Pogbagram

News worth watching even on the weekend

Apple is reportedly using Gemini to train its on-device AI, causing some Siri requests to be redirected to Google Cloud.

According to The Information, Apple plans to fully leverage its edge AI computing power at its Worldwide Developers Conference (WWDC) next month, while also introducing NVIDIA's privacy technology to ensure the security of some cloud-based AI tasks.

Sources familiar with the matter revealed that Apple is using Google's Gemini large language model to train lightweight models that can run locally (i.e., model distillation).

To advance the miniaturization of its models, Apple is seeking to acquire relevant startups, with Liquid AI, an edge AI company based in Cambridge, Massachusetts, being one of its potential acquisition targets.

To address complex cloud-based needs that edge devices cannot handle, Apple has approved the use of NVIDIA's Confidential Compute privacy technology in Google Cloud environments.

This technology ensures the security of cloud-based AI requests by encrypting data and models at the hardware level. This also means that while Apple maintains its "Private Cloud Compute" brand promise, the underlying processing still relies on Nvidia GPUs and Google's cloud infrastructure.

Seres responds to Tesla's FSD entry into China

According to IT Home, at the recent 4th Future Car Pioneers Conference, Seres Automotive Chairman Zhang Zhengping shared her views on the comparison between Huawei's Qiankun Intelligent Driving and Tesla's FSD.

He stated that the ultimate judgment on which of the two intelligent driving systems is superior should be left to the market to objectively evaluate. When discussing specific technological approaches, Zhang Zhengping pointed out:

Ultimately, the market will be the best judge of which intelligent driving system is the most effective. The new generation Wenjie M9 is equipped with six LiDAR sensors, making it comparable to FSD in this aspect. Furthermore, while FSD simulates human visual driving logic, Huawei Qiankun Intelligent Driving achieves a state where "it's not just what the human eye can see."

It is understood that Tesla's FSD, starting from version V12, adopts a pure vision solution, relying on 8 cameras and an end-to-end neural network architecture, with the core being the simulation of human visual logic.

Anthropic prioritizes values-based interviews in its hiring process.

Bloomberg Businessweek reported on Anthropic's hiring process. The report stated that even after its rapid expansion, the AI ​​company still places "whether the candidate aligns with the company's mission" at the core of its interview process; applicants typically undergo multiple rounds of skills assessments and interviews, with most exercises requiring completion without the use of AI .

The report mentions that Anthropic's culture interviews ask candidates about their values, worldview, and how they view the risks of AI. Several candidates and career coaches described this round of interviews as more in-depth than typical company culture interviews, even delving into ethical dilemmas, how the candidate reacted at the time, and how they viewed the situation afterward.

Anthropic President Daniela Amodei previously stated on a podcast that the company asks candidates what "unusual but defendable" beliefs they have, focusing on how candidates defend their judgment in uncomfortable situations.

ChatGPT's market share in consumer AI is being caught up, and the AI ​​market is entering a "three-way competition".

According to the latest traffic data from SimilarWeb, the consumer AI market is shifting from a single leader, ChatGPT, to a three-way competition between OpenAI, Google, and Anthropic.

Over the past six months, ChatGPT’s share has dropped from about 80% to about 60%, Gemini’s traffic relative to ChatGPT has increased from about 20% to about 50%, and Claude’s has increased from about 3% to about 20%.

These data still show that ChatGPT maintains its lead. The report also noted that ChatGPT has approximately 900 million weekly active users, Gemini approximately 500 million, and Claude approximately 150 million to 200 million; Claude has the fastest quarterly growth rate among the top websites, reaching 255%.

Jensen Huang responds to AI spending spree: Wasting a little money is fine, but don't waste time.

According to 36Kr, Nvidia CEO Jensen Huang recently responded to the issue of AI usage costs at an internal company meeting, saying, "It's okay to waste a little money, but the key is not to waste time."

During the internal meeting, some employees expressed concerns that the current widespread use of AI was merely a waste of tokens without bringing any substantial improvement in productivity. Jensen Huang responded that any technology is difficult for people to master in its early stages, so one should not strive for perfection from the outset, but rather take the first step and embrace new technologies.

On the issue of AI usage costs, Jensen Huang has consistently advocated that employees should use AI extensively. He previously stated that he would be concerned if an engineer with an annual salary of $500,000 spent less than $250,000 on tokens; and he would be "furious" if it only cost $5,000.

His logic is that he would rather spend half of his employees' salaries on AI in exchange for a 10-fold increase in efficiency.

Liu Qiangdong proposed the "3 x 12" principle, stipulating that JD.com employees should not work more than 12 hours a day.

According to TechWeb, JD.com recently launched a new internal work time management rule called "3 x 12". This principle was proposed by founder Liu Qiangdong last week and is being gradually implemented across multiple departments within the group starting this week. The relevant notification has already been disseminated downwards from the C1 and C2 levels, and frontline R&D teams have received specific instructions.

The "3 x 12" principle revolves around three quantitative indicators:

• Firstly, employees' daily working hours should generally not exceed 12 hours, with exceptions only during major promotional periods or for special projects.
• Secondly, employees are encouraged to leave the company no later than midnight, and each workplace should guide on-duty employees to leave their posts to rest and arrange for lights to be turned off at midnight.
• Third, employees should not work continuously for more than 12 days to prevent long-term overtime without rest.

JD.com stated that the introduction of these new regulations is a proactive adjustment made in response to the issue of excessive workload in some internal teams. The aim is to explicitly incorporate the physical and mental health of employees into the management orientation and to standardize the daily office and overtime management mechanism through quantified working time standards.

Burning through $500 million a month, corporate AI bills spiral out of control.

According to Axios, businesses are questioning the ever-increasing AI spending, with cost control and return on investment becoming the core contradiction in enterprise AI procurement in 2026.

Microsoft recently canceled most of its Claude Code licenses, with cost being one of the main reasons. Uber's COO has also publicly stated that AI-related spending is "increasingly difficult to justify."

An AI consultant told Axios that one of its clients, a company , burned through $500 million in a single month because it failed to set limits on the use of Claude licenses by its employees.

CloudBees CEO Anuj Kapur told Axios that some companies are laying off employees under the guise of AI automation, but in reality, layoffs may just be "the only lever they can pull" to offset soaring AI bills. Meanwhile, consumer attitudes toward AI are cooling, and resistance is also spreading among employees.

Ministry of Industry and Information Technology: Power batteries have entered the stage of large-scale retirement.

On May 28, the Ministry of Industry and Information Technology held the second meeting of the National Working Group on the Recycling and Utilization of Power Batteries for New Energy Vehicles, stating that my country's power batteries for new energy vehicles have entered a stage of large-scale retirement, and accelerating the improvement of the recycling and utilization system for waste power batteries has become a key task for 2026.

The meeting called for strengthened collaboration in the development of regulations, policies, and standards systems, as well as research on major issues, and called for a joint enforcement campaign to regulate the recycling and utilization of used power batteries. The focus will be on investigating and punishing violations such as the illegal sale of used power batteries, the use of used power batteries to produce substandard products, failure to fulfill information traceability responsibilities, illegal dismantling and environmental pollution, and unlicensed operations.

The Ministry of Industry and Information Technology also requires enterprises to fulfill their primary responsibilities, utilize digital technologies to strengthen the monitoring of power battery flow, and guide industry associations, key enterprises, and research institutions to cooperate in exploring advanced technologies and equipment and new business models.

Unitree's first embody smart experience center in Asia will be located in Jing'an District, Shanghai.

Unitree Robotics announced that its first Asian flagship store, the Unitree Embossed Smart Experience Center, will open tomorrow in Shanghai. Located on the second floor of Jiuguang Department Store at 1618 Nanjing West Road, Jing'an District, Shanghai, the store covers an area of ​​over 100 square meters and is situated across the street from Apple Jing'an. The store will showcase products such as humanoid robots and quadrupedal robot dogs.

BMW will use humanoid robots at its Leipzig plant.

According to the BBC, BMW plans to use humanoid robots in its European car production for the first time this summer, with two Aeon robots manufactured by Hexagon Robotics currently being tested at its Leipzig plant. Michael Nikolaides, BMW's head of process management and digitalization, stated that humanoid robots will be the future of car production.

The Aeon is 1.65 meters tall, weighs 60 kilograms, has a maximum moving speed of 2.4 meters per second, can carry 15 kilograms for short periods, and has a continuous carrying capacity of 8 kilograms. It is equipped with 21 sensors, including cameras, radar, microphones, force sensors, and torque sensors. Its battery lasts approximately 3 hours, after which it needs to be replaced.

