China Desk — 17 Aug 2026, 21:43 UTC

Chinese robotics enters the harder phase: finishing the job

# Leiphone 18h

A factory trial combines mobile robots, a robot dog, and an AGV under a task-level runtime that monitors outcomes and recovers from failures instead of merely issuing commands.

The deployment splits one automotive logistics workflow across multiple machines: sorting and loading, box handling, and transport and dispatch.
The core software loop observes the environment, schedules skills, checks whether the physical result matches the intended result, and either recovers or continues.
The architecture separates a high-level agent harness from an on-site runtime and device-level controllers, keeping open-ended models away from safety-critical real-time actuation.
This addresses a common industrial failure mode: a successful device acknowledgment is not the same as a completed production task.
The company argues that the most valuable data is real-world test and recovery data, not simply more training examples, because the deployment gap appears outside the training distribution.
The deeper industrial implication is process redesign: factories may need workflows organized around robot-stable task units rather than human-shaped workstations.
The sector’s financing surge—reported at 935 billion yuan in the first half, roughly $132 billion—makes this execution gap economically urgent, not academic.

Open at Leiphone Ask Claude ChatGPT

DeepSeek makes the entire agent runtime pluggable

# Leiphone 12h

DeepSeek’s developer preview treats the model adapter, tool registry, session log, and agent loop as replaceable plugins, using reversible effects and dependency tracking to support self-evolving runtimes.

The architecture removes the usual fixed core and separates components such as model adapters, tools, sessions, storage, and the agent loop into replaceable modules.
Its Cordis framework records side effects so listeners, timers, resources, and other changes can be reversed when a component is removed.
Reactive dependency handling lets components activate, deactivate, or remain unchanged as providers appear, disappear, or change.
This gives self-modifying agents a possible rollback mechanism: runtime changes can be experimented with without permanently corrupting the host system.
The design is more consequential than another coding agent because it treats the control plane itself as mutable software.
The open question is whether dynamic composition can remain secure and debuggable when an agent is allowed to modify its own operating environment.

Open at Leiphone Ask Claude ChatGPT

China Telecom backs a physical-AI data platform

# Leiphone 17h

Maniformer raised several hundred million yuan, roughly $30 million to $60 million, to scale real-world robot-data collection, governance, and evaluation across factories, homes, and logistics sites.

The company says its wearable and body-independent collection products capture environment and hand-operation data with millimeter-level trajectory accuracy.
Its software links preprocessing, spatial reconstruction, multimodal annotation, and quality evaluation, reportedly improving processing efficiency more than tenfold.
The platform claims deployments in more than 20 Chinese cities and over five overseas nodes, with data supplied to embodied-model projects.
The investment by China Telecom and a Shanghai industrial group connects cloud, network, physical sites, and data standards in a distinctly infrastructure-led model.
The important bottleneck is not raw hours but whether collected interactions are clean, diverse, legally usable, and tied to measurable task-success gains.
The company reports some closed-loop improvements of 5 to 10 percent per data cycle, but the metric and baseline are not described.

Open at Leiphone Ask Claude ChatGPT

China’s anti-corruption campaign reaches the securities regulator

# The Wire China 1d

A probe into an influential Chinese securities regulator has unsettled the financial community, signaling that anti-corruption enforcement continues to reach institutions central to capital-market reform.

The supplied excerpt identifies an investigation into an influential securities regulator but gives no name, allegation, or formal finding.
The strategic significance is institutional: scrutiny at a market regulator can affect investor confidence, personnel continuity, and the pace of financial-policy implementation.
China’s anti-corruption campaigns often function both as discipline and as a way to reassert central control over powerful bureaucracies.
The missing facts matter greatly; without the regulator’s identity and the scope of the probe, the market impact cannot be assessed precisely.

Open at The Wire China Ask Claude ChatGPT

China’s six-network plan puts power and compute at the center

# Jiemian 7h

China’s State Council is advancing a six-network infrastructure program estimated at about 5 trillion yuan, or roughly $704 billion, with smart power grids the largest component.

