Culture Initium 17h

An Initium feature appears to examine grocery shopping in Singapore and the difficulty of preserving Hong Kong taste abroad, but the supplied article body is incomplete.

The headline frames food shopping as a problem of migration and cultural translation rather than simple product availability.
Because the source payload contains website code instead of the article text, the feature's reporting, argument and evidence cannot be assessed from the supplied material.

Open at Initium Ask Claude ChatGPT

ChinAI	403	0/0		HTTPError: 403 Client Error: Forbidden for url: https://chinai.substack.com/feed
ChinaTalk	200	0/20
Pekingnology	200	2/20
Ginger River Review	200	0/20
Sinocism	200	0/20
QbitAI	200	10/10
36Kr	200	0/30
Solidot	200	5/17
Leiphone	200	7/20
GeekPark	200	3/30
IT Home	200	12/60	+40 over cap
SCMP	200	10/50	+1 over cap
Initium	200	2/15
Jiemian	200	8/30	+21 over cap
China Digital Times	200	0/15
The Wire China	200	0/50
Techmeme	200	3/15
Hacker News	200	1/20
V2EX	200	8/50	+30 over cap

Chang'e-7 slips a year, and only Thailand said why

The moon window closes for the year

A labour economist becomes Marie Antoinette

Nobody will admit to Ox Alpha

Robot records fall, buyers ask about payback

History gets treated as a security problem

Threads we are pulling