Hello Clarifies Highway Toll Rules for Ride-Sharing

Hello released its ride-sharing highway toll rules yesterday. The core rule is that the method of bearing highway tolls is based on the passenger's remarks, and the driver accepting the order indicates acceptance of this method of bearing tolls.

• If the passenger selects "willing to negotiate the fee", the result agreed upon by both parties through the platform chat box and the platform phone shall prevail; the driver should complete the negotiation before accepting the order or within 2 hours after accepting the order and before the passenger gets in the car. If the negotiation is not completed within the time limit, the highway toll will be waived by default.
• If a passenger selects "I will bear all the costs", then the passenger will be solely responsible for the toll fees for the corresponding highway sections at the origin and destination of their trip.
• If you choose "I will not pay the toll", the passenger will not pay the highway toll, and the driver can decide whether to take the highway.

The rules also list violations, including drivers charging passengers who have refused to pay, charging highway tolls twice, charging tolls when no tolls were incurred, charging tolls that do not match the actual trip, and passengers refusing to pay after reaching an agreement.

Regarding violations, drivers who overcharge will be required to refund any fees and may have their growth points and trust points deducted, and their accounts may be suspended for 3 to 7 days, with serious violations resulting in a permanent ban. Passengers who fail to pay highway tolls as agreed will be unable to hail a ride until payment is completed. Hello also reminds passengers that highway tolls should be paid through the ride-sharing platform, not via offline transfer.

MiniMax officially launches A-share IPO

According to information on the official website of the China Securities Regulatory Commission (CSRC), MiniMax signed a tutoring agreement with CITIC Securities on May 29, officially launching its A-share IPO process.

Previously, Zhipu signed a tutoring agreement with Guotai Haitong Securities on February 9th of this year, and together with CICC, which had previously signed an agreement, participated in the tutoring work for listing on the Science and Technology Innovation Board. The start of tutoring for MiniMax means that the "two giants of large models," Zhipu and MiniMax, are expected to be listed on the A-share market simultaneously.

Zhipu's market capitalization rose to HK$880 billion.

According to Caixin, Zhipu AI's market capitalization in Hong Kong once rose to approximately HK$880 billion, with its share price nearly 17 times its IPO price of HK$116.2 in January. As of yesterday's closing, its market capitalization was HK$711.12 billion.

Mo Yan: AI-generated content is essentially secondhand; it cannot replace a writer's original creative ability.

According to an interview with Xinhua News Agency, Mo Yan recently stated when discussing AI and literary creation that he believes AI cannot replace the creative work of writers.

I believe the most valuable quality of a writer is the ability to create original works, to write novels or poems that they have never written before or that no one else has written before, and to create typical characters that have never appeared in other people's works. This is the reason and value of a writer's existence.

Mo Yan further analyzed that current AI-generated content does not possess originality capabilities.

AI is fed with the writings of generations of authors, and it can recombine these massive amounts of material to create works that seem to have never been seen before. But I think it's not original; it's still secondhand.

Mo Yan also acknowledged that the application of AI as an auxiliary tool has become a trend, especially in translation work where its value will gradually be demonstrated; AI will also become increasingly important in literary planning, film and television production, and technical aspects, and some positions will be impacted as a result.

In the interview, he also talked about the differentiating effect of AI on people: tools can make smart people even smarter, but they may also make people who lack judgment rely more on ready-made outputs.

vivo S60 series launched starting at 2899 yuan

The vivo S60 series was officially launched yesterday, with the vivo S60 starting at 3599 yuan and the vivo S60 Vitality Edition starting at 2899 yuan. The series is endorsed by Tian Xiwei and offers three color options: "Starry Sea," "Early Summer Green," and "Midsummer Night." The phones are 7.92mm thick and weigh 199g.

• Screen: 6.59-inch 1260p 144Hz flat screen;
• Imaging: 50MP front-facing camera; S60 rear camera setup includes a 50MP main sensor, an 8MP ultra-wide-angle lens, and a 50MP IMX882 3X periscope telephoto lens; S60 Youth Edition features a 50MP main sensor and an 8MP ultra-wide-angle lens.
• Performance: The S60 version is equipped with a Snapdragon 8s Gen 3 processor, while the Youth Edition is equipped with a Dimensity 7500.
• Battery: Both models are equipped with a 7200mAh Blue Ocean Battery and support 90W wired charging;
• Other features include: standard 3D ultrasonic fingerprint sensor, IP68 and IP69 ratings, stereo dual speakers, infrared remote control, and X-axis linear motor.

vivo is focusing on 4K Live images and Xiaohongshu (Little Red Book) posting scenarios this time. The official statement claims that the S60 supports 4K Live, Starlight Live, Live Color Palette, Inspiration Stickers Live, and 3D spatial effects generated from Live cover frames; vivo also claims that the S60 is the only product in the industry that supports posting 3D spatial images on Xiaohongshu.

Related reading: vivo S60 Review: A phone should be used more, thought less.

The Xiaomi 17T series will be sold in mainland China.

Yesterday, Lu Weibing, president of Xiaomi Group, revealed in an article that the Xiaomi 17T series will be released in mainland China and is expected to go online in early June.

Lu Weibing explained that Xiaomi first launched the Xiaomi T series "Xiaomi 9T" in 2019; subsequently, Xiaomi spent seven years and seven generations of products to compete head-on with the iPhone and Samsung Galaxy S series.

According to its information, the Xiaomi T series has over 15 million users worldwide, covering more than 100 countries and regions in Europe, Southeast Asia, Latin America, the Middle East, and Japan.

Recently, Xiaomi has launched the Xiaomi 17T series in Hong Kong/Taiwan, Japan, and other regions. The new phone offers two versions: the 17T and the 17T Pro. The former features a 6.59-inch flat screen, a Dimensity 8500-Ultra processor, and a Leica triple camera; the latter features a Dimensity 9500 processor, a 6.83-inch flat screen, and a Leica triple camera.

The basic kit is priced at 8999 yuan. Harbin AQUA has released a product, and Zero Zero Tech brings flying cameras to water sports.

Zero Zero Tech has released the Hafu AQUA, positioned as a 100% waterproof flying camera for water sports.

AQUA is a new series in the Hafu family. It weighs less than 250 grams and features IP67 dust and water resistance, positive buoyancy design, and water surface take-off and landing capabilities. It can be used in various scenarios such as wingsuit flying, wakeboarding, kitesurfing, kayaking, sailing, paddleboarding, and jet skiing.

AQUA can withstand winds up to level 7 and has a top speed of 40 km/h. It features a hydrophobic lens, anti-fog technology, and a corrosion-resistant structure. The imaging system uses a 1/1.28-inch CMOS sensor, can record 4K video at up to 100fps, and has more than 15 built-in intelligent camera movement modes, including paddleboard, kayak, and hydrofoil modes.

World wakeboarding champion Dominik Gührs stated that AQUA allows him to independently film water sports without needing to arrange photographers or complex equipment. The product is now available on domestic e-commerce platforms and offline channels, with the basic set retailing for 8999 yuan and the Flying Package for 9999 yuan.

Honor WIN Turbo Released Starting at 2699 Yuan

The Honor WIN Turbo was released yesterday, with a suggested retail price starting at 3299 yuan and a launch price starting at 2699 yuan. Versions include 12GB + 256GB, 12GB + 512GB, and 16GB + 512GB, with launch prices of 2699 yuan, 2999 yuan, and 3599 yuan respectively.

This device emphasizes a large battery and gaming capabilities, offering three color options: "Fast Gameplay," "Win with Expectations," and "Blue Sky Protection." It supports IP68, IP69, and IP69K dust and water resistance ratings. The front features a 6.79-inch 1.5K LTPS OLED flat screen, supporting 8000 nits peak brightness and 3840Hz "zero-risk" dimming.

In terms of performance, the Honor WIN Turbo is equipped with a Dimensity 8500 Racing Edition processor, offering up to 16GB of RAM and 512GB of storage. It also features a built-in 10000mAh Qinghai Lake battery, supporting 80W fast charging and 27W reverse charging.

Claude Code launches dynamic workflow preview version

Anthropic announced the launch of a preview version of "Dynamic Workflow" in the CLI, Desktop, and VS Code plugins for Claude Code, targeting long-term R&D tasks such as complex bug tracking, large-scale refactoring, framework migration, and pre-deployment adversarial testing across service codebases.