The six networks cover water, modern power, compute, next-generation communications, urban underground utilities, and logistics.
A cited securities estimate puts total 2026 investment at about 5 trillion yuan (roughly $704 billion) and the fifteen-year plan at 23 to 25 trillion yuan (about $3.2 to $3.5 trillion).
Modern power-grid investment alone is projected to exceed 5 trillion yuan during the plan period, more than 80 percent above the previous period.
The shift is from traditional infrastructure expansion toward integrating renewable generation, flexible dispatch, compute demand, and industrial electrification.
This is a state-industrial strategy: compute infrastructure is being planned alongside the energy and logistics systems needed to make it economically usable.
The main uncertainty is financing and execution, especially whether social capital can be mobilized without creating another round of underused local projects.

Open at Jiemian Ask Claude ChatGPT

China’s EV industry grew through local deals more than central planning

# Analysis ChinaTalk 11h

A new analysis argues that China’s EV champions emerged from alliances among local governments, private capital, foreign technology, and entrepreneurs, producing global strength alongside severe overcapacity.

The account traces the industry from roughly 130 small carmakers in the 1990s to globally competitive firms such as Chery and Geely.
Its central revision is that local officials and private investors often worked around state-owned-enterprise dominance and central regulation rather than simply executing a national master plan.
Local political connections, financing, and foreign technology helped firms survive long enough to build capabilities and scale.
The same decentralized competition produced excess capacity, making China’s EV success both an industrial achievement and a structural problem.
The analysis matters beyond autos because similar local-government and private-capital arrangements may shape robotics and AI.

Open at ChinaTalk Ask Claude ChatGPT

Two humanoids play a complete autonomous table-tennis game

# AI QbitAI 12h

A Hong Kong University team’s robots completed an 11-point game without teleoperation, highlighting the hard systems problem of linking perception, prediction, planning, and whole-body control.

The SMASH 2.0 stack extends the earlier system to short and long balls and adds autonomous serving, both necessary for a genuine match.
The team says the robots use the same system and preset strategies but cannot yet learn an opponent’s style during play.
Human motion-capture data supplies much of the training signal, while successful and failed matches are retained for future real-robot reinforcement learning.
Table tennis is a demanding benchmark because the robot must observe a moving object, decide where to move, and coordinate balance and contact in one loop.
The demonstration is a stage in research, not evidence of competitive human-level play; spin handling and repeatability are the next meaningful tests.

Two humanoids complete an autonomous 11-point table-tennis game QbitAI

Open at QbitAI Ask Claude ChatGPT

HelixWorld adds spatial audio to interactive world models

# AI QbitAI 14h

A Hong Kong-linked research team claims real-time interactive worlds at 24 frames per second with 48 kHz stereo audio, generated jointly so sound responds to movement and scene geometry.

HelixWorld combines video and audio in one Transformer rather than adding a soundtrack after video generation.
The system uses first-person footage and game-engine data to learn how movement, occlusion, distance, materials, and listener orientation affect sound.
The reported output is 24 frames per second with 48 kHz stereo audio, but the supplied text does not provide latency or hardware requirements.
Joint generation matters because an interactive world must update the acoustic scene when the user turns, walks, or crosses a material boundary.
The planned open release could make the approach easier to test, but reproducibility will depend on the quality and licensing of its multimodal data.

Open at QbitAI Ask Claude ChatGPT

China’s robot industry faces a reality check at its flagship expo

# Analysis GeekPark 18h

Ahead of the World Robot Conference, Chinese coverage argues that valuation and demos are outrunning orders, and that the decisive tests are endurance, data, components, and non-humanoid forms.

Unitree’s planned listing is cited at an implied market value of about 61 billion yuan, roughly $8.6 billion, while its 2025 humanoid shipments were only about 5,500 units.
The industry is growing rapidly, but the source says no core layer—data, bodies, or algorithms—has reached technological convergence.
The article asks whether wheeled or specialized robots may commercialize faster than humanoids when the task does not require a human shape.
It also treats data collection hardware, force sensing, motors, and dexterous hands as more meaningful indicators than stage demonstrations.
The central commercial metric is not whether a robot can perform one impressive action, but whether it can work for hours without intervention at a viable cost.
This is a valuable Chinese industry self-critique because it places the sector’s spectacular financing and public attention against thin current output.

Open at GeekPark Ask Claude ChatGPT

Chinamaxxing may be as much Western fatigue as China fascination

# Culture SCMP 7h

Chinese coverage treats viral interest in White Rabbit candy and other symbols as a debate over whether Gen Z is embracing China or simply tiring of Western cultural defaults.

A White Rabbit package post reportedly drew more than six million views and 260,000 likes, turning a familiar Chinese confection into a Western design object.
The article questions whether this is genuine curiosity about China or a broader search for alternatives to dominant Western brands and aesthetics.
The framing matters because Chinese cultural exports often gain traction first as visual or nostalgic objects, before audiences engage with their political context.
The trend is not evidence of a durable geopolitical realignment; it is better read as a measure of online taste, novelty, and cultural fatigue.
If the attention converts into sustained consumption and travel interest, it becomes soft-power evidence; otherwise it is a short-lived social-media cycle.

Open at SCMP Ask Claude ChatGPT

A Beijing professor bets on latent-space models for robots

# Analysis Leiphone 18h

BeingBeyond has collected more than 500,000 hours of first-person human video and is pursuing a latent world-action model that predicts robot actions and world state without generating pixels.

The approach represents a sharp trade-off: it claims roughly one percent of the training cost of video-generation routes and latency suitable for control, but produces less visually impressive demos.
The model combines vision, touch, action, and predicted future state in a shared latent space, aiming to learn physical interaction rather than merely render plausible frames.
The founder argues that embodied AI lacks a settled equivalent of the Transformer and next-token prediction, so the industry may be scaling before it knows the right abstraction.
The company’s large first-person video collection reflects a bet that human motion and interaction footage is the most scalable source of embodied data.
This is a useful counterpoint to capital-heavy world-model narratives: lower-dimensional prediction may win if control latency and data efficiency dominate visual fidelity.
The major unknown is transfer from human video to robot bodies, sensors, and action spaces; the source provides no broad task-success results.

Open at Leiphone Ask Claude ChatGPT

China’s space program has a failure-tolerance problem

# Analysis SCMP 13m

Chinese coverage reports only a flight anomaly after a Long March-7A breakup, while trade reports quickly blamed an engine; the gap exposes how limited public accountability remains.

The source says CASC and Xinhua did not identify a cause, while trade reporting rapidly converged on an engine fault that remained unconfirmed.
The strategic issue is not one failed launch but how quickly an opaque system manufactures consensus around failures without releasing evidence.
China’s launch cadence depends on rapid iteration, yet a culture that treats uncertainty as reputational risk can discourage engineers from surfacing inconvenient causes.
The episode is a useful test of whether China’s commercial-space expansion will adopt more transparent failure analysis or preserve a state-enterprise information model.
The story matters more if the investigation eventually publishes engineering findings; without that, the engine theory remains speculation rather than diagnosis.

Open at SCMP Ask Claude ChatGPT

Pentagon orders 30 universities to audit China research ties

# Policy SCMP 6h

The Pentagon is conditioning future funding on audits of foreign research relationships, including Chinese partners; Chinese coverage presents this as a widening institutional barrier to collaboration.

The order reportedly covers 30 US universities and requires reviews of academic, financial, and research ties to foreign entities of concern.
The inclusion of former Confucius Institute-linked organizations broadens scrutiny beyond classified work to the institutional networks around research partnerships.
For Chinese universities and companies, the practical effect is higher diligence cost and more uncertainty around ordinary collaboration, even when the work is not security-sensitive.
The policy converts geopolitical suspicion into a funding-eligibility mechanism, giving universities a strong incentive to over-comply.
Its importance will turn on how broadly the Pentagon defines sensitive relationships and whether other agencies adopt the same model.