This mechanism transforms the original single-threaded programming agent into an orchestration-based workflow: the main agent writes orchestration scripts for tasks, breaking down the target into dozens to hundreds of parallel sub-agents;

Before merging the code, the system will also arrange for independent sub-agents to conduct multi-perspective adversarial testing and introduce two rounds of code review by Reviewer Agents. To support continuous operation, the dynamic workflow also adds incremental automatic archiving and breakpoint resume functionality.

Step 3.7 Flash Open Source Project

StepStar has released and open-sourced its next-generation Flash big model, Step 3.7 Flash, positioned as a model foundation for production-grade agents, code, web search, and multimodal workflows.

The model architecture consists of a 196B language backbone and a 1.8B visual Transformer, with a total of approximately 198B parameters and approximately 11B activation parameters. It supports 256K context and 3 levels of inference.

According to the official documentation, Step 3.7 Flash can autonomously select, crop, and reread local information in the UI or charts, and actively retrieve and verify information when it is uncertain. In terms of performance, the model achieved 67.1% in the ClawEval-1.1 autonomous task execution evaluation, 79.2% in SimpleVQA Search, 95.3% in V* Python, and 56.3% in SWE-Bench Pro.

Liquid AI open-source edge MoE model LFM2.5

Liquid AI has released and open-sourced its edge-side hybrid expert model, LFM2.5-8B-A1B. The model has a total of 8 parameters, but only 1 parameter is activated per run. Target devices include mobile phones, PCs, and robots. It supports local fine-tuning and offline operation on personal devices and single consumer-grade graphics cards.

This update shifts LFM2.5-8B-A1B to a "think first, answer later" reasoning model and expands the context size from 32K to 128K. Liquid AI also expands the vocabulary to 128K to improve the efficiency of non-English text processing; its data shows that processing efficiency for Thai and Hindi has improved by 238.2% and 120.4% respectively, and Arabic by 38.8%.

The team also made specific optimizations to address common issues in small models, such as circular reasoning and hallucinations. Liquid AI stated that targeted reinforcement learning makes the model more inclined to admit ignorance when it exceeds the knowledge boundary, increasing the hallucination-free rate from 7.46% in the previous generation to 63.47%.

Xiaomi open-sources ControlFoley, a controllable video and audio effects model.

Xiaomi's official WeChat account released and open-sourced ControlFoley, a multimodal controllable video audio generation framework designed for automatic video dubbing. It extends the "matching sound to visuals" approach to multi-condition control: creators can simultaneously input video, text commands, and reference audio, specifying the desired sound content and timbre style.

ControlFoley primarily addresses three types of control problems:

• When there is a semantic conflict between text and visuals, the model can prioritize responding to the user's textual intent.
• When the user provides reference audio, the model attempts to preserve the reference timbre without disrupting audio-visual synchronization;
• When the combination of input conditions changes, unified multimodal representation and randomized modal dropout training enable the model to remain stable in tasks such as TV2A, TC-V2A, and AC-V2A.

Xiaomi's experimental results show that ControlFoley achieves better semantic alignment, time synchronization, and sound quality performance compared to open-source baselines on multiple datasets.

Hugging Face: huggingface.co/YJX-Xiaomi/ControlFoley

Tencent's Hy-Memory plugin for Agent memory has been released.

Tencent Hunyuan released Hy-Memory, positioned as a memory plugin for long-term collaborative intelligent agents such as OpenClaw. It aims to solve the problems of memory fragmentation, attention dilution, and old preference overwriting that easily occur when long-cycle agents run across days and sessions.

• Hy-Memory's architecture consists of a 6-layer memory framework, breaking down memory into layers such as facts, profiles, mental models, and forward-looking intentions.
• At the same time, a dual-system mechanism is introduced: System 1 is responsible for extracting real-time facts and conversation summaries in milliseconds, while System 2 refines user mental models and knowledge networks in the background.
• The system also uses supersedes pointers to link old and new memories into an evolutionary chain, allowing the search results to simultaneously retain the latest conclusions and the path of preference changes.

In terms of evaluation, Tencent stated that Hy-Memory achieved a score of 85.2 in LongMemEval, and led in metrics such as preference, temporal reasoning, and knowledge update.

In terms of performance, its write speed is 8 times that of Graphiti, the number of memory records is about 1/3 of mem0 and 1/4 of Graphiti, and the token consumption in long context processing is reduced by 35%. Hy-Memory offers three configurations: Lite, Pro and Ultra, and uses the Chroma native embedded vector library by default.

Starting next month, those who refuse to pay the fare after violating travel rules will be restricted from purchasing tickets.

According to CCTV News, citing China State Railway Group, the new version of the "China State Railway Group Co., Ltd. Railway Passenger Transport Regulations" will be officially implemented on June 1, and for the first time, it sets restrictions on ticket purchases for passengers who violate regulations and refuse to pay for their tickets.

The new regulations cover violations including traveling without a ticket, traveling with an altered, forged, or defaced travel pass, discrepancies between the ticket, pass, and passenger, traveling in a higher-priced seat class with a lower-priced ticket, and holding a discounted or preferential ticket without a valid discount voucher or meeting the discount conditions.

In the event of the above-mentioned circumstances and the passenger refuses to pay the outstanding fare, railway station and train staff will inform the passenger on-site of the violation and subsequent ticket purchase restrictions. When the restricted passenger attempts to purchase a ticket again, the railway transport company will inform the passenger of the reason for the restriction through 12306 or the ticket window. After the passenger pays the outstanding fare plus a surcharge of 50% of the outstanding fare, the ticket purchase restriction will be automatically lifted online or offline.

After purchasing a replacement ticket, an electronic railway invoice can still be issued within 180 days from the end of the original trip. The new passenger regulations also adjust the student ticket discount range, the refund rules for connecting tickets, and the rules for sending lost items.

Xiaohongshu videos support Dolby Atmos

Yesterday, Dolby Laboratories and Xiaohongshu announced the launch of Dolby Atmos content sharing functionality on Xiaohongshu. The two companies introduced Dolby Vision to Xiaohongshu in 2024, and with the addition of this audio dimension, creators can now publish video content featuring both Dolby Vision and Dolby Atmos on supported devices.

Xiaohongshu users can preview, share, and enjoy Dolby Atmos content on the Xiaohongshu mobile app using an iPhone or Android phone that supports Dolby Atmos; they can also access the Xiaohongshu website via Safari on a Mac or iPad to share and watch.

Mixue Ice Cream launches new mint series products

Mixue Ice Cream announced the launch of its new mint series yesterday, featuring refreshing summer flavors. The series includes four products: mint ice cream, mint crispy mini sundae, mint milk green tea, and Snow King Snow Top Coffee Mint Flavor.

According to the official description, the Mint Milk Green Tea combines the flavors of mint, jasmine green tea, and milk; the Mint Ice Cream offers two flavor options: mint and double-flavor; and the Mint Crispy Sundae consists of mint ice cream, chocolate shell, and Oreo cookie crumbs.

I really want to remove the controversial NFC juice from the shelves.

According to Jiemian News, following CCTV News' exposé of irregularities in juice labeling, a series of juice drinks under the brand Haoxianglai has attracted attention. The program mentioned that the product's outer packaging prominently displays "100% juice" and "NFC (Not From Concentrate)," but the first two ingredients listed are water and concentrated juice, indicating a discrepancy between the label information and the actual ingredients.

A public relations representative from Wancheng Group, the parent company of Haoxianglai, responded that the product in question involved a labeling defect, which consumers may have mistaken for pure NFC (Not From Concentrate) juice. The company has already removed the product from shelves and will repackage it to avoid any ambiguity. The report also cited previous cases involving brands like Beibingyang and Master Kong, which had faced controversies regarding juice content and packaging labeling.

It's the weekend!

One Fun Thing | Nintendo made a weird game that flips through your photo album

According to The Verge, Nintendo launched the mobile game Pictonico this week. It looks like a lightweight version of WarioWare: each round consists of 10 mini-games lasting only a few seconds, with simple on-screen commands like "chomp," requiring players to immediately understand the controls, such as grabbing a mouth and making it chew food.

What's interesting is that the game uses photos from your phone's photo album to customize the experience. The Verge mentioned that players will see photos from their photo album being re-added into the mini-game scenes during absurd actions such as plucking hair, licking lollipops, and peeling bananas.