Open at SCMP Ask Claude ChatGPT

A Chinese model router supports GLM-5.3 on release day

# AI QbitAI 13h

PhanRouter added Zhipu’s GLM-5.3 on launch day through a unified API, illustrating how Chinese infrastructure vendors are competing on model availability, chip compatibility, and enterprise switching costs.

The service claims a unified endpoint, routing, caching, and lower pricing than direct access, allowing users to change models without changing integration code.
The source says the model has 743 billion total parameters and relies on post-training reinforcement learning, but the performance figures are vendor claims.
PhanRouter reports support for Huawei Ascend, Cambricon, and more than 220,000 models, indicating a domestic compatibility layer around fragmented hardware and model supply.
Same-day availability can reduce adoption friction for enterprises that want new models without rebuilding deployment pipelines.
The report is company-sponsored, so service-level and benchmark claims need independent validation.

Open at QbitAI Ask Claude ChatGPT

A few editors control most Chinese Wikipedia deletions

# Culture Solidot 8h

A study of 47,030 deletion debates from 2020 to 2024 finds decision power concentrated among a shrinking group of Chinese Wikipedia administrators, with one editor handling 89.9 percent of 2023 cases.

The study finds that six administrators made 85 to 100 percent of final decisions, despite hundreds of editors participating in debates each year.
One administrator reportedly handled 6,312 cases in 2023, or 89.9 percent of the total, showing how formal openness can coexist with extreme operational concentration.
The source says traditional-character editors outnumber simplified-character editors by 1.2 to 1.8 times even though the simplified-character user base is much larger.
Notability was the most common deletion rationale, and more than a quarter of deletions reportedly lacked a clear explanation.
The result resembles a governance bottleneck: participation is broad, but authority and personal preferences are concentrated in a few maintainers.
This is not state censorship, but it is a useful case study in how volunteer institutions can centralize power through attrition and workload.

Open at Solidot Ask Claude ChatGPT

The US may be arguing about the wrong AI bottleneck

# Analysis SCMP 9h

Chinese coverage emphasizes the shock and market reaction to Kimi K3 and Qwen3.8 while questioning unaudited claims, framing the US debate as a failure to understand China’s model-development path.

The article says global chip stocks lost about $3 trillion in market value after the Chinese releases, before complete benchmark and licensing evidence was public.
Its central argument is that the strategic issue is not one model’s score but whether China can build a self-reinforcing technology path under export restrictions.
The Chinese framing treats model releases as evidence of industrial adaptation and US analytical overreaction, while also warning that performance claims need auditable model cards and licenses.
The gap between market response and public evidence is itself important for executives planning around rapidly changing AI supply assumptions.
The story is analysis, not independent validation of the models or the market-loss attribution, so its strongest value is the framing of China’s technological autonomy debate.

Open at SCMP Ask Claude ChatGPT

A Chinese startup tests full-body intelligence with a racing humanoid

# AI Leiphone 12h

Symbiosis Robotics used a bipedal robot to drive a go-kart, combining visual perception, balance, steering, and pedal control as a stress test for end-to-end physical intelligence.

The demonstration requires simultaneous control of hands, feet, balance, and environmental response rather than a single preprogrammed motion.
The company describes an end-to-end vision-to-action foundation model intended to reduce information loss between perception, planning, and control modules.
That architecture can improve cross-task learning, but it also couples model errors directly to hardware and requires large volumes of real-robot data.
The source says the team’s background spans vision-language-action models, whole-body control, force-position control, and cross-platform learning.
A video proves only that one constrained task worked; the important follow-up is reproducible success rates, model details, and performance outside the track.

Open at Leiphone Ask Claude ChatGPT

Waymo has imported 3,200 Chinese Zeekr vehicles

# Policy Solidot 13h

Waymo reportedly imported 3,200 Zeekr vehicles despite a 127.5 percent US tariff because the Chinese platform remains cheaper than its previous vehicle base, exposing a gap between trade policy and autonomous-fleet economics.

The report estimates a Zeekr vehicle at about $38,000 before tariffs and roughly $86,500 after the tariff, still below the more than $200,000 cost of a modified Jaguar I-Pace.
Waymo has reportedly imported 2,600 of the vehicles this year, making the choice a fleet-scale sourcing decision rather than a test purchase.
The case shows that tariffs can raise costs substantially without eliminating Chinese vehicles’ systems-level cost advantage.
It also reveals how autonomous fleets value a vehicle platform: sensor integration, cabin layout, supply continuity, and modification cost may outweigh origin-based procurement rules.
The policy question is whether future restrictions will target finished vehicles, autonomous hardware, software, or all three.

Open at Solidot Ask Claude ChatGPT

Beijing is building an autonomous theory of international relations

# Policy Pekingnology 9h

Chinese academic institutions are being asked to develop an independent knowledge system, including an international-relations theory grounded in Chinese concepts and diplomatic experience.

The agenda follows Xi Jinping’s call for disciplines with Chinese characteristics and is being organized through Party and university institutions.
The cited essay surveys relational theory, moral realism, world-order theory, and symbiosis theory as alternatives or complements to Western international-relations frameworks.
This is not merely an academic branding exercise: it links scholarship to the state’s broader effort to define policy concepts on its own intellectual terms.
The project may produce useful analytical tools for understanding Chinese diplomacy, but its proximity to official ideological priorities can narrow what counts as acceptable theory.
The important signal for foreign institutions is that Chinese policy language will increasingly be backed by domestic theoretical frameworks rather than translated Western categories.

Open at Pekingnology Ask Claude ChatGPT

Qwen models reportedly passed three billion downloads

# AI Solidot 5h

A Hugging Face report puts Alibaba’s Qwen downloads above three billion in six months, with far more derivatives than rival open-weight families; licensing limits complicate the meaning of openness.

The reported download total is three billion for Qwen, compared with 418 million for Google’s open-weight models and 227 million for Meta’s, according to the source.
The ecosystem count is also notable: 151,448 derivative models reportedly build on Qwen, versus 82,506 for Google.
The source says many large Chinese models use permissive licenses but retain noncommercial or revenue-sharing restrictions, so download volume does not equal unrestricted adoption.
Qwen’s position suggests that distribution, derivative experimentation, and low-friction access may matter as much as benchmark leadership in building model influence.
The number is useful but should be read as a platform metric from a third-party report, not proof that all downloads represent active production use.

Open at Solidot Ask Claude ChatGPT

Chinese researchers develop a glucose-sensing diabetes probiotic

# Signal SCMP 9h

A Shanghai team engineered a bacterium that senses high blood sugar and releases a glucose-lowering hormone, matching Ozempic in animal tests but still far from human approval.

The reported design turns a probiotic into a feedback-controlled delivery system: it detects elevated glucose and releases a therapeutic hormone.
The source says animal results were comparable to Ozempic, but animal equivalence does not establish safety, dosing, or efficacy in humans.
The team has filed patents and is scaling production, suggesting an attempt to move quickly from academic result to regulated product development.
The approach could reduce injection burden if it survives clinical testing, but engineered organisms introduce containment, persistence, and manufacturing questions.
The next meaningful milestone is a transparent clinical protocol, not a US shelf date.

Open at SCMP Ask Claude ChatGPT

Why the Arctic sea route matters to China and Russia

# Analysis SCMP 6h

China’s growing use of Russia’s Northern Sea Route is making the Arctic a strategic shipping option, but climate risk and seasonal ice still limit it as a Suez substitute.

Also: Solidot

The route spans roughly 5,600 kilometers along Russia’s Arctic coast and became more attractive after Red Sea attacks disrupted Suez traffic.
Chinese container ships have used the route for several years, making it more than a theoretical corridor for Beijing’s Europe trade.
The route’s appeal is lower distance and reduced exposure to Middle East chokepoints, but access depends on ice conditions, specialized vessels, and Russian control.
More traffic could deepen China-Russia logistics ties while increasing environmental and geopolitical scrutiny in the Arctic.
The commercial case strengthens if seasonal reliability improves; otherwise the route remains a hedge rather than a replacement for Suez.