It doesn't require sitting down to figure out how to play; it's more like a fragmented toy that mixes a private photo album, a few seconds of reaction time, and Nintendo-style absurdity. It transforms the phone's photo album from something to be "browsed" into something to be "used as a prop in a mini-game."

What to watch this weekend | The Mandalorian and the Gugu

Star Wars: The Mandalorian and the Gugu is the first theatrical film in Disney's The Mandalorian TV series. The story takes place after the third season, and is directed by Jon Favreau, co-written by him, Dave Filoni, and Noah Krull, and stars Pedro Pascal.

The story focuses on bounty hunter Din Djarin and his non-blood-related son Grogu, who are on a top-secret mission in a galaxy in turmoil following the fall of the Empire, a mission that will determine the fate of the New Republic.

The film continues the core narrative axis of the series, "father-son relationship," and advances with a game-like "level-clearing and monster-fighting" structure. The scope is relatively restrained and does not focus on the macro-level confrontation between the new republic and the remnants of the empire.

As of May 29, 2026, the film received a score of 7.4 on Douban (27,407 reviews), which is better than 80% of action films and 83% of science fiction films.

A Guide to Buying Books Without Reading Them: "The Big Company Syndrome: From Individual Dilemmas to Organizational Performance"

The author of "Big Company Disease," Xi Fanjun, has over ten years of experience working in large technology companies, specializing in strategic communication and technology governance. He now focuses on research into organizational structure and public narrative.

This book takes a sociological approach, drawing on theoretical resources from Weber, Foucault, Bauman, Bourdieu, and others, to systematically dissect the structural roots behind contemporary workplace phenomena such as "involution," "lying flat," and "the crisis of age 35."

The book's core argument is that individual anxiety and burnout are not due to insufficient effort, but rather an inevitable product of the alienation of modern organizational structures. Xi Fanjun distills a series of original concepts such as "liquid man," "safe man," "performative governance," and "the cage of language," dissecting how large companies reshape human behavior and erode people's sense of meaning through processes, performance metrics, and rhetoric.

The book points out that when a company evolves into a "self-sustaining system," middle managers become "loyal hostages," while senior employees face a higher risk of marginalization—"the right process, no one survives" is the core metaphor that runs throughout the book.

Game Recommendation | Steam's free game "Drift86" is perfect for trying out drifting this weekend.

Independent developer RewindApp's arcade drift racing game *Drift86* is currently available for a limited time free on Steam. Originally priced at 11 RMB in China, players can now claim it for free and add it permanently to their library.

Drift86 features a light drifting gameplay, including over 40 vehicles and 30+ maps, and supports single-player and multiplayer modes. The game's atmosphere is influenced by Initial D, featuring Eurobeat music and arcade-style cornering rhythms, with a focus on quickly entering a cycle of drifting, scoring, and racing.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

From digital innovation to artistic communities, a new paradigm is transforming the way we produce, share, and experience creativity.

For decades, the worlds of technology and artistic culture have traveled on parallel tracks, each believing they spoke a language the other couldn't understand. Today, that gap has dramatically narrowed. Algorithms, digital platforms, and audiovisual production tools have firmly entered the spaces of contemporary art, transforming not only creative processes , but also the way communities organize around them.

It's not simply a matter of using software instead of a canvas. The change is more profound: technology has redefined who can create, how they can do it, and, above all, with whom. Digital systems have become fully-fledged cultural infrastructures, capable of supporting networks of people, ideas, and practices that previously struggled to find physical space or visibility.

A new generation of hybrid cultural spaces

Among the organizations that best embody this paradigm shift is BrainArt , a project that operates at the intersection of artistic expression, technology, and community activation. Its underlying philosophy mirrors exactly what many digital innovation theorists describe as the future of creative ecosystems : no longer separate places where "art" and "tech" are made, but hybrid spaces where both dimensions feed off each other.

BrainArt demonstrates how a structured cultural community can adopt digital tools without losing its identity, but rather strengthening it. Content production, project communication, and relationship building take place through channels and approaches borrowed from the digital world, yet at the service of a very specific human and territorial vision.

Technology as an enabler, not an end

The point that emerges most clearly from experiences like Brain Art is that technology works best when it is not the goal, but the instrument of a larger goal . This seemingly simple principle is actually one of the most difficult to apply in the context of contemporary digital innovation, where the temptation to pursue technological novelty for its own sake is constant.

Cultural organizations that successfully use digital technology do so because they start with a specific question: what do we want to communicate, to whom, and why? Only then do they ask: with which platform, in which format, with which technology? This order —vision first, tool second— is precisely what distinguishes good practice from a mere experiment.

The value of communities as a system

There's another lesson the tech world can learn from observing realities like Brain Art: the systemic value of communities . In computer science, we often talk about systems, networks, and interconnected nodes. But we rarely reflect enough on how these metaphors translate into concrete social practices.

An active artistic community isn't just a group of people with a common interest. It's a generative system , capable of producing resources, relationships, skills, and innovation in a distributed and continuous manner. It's exactly the kind of structure the digital ecosystem should aspire to support and replicate: not vertical and centralized, but horizontal, resilient, and capable of self-organization.
Towards a more conscious innovation

The dialogue between technology and culture isn't a luxury reserved for large institutions or international festivals. It happens, every day, in concrete ways, in laboratories, production studios, and in the digital spaces frequented by those who create art with the tools of their time.

Recognizing and valuing these practices is one of the most significant actions the tech community can take today. Not to romanticize art, nor to reduce it to a mere use case, but to understand that authentic innovation and shared creativity ultimately speak the same language.

The article "Creativity and Technology: An Ecosystem Built Together" was written on: Tech | CUENEWS .

Apple's iOS 27 AI Siri interface revealed

Yesterday, Bloomberg reporter Mark Gurman released a set of UI mockups for iOS 27, showcasing the new Siri interface, new chatbot-style applications, and other major changes Apple plans to announce in iOS 27 at WWDC26.

It is understood that the new version of Siri will have two launch options:

• Traditional method: Saying "Siri" or pressing and holding the iPhone's power button will trigger the new Siri animation in Lingdong Island.
• The new "Search or Ask" feature: iOS 27 will allow users to swipe down from the top center of the iPhone from anywhere on the system to launch the new "Search or Ask" interface.

It is understood that "Search or Ask" supports voice and text input, has screen content awareness and personal data analysis capabilities, and can perform multi-level tasks across applications based on context.

In addition, Apple will launch a standalone Siri app similar to ChatGPT, which will retain historical conversations and support the uploading of multimodal attachments for AI analysis.

In terms of imaging and native applications, Siri will be directly integrated into the camera app for the first time, using a third-party AI agent for object recognition and reverse image search.

It's worth mentioning that the camera interface in iOS 27 will introduce a customizable control panel to meet professional shooting needs; while the Photos app will add computational photography features such as AI Reframe and AI Extend.

Beyond its AI features, iOS 27 will also focus on fixing bugs related to last year's "Liquid Glass" design.

Claude Opus 4.8 Released: Fast Mode Costs Reduced to One-Third

Early this morning, Anthropic officially released its flagship model, Claude Opus 4.8, upgrading its programming, AI agent tasks, reasoning, and knowledge work capabilities while maintaining the same price.

Standard mode pricing remains at $5 per million input tokens and $25 per million output tokens; fast mode pricing is $10 per million input tokens and $50 per million output tokens, with a running speed approximately 2.5 times faster than the previous generation and model costs reduced to one-third of the previous generation.

The core improvement in this upgrade lies in judgment and honesty. Official assessments show that Opus 4.8 is about one-quarter as likely as its predecessor, Opus 4.7, to pass tests with unexplained code defects, and is more willing to proactively label uncertainties to reduce unfounded conclusions.

In benchmark tests, Opus 4.8 scored 69.2% on SWE-Bench Pro, higher than Opus 4.7's 64.3%, and surpassed GPT-5.5 and Gemini 3.1 Pro in multiple tests including Agentic Coding, Agentic Computer Use, Knowledge Work, and Finance Agent. However, on the Terminal-Bench 2.1 terminal programming benchmark, GPT-5.5's 78.2% still outperformed Opus 4.8's 74.6%.

Anthropic also stated that it plans to roll out the more capable Claude Mythos series models to all customers in the coming weeks.

Related reading: Claude Opus 4.8 just launched, immediately claiming to be DeepSeek and Qwen

The cumulative number of devices connected to the HarmonyOS ecosystem exceeds 1.3 billion.

Yesterday, at the HarmonyOS Developer Conference 2026, a Huawei representative stated that the cumulative number of HarmonyOS smart hardware ecosystem devices has exceeded 1.3 billion .

In addition, Xie Shaofeng, chairman of the Open Atom Foundation, said that the open source HarmonyOS community has exceeded 140 million lines of code, with more than 13,000 contributors, and supports various forms such as mobile phones, tablets, computers, TVs, wearables and industry smart terminals.

DeepSeek crashed again.

Yesterday morning, numerous users reported DeepSeek service anomalies on social media platforms. The DeepSeek status page indicated performance degradation in web chat and API services, with the event marked as "partially interrupted" at 10:21 AM. Affected components included the API service and web chat service.

The status page updated at 10:27, showing that the cause had been identified and a fix was underway. The incident lasted approximately 22 minutes; an update at 10:50 showed that service had been restored.

Xiaomi surpasses Volkswagen and Toyota to become the world's seventh largest new energy vehicle manufacturer.

TrendForce recently released industry insights showing that:

In the first quarter of 2026, global sales of new energy vehicles, including battery electric vehicles (BEVs), plug-in hybrid electric vehicles (PHEVs), and hydrogen fuel cell vehicles, totaled 3.94 million units, a 2% year-on-year decrease, accounting for 19% of global automobile sales in the first quarter.

Among them, China's new energy vehicle market has entered a period of adjustment; meanwhile, the Western European market is showing signs of recovery, and pure electric vehicle sales in Japan and South Korea have also seen significant growth. It is worth noting that:

• Tesla surpasses BYD to regain the top spot in pure electric vehicle sales;
• Leapmotor, ranked fifth, achieved annual sales growth against the trend thanks to its rapidly expanding product portfolio and cost-effective strategy.
• With a market share of 2.9%, Xiaomi surpassed Volkswagen and Toyota to rank seventh in the pure electric vehicle market.

The OpenAI Foundation establishes a $250 million AI fund.

The OpenAI Foundation, the nonprofit that controls OpenAI, announced yesterday that it will invest an initial $250 million to help workers, communities, and economies cope with the job and income structure changes brought about by AI.

This funding will be used to research the impact of AI on the labor market, support workers and communities facing short-term job displacement pressures, and explore ways to more broadly distribute the economic benefits of AI. Public sources confirm the initial funding size, funding objectives, and the foundation's entity; the distribution of economic benefits will be one of the foundation's key focuses.

Last year, the OpenAI Foundation pledged $50 million to support public interest projects. This time, the $250 million fund expands its focus from general philanthropy to the impact of AI on the workforce and the distribution of benefits.

Guangdong responds to rumors about AI grading of college entrance exam papers

According to Guangdong Education Media, the Guangdong Provincial Education Examinations Authority responded, stating that the rumor that "AI is used for grading the summer college entrance examination in Guangdong" is false. The report mentioned that Guangdong has indeed used intelligent technology in college entrance examination management in recent years, but its purpose is to assist in examination security aspects such as invigilation, security checks, radio signal jamming, and inspection of secure rooms.

The Guangdong Provincial Education Examinations Authority stated that subjective questions will still be graded manually by teachers, with two teachers independently grading each question. Questions with scores exceeding a threshold will undergo a third grading process. The system will only be responsible for random sampling, real-time monitoring, and error correction. In 2025, the Guangdong grading center will be located at South China Normal University, with over 6,200 teachers selected from across the province to participate in the grading.

Huang Renxun will join Tsinghua University's School of Economics and Management Advisory Committee

According to the Financial Times, Jensen Huang has accepted an invitation to join the Tsinghua SEM Advisory Board.

The Tsinghua University School of Economics and Management Advisory Board was established in 2000. Its members have long included multinational business leaders and representatives in the field of economics and management. Apple CEO Tim Cook previously served as the chairman of the board. The board's mission is to help the school become a world-class school of economics and management.

It is understood that the committee brings together the heads of tech giants such as Apple, Microsoft, Meta, Tesla, Alibaba, and Tencent, as well as the leaders of financial giants such as JPMorgan Chase, Blackstone Group, and Bridgewater.

A Google engineer has been charged with using internal search data to gamble and win $1.2 million.

According to TechCrunch, the U.S. Department of Justice has indicted Google software engineer Michele Spagnuolo, accusing him of profiting $1.2 million on Polymarket by using non-public business information from within Google.

The indictment alleges that Spagnuolo used the account "AlphaRaccoon" on Polymarket to place bets on Google's 2025 Year in Search campaign results, with an exposure exceeding $2.7 million. TechCrunch reported that these bets were related to search trend data that Google has not yet publicly released.

Li Auto spun out of profit in the first quarter

Yesterday, Li Auto released its first-quarter results for fiscal year 2026. During the period, total operating revenue was RMB 22.983 billion, a year-on-year decrease of 11.4%; net loss attributable to the parent company was RMB 2.29 billion, compared with a profit of RMB 650 million in the same period last year; gross profit was RMB 1.808 billion, a year-on-year decrease of 66.0%, and the gross profit margin was 7.9%, a year-on-year decrease of 12.6 percentage points.

Operationally, Li Auto delivered 95,142 vehicles in the first quarter, a year-on-year increase of 2.5%. The vehicle gross margin decreased to 6.1%, with net operating cash outflow of RMB 6.091 billion and net free cash outflow of RMB 7.388 billion. The company expects to deliver 95,000 to 100,000 vehicles in the second quarter, corresponding to revenue of RMB 24.1 billion to RMB 25.4 billion.

As of April 30, Li Auto had 511 retail centers and 550 after-sales service centers in 160 cities, and 4,077 supercharging stations in operation.

XPeng reported a loss of 1.78 billion yuan in the first quarter, and its humanoid robot is expected to debut in the third quarter.

XPeng Motors released its unaudited results for the first quarter of 2026. Total revenue for the period was RMB 13.03 billion, a year-on-year decrease of 17.6% and a quarter-on-quarter decrease of 41.4%. Vehicle sales revenue was RMB 11 billion, a year-on-year decrease of 23.5%, mainly due to a decline in deliveries. The company reported a net loss of RMB 1.78 billion in the first quarter, compared to a loss of RMB 660 million in the same period last year and a profit of RMB 380 million in the previous quarter.

• The overall gross profit margin for the first quarter was 20.6%, higher than 15.6% in the same period last year, but lower than 21.3% in the previous quarter.
• The gross profit margin for automobiles was 12.1%, an increase of 1.6 percentage points year-on-year;
• Research and development expenses amounted to 2.91 billion yuan, a year-on-year increase of 46.8%, which the official explanation was increased investment in new models and AI-related technologies.

XPeng expects to deliver 100,000 to 106,000 vehicles in the second quarter, representing a quarter-on-quarter increase of 59.5% to 69.1%; total revenue is expected to be between RMB 19.6 billion and RMB 20.8 billion.

He Xiaopeng, Chairman of XPeng Group, stated in the Q1 earnings call that the development of the next-generation XPeng IRON humanoid robot's software and hardware for mass production is progressing smoothly and is about to enter the ET2 software and hardware integration phase. The company plans to officially launch the robot in Q3 of this year and aims to achieve mass production of high-end humanoid robots by the end of this year.

Weibo's net revenue in the first quarter was US$421.3 million.

Weibo released its unaudited results for the first quarter of 2026, with net revenue of US$421.3 million, up 6% year-over-year or 1% year-over-year at constant exchange rates; net income attributable to Weibo shareholders was US$34.7 million, or US$0.14 per diluted share.

Advertising and marketing revenue remained the main driver, reaching $369.8 million in the first quarter, a 9% year-on-year increase; of which, excluding Alibaba's advertising revenue, advertising and marketing revenue was $326.5 million, a 10% year-on-year increase. Value-added service revenue was $51.6 million, an 11% year-on-year decrease, mainly due to a reduction in revenue contribution from game-related businesses.

In terms of user scale, Weibo had 562 million monthly active users and 254 million daily active users in March. CEO Wang Gaofei mentioned in the financial report that the company optimized the homepage feed experience, improved the supply and distribution of high-quality video content, and enhanced search and advertising conversion efficiency through AI.

MiniMax's annualized revenue doubled in two months

According to CBN, MiniMax disclosed its latest business data yesterday: the number of global enterprise and developer clients served by the company has exceeded 1 million, a five-fold increase compared to six months ago; the global user base is approximately 300 million.