A Chinese container ship opens another Arctic route to Europe Solidot

Open at SCMP Ask Claude ChatGPT

Can China build its own technology path under US restrictions?

# Policy SCMP 11h

A Chinese analyst argues that technological autonomy, illustrated by Huawei’s Atlas computing systems, may determine the AI contest more than access to any single foreign chip.

The argument is presented in a book preface about Huawei’s Atlas 950 SuperPod computing system, linking hardware integration to a broader national technology strategy.
The phrase technological path implies more than import substitution: it includes domestic architecture, software, supply chains, and institutions that can evolve together.
This framing is consistent with China’s policy push toward self-reliance under sustained export controls.
The source offers an analyst’s thesis rather than new performance data, so its value is in revealing how Chinese policy circles conceptualize the contest.
For executives, the practical question is whether Chinese systems can become good enough through integration and scale even when they lag at individual components.

Open at SCMP Ask Claude ChatGPT

RoboScience unveils a wheeled humanoid warehouse robot

# AI Leiphone 13h

RoboScience’s REX G1 combines a mobile base, lifting torso, two arms, and an object-centered vision-language-object-action model for warehouse and factory work.

The robot has 22 degrees of freedom, can carry 10 kilograms with both arms, and claims 0.1 millimeter repeatability.
Its adjustable body is designed for narrow aisles, with on-board compute intended to keep perception, planning, and control local when factory networks are unreliable.
The Visics architecture separates a physical world model that predicts object motion from an operation model that converts those predictions into control signals.
This is a more credible industrial direction than a humanoid optimized only for walking, but the supplied claims are company figures without independent deployment data.
The real test is sustained task completion, recovery from exceptions, and total cost per completed warehouse operation.

Open at Leiphone Ask Claude ChatGPT

Taiwan and a US startup will develop underwater drones

# Policy SCMP 8h

Taiwan’s government weapons institute signed with US startup Vatn Systems to develop autonomous underwater vehicles, an asymmetric-defense move shaped by pressure from Beijing.

The agreement runs through Taiwan’s National Chung-Shan Institute of Science and Technology, the island’s state-funded weapons developer.
Autonomous underwater vehicles offer a lower-cost, distributed capability that is harder to neutralize than a small number of large platforms.
The partnership also shows Taiwan using foreign startup technology alongside domestic defense institutions rather than relying only on traditional procurement channels.
For Beijing, the important signal is the widening of Taiwan’s unmanned systems network into the maritime domain.
The military value will depend on autonomy, communications resilience, and deployment at scale, none of which the supplied excerpt establishes.

Open at SCMP Ask Claude ChatGPT

Pinduoduo’s village delivery network sends garlic straight to market

# Culture Leiphone 18h

A rural logistics pilot lets village merchants load garlic onto returning delivery trucks, cutting a seller’s reported shipping cost from about 7 yuan to 3 yuan per parcel and turning logistics into a local economic platform.

The model uses delivery vehicles that bring parcels into villages and collect agricultural shipments on the return trip, reducing empty backhauls.
A local seller says the change cut monthly transport costs by more than 3,000 yuan and reduced the parcel rate from about 7 yuan to roughly 3 yuan.
The village hub reportedly handles 10,000 to 12,000 inbound parcels and about 2,000 outbound parcels daily, most of them garlic.
This is a concrete example of platform economics solving a rural coordination problem that individual couriers and small sellers could not finance alone.
The wider effect is labor and market access: farmers, packers, livestreamers, and village shops can participate without moving closer to county logistics centers.
The model also increases platform dependence, since the logistics subsidy and seller demand are tied to one marketplace.

Open at Leiphone Ask Claude ChatGPT

Ecuador’s president courts China without threatening US ties

# Policy SCMP 45m

Ecuador’s China visit seeks shrimp access and energy and mining investment; Beijing is using trade and state-to-state diplomacy to deepen ties without forcing Quito into an overt US choice.