In terms of revenue, MiniMax's annualized recurring revenue (ARR) has grown by more than 100% in the past two months. The company previously disclosed at its March earnings conference that its ARR had exceeded $150 million by February 2026; based on this data, the doubling cycle of ARR has been compressed to about 60 days.

Usage also increased in tandem. MiniMax stated that the daily token consumption of the M2 series models increased sixfold within two months, and the number of new registered users on the open platform increased more than fourfold.

OpenRouter raises $113 million in Series B funding.

AI model API platform OpenRouter officially announced the completion of a $113 million Series B funding round, led by Alphabet's growth fund CapitalG, with participation from NVentures, ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures, Databricks Ventures, AMP PBC, Pace Capital, and existing investors a16z and Menlo Ventures.

According to the official statement, the platform's weekly token processing volume has increased from 5 trillion to 25 trillion in the past six months, and is expected to exceed 1,000 trillion tokens this year, serving more than 8 million developers and covering more than 400 models.

The new funding will be used to expand infrastructure, enterprise capabilities, and intelligent routing. OpenRouter noted that the platform already supports image, audio, speech, transcription, embedding, and video models in addition to text, and provides workspaces, cost management, guardrails, zero data retention policies, vendor-grade failover, and cost/latency optimization.

Luo Fuli: MiMo API price reduction stems from lower inference costs

Xiaomi MiMo officially announced a permanent price reduction for its MiMo-V2.5 series API, with reductions of up to 99% compared to the original API pricing, and the distinction between different context window lengths has been removed. Luo Fuli, head of Xiaomi MiMo, subsequently posted on the X platform explaining the technical reasons behind this price reduction.

Luo Fuli stated that MiMo's original inference cost is far below the industry average, and the new price "can basically maintain a break-even point." She attributed this to both the model and the system: the model itself has been optimized in terms of inference efficiency, and the backend service has also reduced the cost per token through engineering optimization.

Luo Fuli emphasized that MiMo-V2.5 no longer differentiates prices based on context window length, aiming to lower the barrier to entry for using long context windows and make it easier for developers to try it out in real-world applications. It's noteworthy that Luo Fuli also advised competitors in her post against "blindly lowering prices":

We previously advised LLM companies against "blindly lowering prices" precisely because few model architectures and inference optimizations can prevent API costs from operating at a loss.

She believes that MiMo's low cost stems from its specific architectural design and reasoning optimization capabilities. If other companies do not have a similar foundation, blindly following suit may lead to losses.

She also elaborated on the industrial significance of this logic from a more macro perspective: reasonably priced, high-performance model APIs will drive the demand for truly large-scale inference, thereby accelerating the development of the AI ​​infrastructure chain, including chips, servers, liquid cooling, power supply, and data centers, forming a virtuous cycle in the industry.

Sony releases new Bravia TVs for 2026.

Yesterday, Sony launched new Bravia TVs at its 2026 Home Entertainment New Product Launch Event, including the 9 Series II, 8 Series II, and 7 Series II, with prices starting at 10,999 yuan.

The 9-series second generation and 7-series second generation models come standard with True RGB color technology. This technology uses independent red, green, and blue LED backlights, improving color purity and brightness through finer backlight control. The 8-series second generation continues the OLED approach, targeting users who prioritize contrast and black levels.

Related reading: Is the Sony RGB TV priced at 149,999 yuan a hit or a flop? We went to Sony headquarters to take a look.

The Song Ultra DM-i is launched, starting at 129,900 yuan.

BYD yesterday launched the Song Ultra DM-i, the first B-class plug-in hybrid SUV equipped with the "Didi Shrimp" intelligent system. The new car has five models, priced from 129,900 yuan to 159,900 yuan.

The Song Ultra DM-i measures 4850/1910/1670mm in length, width, and height, with a wheelbase of 2840mm. Equipped with fifth-generation DM technology, it boasts a maximum pure electric range of 310km and a maximum combined range of 1845km, with a fuel consumption of approximately 3.3L/100km when the battery is depleted. The powertrain consists of a 1.5L engine and an electric motor, with the engine delivering a maximum power of 70kW and the motor a maximum power of 175kW.

In terms of configuration, the new car offers features such as a queen-like passenger seat, a double bed mode with the front and rear seats folded down, a 16-speaker "Sky Sound" audio system, Cloud Carriage-C, and TBC high-speed tire blowout stability system.

Even the entry-level model comes equipped with 7 airbags, a panoramic sunroof, a power sunshade, a power tailgate, 19-inch wheels, a 15.6-inch central control screen, wireless charging, and a dashcam.

Guangfan AI wearable devices are integrated with Tencent Mobility.

Yesterday, Guangfan Technology announced a partnership with Tencent Mobility Services. Guangfan's AI-powered wearable devices will be integrated with Tencent Mobility Services, with related functions launching on June 5th. Users can initiate ride-hailing requests via voice, and the system will guide them through the process of price estimation, order placement, and automatic payment based on their current location, destination, and other information.

After a user initiates a ride-hailing command via voice, the system can guide the user through the process of price estimation, ordering a ride, and automatic payment based on the user's current location and destination information.

The Guangfan AI All-Sense Wearable Device, Guangfan Technology's first product, officially went on sale on May 15th. Official information shows that after its launch, the product topped the JD.com AI Headphones best-selling list and maintained that position for eight consecutive days. The first batch of products has already sold out, and a new round of pre-orders has now begun.

vivo officially announces 6 major selling points of its first over-ear headphones.

vivo yesterday announced six key selling points of its first wireless over-ear noise-canceling headphones, which will be officially launched on May 29th at 7:30 PM. The official specifications claim that these headphones feature 58dB active noise cancellation, 75 hours of battery life, and a lightweight design of only 238g.

The product also features ultra-soft leather earcups and supports seamless connectivity across three devices in the ecosystem, making it suitable for long commutes, office work, and audio-visual scenarios.

Xiaomi Recruitment Agent Launched

Yesterday, Xiaomi officially announced the launch of "Xiaomi Recruitment Agent" on its recruitment website, leveraging Xiaomi's MiMo big data model. According to reports, Xiaomi Recruitment Agent provides job seekers with services such as policy explanations, intelligent job recommendations, and recruitment progress tracking on the Xiaomi recruitment website.

OPPO Watch X3 Ideal Custom Edition Launched

Li Auto officially announced yesterday that the OPPO Watch X3 Li Auto Custom Edition is now available on the Li Auto Mall at a limited-time price of 2329.1 yuan.

This watch is designed to integrate with Li Auto's car keys and vehicle control scenarios. The main product is based on the OPPO Watch X3, while the Li Auto customized version focuses on unlocking on the wrist, accessing vehicle control, and providing owner benefits.

Broadcom releases its first batch of Wi-Fi 8 router integrated chips

Recently, Broadcom released three Wi-Fi 8 router integrated chips: BCM6772, BCM6774, and BCM6776, targeting next-generation home, enterprise, and carrier routing devices.

According to reports, Wi-Fi 8 focuses on improving stable connections, low latency, and multi-device concurrent experiences in complex wireless environments. Broadcom's chips are designed for routers of various sizes, allowing manufacturers to integrate wireless, network processing, and system functions into a single chip solution, reducing overall design complexity.

Alibaba DAMO Academy releases GPU version solver

According to a report from the official WeChat account of Alibaba DAMO Academy, the "MindOpt" solver developed by Alibaba DAMO Academy has officially released a GPU version. It utilizes the parallel acceleration capabilities of GPUs and introduces new algorithms to handle the "long tail effect" in large-scale optimization problems.

According to DAMO Academy, in approximately 2,000 general-purpose test cases, MinDie can stably solve over 99% of problem types to high accuracy, and supports linear programming problems with hundreds of millions of variables that are traditionally difficult to handle. The solver primarily serves complex computing scenarios such as power dispatching, flight scheduling, high-end manufacturing, and financial management.

Traditional linear programming solvers are mostly CPU-based, and as the problem size increases, matrix factorization, memory requirements, and parallelism become bottlenecks. GPU versions advance large-scale optimization problems from "whether it's feasible" to solving them with time and resource costs closer to those usable in industrial systems.

Alibaba's Qoder launches Cloud Agents

Alibaba's Qoder has launched Cloud Agents, a fully managed AI agent runtime platform that provides enterprises with capabilities such as agent foundation, model services, and runtime environment. Enterprises can use API calls to enable agents to execute tasks in the cloud.

According to the Qoder documentation, developers can define agents and initiate sessions via API. Complex tasks run in the cloud and return results in real time. Alibaba Cloud claims that the platform can reduce the agent deployment cycle from approximately one month to one day.

The independent variable robot, Wall-OSS-0.5, is open-source and can be directly deployed after pre-training with a large embodied model.

Independent Variable Robotics released and open-sourced its embodied large model, Wall-OSS-0.5, yesterday. The official positioning is that it's a "pre-train once, deploy anywhere" vision-language-action (VLA) model, with its core selling point being that it can be directly deployed after pre-training without any task fine-tuning. Wall-OSS-0.5 introduces four technologies:

• The Gradient-bridge collaborative training mechanism enables the visual, language, and action modalities to learn collaboratively in a unified optimization direction, allowing the physical knowledge accumulated during visual and language pre-training to be extended to the action dimension.
• A unified visual-action representation for VLA binds the action codebook to visual features based on residual vector quantization, using future frame changes as the reconstruction target, so that each action token corresponds to an observable physical change, thereby simultaneously improving the accuracy of real-device actions and multimodal reasoning capabilities.
• Action-space supervision calculates the loss in the decoded real action space, allowing the gradient to directly reflect the control error and automatically weights the key moments of physical interaction to accelerate convergence.
• DMuon, the first distributed Muon optimizer for large-scale pre-training of embodied systems, reconstructs communication patterns through dedicated home scheduling and eliminates redundant computations with a symmetric-aware CuteDSL kernel, reducing the additional overhead introduced by Muon by approximately 100 times.

Official test results show that Wall-OSS-0.5 achieved an autonomous completion rate of over 80% in 4 out of 17 real-machine zero-sample tasks; in a fair comparison of more than ten real robot tasks, its performance in operation and reasoning tasks outperformed similar open-source models.

GitHub: github.com/X-Square-Robot/wall-x

Hugging Face: huggingface.co/x-square-robot/wall-oss-0.5

WBench Open Source Interactive World Model Benchmark

Fudan University and Meituan's LongCat team have open-sourced WBench, a benchmark for evaluating interactive video world models. The benchmark includes 289 multi-round test cases, covering five evaluation dimensions, and is calibrated against human judgment using 22 metrics.

This benchmark focuses on the model's physical rules, spatiotemporal consistency, instruction compliance, and control capabilities in continuous interactions, evaluating quality beyond that of a single video generation. The paper abstract summarizes these five dimensions as video quality, scene compliance, interaction compliance, consistency, and physical compliance.

GitHub: github.com/meituan-longcat/WBench

Hugging Face: huggingface.co/datasets/meituan-longcat/WBench

Doubao has reportedly integrated into Douyin's lifestyle services ecosystem, enabling direct group-buying.

According to Dujia, ByteDance's AI assistant Doubao is integrating with Douyin's lifestyle services to provide group-buying package recommendations, in-app merchant pages, payment and offline verification processes within the platform.

After a user initiates a group-buying inquiry on Doubao, Doubao will recommend corresponding stores and package details; after clicking, the user can enter the merchant page within the app to complete the package selection and purchase.

Meituan Releases AI Solution for Instant Retail Merchants

Yesterday, Meituan released its first AI-powered solution specifically for instant retail merchants, "Claw". This solution targets instant retail merchants and covers scenarios such as business analysis, product management, marketing campaigns, and operational recommendations.

The challenges of instant retail stem from the rapid changes in products, inventory, fulfillment, and local demand. Merchants need to frequently adjust product supply, pricing, and promotions. Meituan has placed AI tools in its operational backend, aiming to transform platform-side data capabilities into actions that merchants can directly execute.

Beijing and Shanghai jointly summoned three major food delivery platforms for talks.

According to CCTV News, the Beijing Municipal Administration for Market Regulation and the Shanghai Municipal Administration for Market Regulation signed the "Cooperation Framework Agreement on Food Safety Governance of Third-Party Online Catering Platforms," ​​establishing a collaborative governance mechanism for food safety in online catering and, for the first time, jointly providing administrative guidance to the three major food delivery platforms: Meituan, Taobao Flash Sale, and JD.com Waimai.

The market supervision departments of the two regions will collaborate in areas such as data sharing, information exchange, joint risk control, and law enforcement coordination to promote the implementation of a dual-track system of "online review + offline verification" on food delivery platforms, and urge platforms to verify and compare the food business licenses and other qualification information of online catering merchants with provincial market supervision data.

This administrative guidance corresponds to the "Regulations on Supervision and Management of Online Catering Service Operators' Implementation of Their Primary Responsibility for Food Safety," which will take effect on June 1st. The regulatory authorities require platforms to improve their systems for substantive qualification verification, online information disclosure, food delivery management, and food safety management.

"The Annual Party Can't Stop 2!" is scheduled to premiere on August 7th.

Yesterday, the workplace comedy film "The Annual Meeting Can't Stop 2!" officially announced its release date, set for August 7th, starring Zhang Ruoyun and Bai Ke.

The first film, "The Annual Meeting Can't Stop!", centered its comedy around the organizational structure, workplace slang, and annual meeting culture of a major internet company, gaining attention from the perspective of "working class." The sequel continues the workplace comedy genre, and the pairing of Zhang Ruoyun and Bai Ke maintains the film's clear target audience of young and working professionals.

X-Men '97 Season 2 is set to premiere on July 1st.

The second season of the Marvel animated series "X-Men '97" will premiere on Disney+ on July 1st this year. The official poster and trailer for the second season were released yesterday.

The second season consists of nine episodes and continues the storyline of the first season. Set in the 1990s, it tells the story of a group of mutants with superpowers protecting humanity in a hostile world.

The first season of X-Men '97 premiered on Disney+ in 2024 and became one of the most-watched animated shows on the platform, earning an Emmy nomination.

The Pittsburgh Medusa has been renewed by HBO for a third season.

According to Deadline, HBO's medical drama "Hyperpixburgh" has been renewed for a third season before its second season even aired. The second season, which aired from January to April of this year and consisted of 15 episodes, will continue to focus on emergency room doctor Robby and the hospital's emergency system.

"Pittsburgh Medical Soul" focuses on real-time and high-pressure emergency room scenes. Each season presents the hospital's internal cases, resource allocation, and character relationships through a series of consecutive shifts.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

In the past decade, almost no supercar has generated as much discussion as the Ferrari Luce today.

Not only because it's Ferrari's first-ever all-electric car —a supercar manufacturer with performance, design, and style etched into its DNA—but also because of its indescribable appearance :

Image | Ferrari

In Italian, "luce" means "light," which not only symbolizes that this electric car illuminates the way forward, but also adds a layer of religious sanctity to Ferrari's first all-electric vehicle.

God said: "Let there be light!"

The Luce is Ferrari's first 5-seater model, powered by four electric motors. Its 122 kWh battery pack provides a range of approximately 530 kilometers on a full charge and a peak power of 1,050 horsepower .

Even with a weight of 2.2 tons (close to the Xiaomi SU7 Max), the Luce still boasts a top speed of approximately 310 km/h and a 0-100 km/h acceleration time of around 2.5 seconds .

Pope Leo XIV visits Ferrari Luce | The Verge

In terms of price, Ferrari announced that the starting price of the Luce is 550,000 euros (approximately RMB 4.329 million), which is in line with the prancing horse logo on the front of the car in terms of both performance and price.

However, it was this prancing horse logo that brought Luce a massive online debate after its official debut.

impure horses

Even before the KOLs and media test drives, netizens were already generating a constant stream of creative memes and images.

Fiat Luce | Threads

Fiat Multipla Luce | Threads

Pure "vehicle" body attack | Threads

However, just two months ago, when Ferrari announced that the interior of the Luce was designed by LoveFrom, led by former Apple design chief Jony Ive, the general consensus was quite positive:

Image | TopGear

But when people saw the Luce, designed by legendary industrial designers Johnny Ivor and Marc Newson, their opinions took a 180-degree turn, which seems understandable.

The Luce is a very nice tram, but it's definitely not a nice Ferrari.

Image | Ferrari

In short, Luce is a car that pursues the ultimate in tram design: an integrated cockpit, extremely low wind resistance, and coach doors inherited from the Thoroughbred, among other features.

However, the Luce is different not only from hybrid Thoroughbreds, but also from all of Ferrari's previous design languages—

It lacks the roundness of a classic Ferrari, the squareness of a traditional Ferrari, and the sharp lines of a modern Ferrari; instead, it resembles a consumer electronics product.