Ecuadorian President Daniel Noboa is seeking the reinstatement of 14 shrimp processors and new Chinese investment in energy and mining, according to the source.
His meetings with China’s top leadership show how commercial disputes and investment are handled through high-level political channels.
The visit reflects Beijing’s preference for incremental economic leverage: market access and infrastructure finance can expand influence without requiring a formal geopolitical realignment.
For US companies, the relevant signal is competition for Ecuadorian commodities and projects, not an immediate alliance shift.
The outcome will depend on whether processor access and investment commitments materialize rather than on the symbolism of the visit.

Open at SCMP Ask Claude ChatGPT

Selling Chinese electronics in Indonesia is not an easy win

# Signal 36Kr 18h

A Chinese electronics distributor leads Indonesia’s accessory market through 40,000 offline dealers, but currency losses and rising platform fees show why overseas scale can still produce weak economics.

The company reported 1.217 billion yuan of 2025 revenue, roughly $171 million, with 94 percent from Indonesia and 93.4 percent from its own brands.
Its 2026 first-half revenue rose 22 percent, but operating profit was nearly flat as selling costs and third-party platform commissions climbed.
A reported 179.2 million yuan currency loss, roughly $25 million, consumed about 72 percent of first-half net profit as the Indonesian rupiah weakened against the yuan.
The company’s 2.2 percent share makes it the leading accessory retailer but also shows how fragmented the Indonesian market remains.
The offline network is the strategic asset: 42 local branches, nine warehouses, and tens of thousands of small distributors create reach that marketplaces alone cannot reproduce.
The case illustrates a broader Chinese globalization problem: moving factories or products abroad is easier than building resilient local pricing, treasury, software, and channel operations.

Open at 36Kr Ask Claude ChatGPT

Doubao lets users control a computer remotely from a phone

# AI Leiphone 12h

ByteDance’s Doubao now lets a phone direct an authorized desktop agent to read files and operate software, extending Chinese consumer AI from chat into persistent local-device execution.

The system links a phone and one or more authorized computers, with the desktop reading local files and returning progress and results.
This is an agent architecture with real local privileges, so identity binding, permission scope, and revocation matter more than the convenience demo.
The use cases described include file conversion, continuing data analysis, and combining files across home and work computers.
The feature signals Chinese consumer AI products moving toward remote task execution rather than merely generating answers.
The source gives no information about audit logs, approval gates, or protections against a compromised phone or malicious instruction.

Open at Leiphone Ask Claude ChatGPT

Why China’s virtual singer Luo Tianyi still has staying power

# Culture 36Kr 20h

Fourteen years after launch, Luo Tianyi remains a cross-generational Chinese pop-culture property because a creator community, voice technology, and long-term operations built her identity together.

The character’s anniversary concerts now attract students, parents, and older fans, showing a reach beyond the original niche audience.
Unlike AI companions or virtual streamers, the virtual singer began as a voice-and-composition tool whose personality was built through user-generated songs and fan circulation.
The article credits the ecosystem of independent producers with supplying recurring music, memes, food associations, and historical or regional imagery.
Commercial partnerships with tourism, theme parks, and ceramics show how a virtual character can function as a cultural licensing platform rather than just a media avatar.
The strategic lesson for AI companions is that durable attachment may depend less on conversational novelty than on a deep creator ecosystem and accumulated shared memory.
Its challenge is avoiding stagnation: newer AI characters may offer more interaction, but they do not yet possess the same fourteen-year cultural archive.

Open at 36Kr Ask Claude ChatGPT

Chinese brain-interface company raises more than $14 million

# Signal 36Kr 21h

Weiling Medical raised more than 100 million yuan, roughly $14 million, to validate an implanted brain-computer interface for neurological rehabilitation rather than consumer control.

The company is pursuing an implanted platform with flexible cortical electrodes, an external controller, neural stimulation, and assistive devices.
Its thin electrodes sit on the cortical surface rather than penetrating brain tissue, according to the source, allowing broader signal collection while targeting specific clinical circuits.
The company says an early patient with severe brainstem injury regained limited arm and wrist movement after implantation and several weeks of training, but this is an individual case, not clinical proof.
The founder’s strategy explicitly prioritizes serious medical indications before consumer applications, aligning development with a clearer clinical need and regulatory pathway.
The commercial challenge is not only decoding signals but demonstrating durable functional benefit, surgical safety, and repeatable rehabilitation protocols across patients.
The funding reflects China’s growing medical BCI ecosystem, but the field remains at the clinical-validation stage.

Open at 36Kr Ask Claude ChatGPT

Chinese manufacturers move from exporting products to exporting factories

# Analysis 36Kr 8h

Chinese companies are shifting overseas from selling finished goods to relocating production, but local networks, software, labor, and supply-chain coordination are proving harder to export than equipment.

The article describes tariff pressure pushing production toward Vietnam, Thailand, Mexico, and the Middle East rather than relying only on cross-border shipments.
The difficult layer is operational: domestic manufacturing systems often meet weaker local networks, unfamiliar workflows, language problems, and fragmented reporting.
Successful projects are using cloud-edge-device architectures and AI-assisted planning, but the source presents them as exceptions built through costly trial and error.
The strategic implication is that globalization now requires a distributed operating system for plants, data, suppliers, and brands, not just a new factory address.
Small and midsize firms face the greatest gap because they lack the capital to recreate China’s dense supplier and management infrastructure abroad.

Open at 36Kr Ask Claude ChatGPT

A fill-in-the-blank question becomes a debate about how to live

# Culture 36Kr 12h

A fast-food promotion in China turned a simple savings-and-travel sentence into a social-media argument over delayed gratification, revealing young consumers’ competing responses to insecurity and present-day desire.

The promotion accepted multiple answers, but users continued arguing online over whether saving must come before travel or can happen alongside it.
The three recurring positions were save first, do both simultaneously, and reject the premise in favor of immediate enjoyment or saving without travel.
The article reads the debate as mood data rather than a generational rule: the preferred answer changes with income, work pressure, and current financial anxiety.
That makes the episode more revealing than the promotion itself; consumer culture is using a trivial game to negotiate uncertainty, ambition, and the value of time.
For brands, the lesson is that interactive campaigns can become identity tests when they touch household finances and life-stage pressure.

Open at 36Kr Ask Claude ChatGPT

Seven recent stories show US-China ties narrowing around investment and tech

# Analysis SCMP 7h

The roundup highlights stalled bilateral investment mechanisms, technology restrictions, and disputes involving Chinese firms; its value is the pattern rather than any single new event.

The source presents the US-China investment board as unlikely to deliver a major result before an expected leader-level meeting.
Its selection of stories places capital flows, research and technology controls, and company-level disputes in one increasingly securitized relationship.
The roundup is analysis rather than a single primary report, so individual claims should be checked against the underlying stories.
The broader signal is that even apparently commercial channels now depend on political confidence and regulatory clearance.

Open at SCMP Ask Claude ChatGPT

ChinAI	403	0/0		HTTPError: 403 Client Error: Forbidden for url: https://chinai.substack.com/feed
ChinaTalk	200	1/20
Pekingnology	200	1/20
Ginger River Review	200	0/20
Sinocism	200	0/20
QbitAI	200	5/10
36Kr	200	9/30
Solidot	200	11/18
Leiphone	200	20/20
GeekPark	200	4/30
IT Home	200	12/60	+34 over cap
SCMP	200	10/50	+5 over cap
Initium	200	2/15
Jiemian	200	8/30	+13 over cap
China Digital Times	200	0/15
The Wire China	200	5/50
Techmeme	200	1/15
Hacker News	200	1/20
V2EX	200	8/50	+33 over cap

Xi took the nineties back this morning; the July data explains why he needed them

Forty minutes for Jiang, then a warning

Unitree lists Wednesday at 219 times earnings

DeepSeek starts charging peak rates today

Video models converge, only one opens the weights

Want Want's sugar problem, and a fill-in-the-blank

Waymo pays 127% and still saves money

Threads we are pulling