Ferrari Purosangue 2026｜TopGear

To make matters worse, right after the official announcement of Luce, Italian media outlet Askanews interviewed former Ferrari chairman Luca Cordero di Montezemolo and asked him for his opinion on Luce.

The 78-year-old, who served as chairman of Ferrari for 23 years and oversaw several legendary models including the Enzo, 458, and LaFerrari, gave a rather helpless answer:

If speaking my true thoughts would hurt Ferrari, we are risking destroying a myth, and I feel very sorry… Hopefully they will at least remove the prancing horse emblem from the Luce.

Image | Askanews

Clearly, Montezemolo couldn't hold back in the end and added a comment before leaving the camera's view:

…Even Chinese car manufacturers don't want to copy its design.

If you calm down and look at the unconventional placement of the Ferrari logo on the Luce's official website, you'll understand why the Luce received such criticism—it's not surprising at all.

Luce is a Ferrari that needs to alleviate its identity crisis by "putting a prancing horse emblem on the side door".

Image | Ferrari

However, is the problem really with the "Ferrari" part of the Ferrari Luce? Actually, no.

As Montezemolo said, simply replacing the prancing horse emblem on the Luce with a more iconic logo instantly makes the Luce's design make sense.

Image | Threads

Of course, netizens have also come up with solutions to the most important problem with that tram:

Image | Threads

Luce looks like iCar

Ultimately, the fundamental reason why Luce's design stands out from the rest of the Ferrari family is that it is Ferrari's first all-electric car.

For a product that is "unprecedented" in its product line, Ferrari most likely never intended for the Luce to resemble conventionally powered models in appearance from the very beginning .

This viewpoint can be indirectly corroborated by Ferrari's official website's description of the Luce's design:

"What we're developing with LoveFrom is far more than just an electric Ferrari," said John Elkann. "What we're doing here is not only extremely difficult in conception, but also incredibly challenging in implementation. This will create a truly unique Ferrari Luce."

Image | Ferrari

This raises a question:

Designed by Johnny Iver, Marc Newson, and LoveFrom Studio, is the Ferrari Luce a more consumer-oriented Ferrari of the future , or a consumer product that resembles a future Ferrari ?

As we can see, after the interior design of Luce was announced in February, everyone praised its pioneering design that combined screens and physical buttons, but after seeing its equally unconventional design, it was met with a barrage of criticism.

Image | Ferrari

This kind of evaluation is inherently contradictory when applied to the same car: you can't simultaneously expect Luce to be both avant-garde and conservative in its design philosophy.

Take the Thoroughbred for example. Although it is a hybrid car and retains the sharp design style of modern Ferrari, it would still feel out of place to transfer the intelligent interior of the Luce to the Purosangue.

Ferrari Purosangue Interior | TopGear

At the same time, this discussion about the nature of Luce products also touches on that ultimate question:

Who exactly is the Ferrari Luce designed for?

550,000 euros or 4.3 million RMB—this figure alone excludes 99.9% of the world's inhabitants.

Luce's target customers are those whose garages are already full of other luxury cars:

Image | Rockstar Games

Although Ferrari has not yet introduced the purchase model for the Luce, we know from past experience that Ferrari will conduct background checks on potential buyers before granting them eligibility to purchase the car.

According to statistics, of the nearly 14,000 people who will buy Ferraris globally in 2025, more than 80% will already own a Ferrari.

While we joke about how the Luce needs to be flipped over to charge and has less load-bearing capacity than a Fiat, customers who already own a bunch of V12 engines might just place an order simply because "the Luce looks different from traditional Ferraris."

Image | Car Magazine

After all, as an electric vehicle, the Luce's exterior design primarily pursues a low drag coefficient, which is not entirely the same as the design specifications of traditional sports cars (and FUVs) that prioritize downforce.

Furthermore, the chassis structure and center of gravity distribution of the tram also determine that, after removing the space for the engine and gearbox, the cabin area will eventually present a teardrop-shaped outline, no matter how it is designed.

Image | Ferrari

This also explains why Luce doesn't look like a Ferrari, but rather resembles a domestically produced new energy vehicle:

When designing real-world products, the optimal solution determined by physics is sometimes the only solution, and the pursuit of the ultimate often leads to convergent results.

Are there any other options?

No matter how much car enthusiasts think the Ferrari Luce is "forgetting its roots," the fact remains: its appearance is already finalized, and the Luce is there, ready to catch you.

Image | Threads

However, we can't help but ask: even with Johnny Ive personally in charge, is Luce really the optimal solution for a "Ferrari electric car"?

The answer is clearly no—while you can't expect a product design to be both avant-garde and conservative, there are ways to design a product that is both avant-garde and retains the iconic brand DNA.

One of the best examples of "avant-garde iconic design" is the modern N Vision 74.

Image | Threads

Although the N Vision 74 and Luce are positioned very differently, they face highly similar design challenges:

How can a completely new design provide a distinctive new model for the brand's development of new powertrain technologies (hydrogen fuel cell vs. pure electric)?

The N Vision 74, which drew inspiration from the 1974 Hyundai Pony concept car, is clearly more widely accepted than the Luce in conveying innovation and brand style.

In an interview with tech blogger Cleo Abram, Ferrari's design director Flavio Manzoni, when commenting on the public's reaction, quoted a line from Austrian composer Gustav Mahler:

Tradition is not the worship of ashes, but the inheritance of the flame.

In other words, from Ferrari's perspective, Luce, even if it no longer inherits the iconic design of modern Ferrari, still carries on Ferrari's spirit of innovation and technology.

Enzo Ferrari | Camisasca Automotive Manufacturing Inc.

However, "inheritance" is never an easy thing. Only when the brand's genes are inherited well can it be called inheritance. If it is not inherited well, it will become like New Coke back then, which is called a disaster.

Therefore, Cleo Abram made a rather fair summary at the end of the interview video:

We don't know if Luce will ultimately succeed or fail, but either way, we can learn something from it.

If Luce fails, we can learn how to improve; if Luce succeeds, we will see the entire industry start to learn from Luce, to the point that it will affect the electric cars we can buy in the next decade.

As for domestically produced new energy vehicles, Montezemolo's words are still true—

There's no need to copy the exterior design; just learn from Luce's interior design.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

After using the OPPO Reno16 Pro for a week, I feel like it doesn't look like a phone from 2026.

It's not that it's backward. Quite the opposite, it's that it's too unrestrained.

Since the beginning of this year, the cost of consumer electronics has been steadily increasing: storage, chips, and batteries—each is an unavoidable expense. For mobile phone manufacturers, the options are actually not complicated: either "reduce prices and reduce features," or "slightly increase prices but not features," or "small increases." Products that actually choose to "increase features and raise prices" are relatively rare.

The OPPO Reno16 Pro is a bit different; it doesn't feel cramped. This year, it's a rare "relaxed" phone. "Relaxed" doesn't refer to its size, but rather that while many products are tightening costs, Reno is still willing to include the best features and present the increased price in a visible way.

After all, "no significant regression" should not be a point of celebration for the mobile phone industry today.

The OPPO Reno16 Pro doesn't feel like a phone from 2026.

The Reno product line answers a very interesting question:

How to make a phone that mainstream consumers like?

This means that Reno cannot be haute couture or a conceptual item; it should be more like a trendy new product of the season—not necessarily at the forefront, but the aesthetic must be novel and the style must be popular enough.

In recent years, the definition of a "good phone" in the eyes of consumers has gone beyond simply being "easy to use" in terms of functionality. With similar configurations and prices, the emotional value brought by "fun" has become more important.

This is precisely Reno's comfort zone. Upon holding the OPPO Reno16 Pro, it's hard not to be drawn to the back of the phone: the Reno16 series uses "stars" as its theme, employing a 3D ice-clear floating process to achieve a naked-eye 3D visual effect on a flat glass surface, with planetary textures appearing to float in the air.

Mobile phones have become our "life partners". If something is going to accompany you for four years, it's best if it doesn't just have good features, but also some personality and warmth.

Since the very beginning of the Reno brand, it has been associated with this kind of temperament.

Rather than presenting the phone as a spec sheet, Reno has a more "soft" product strength, aiming to make you "like" it rather than "choose" it.

The Reno is like a Mini Cooper; it's not the fastest car, but most people buy it simply because it makes commuting, this "daily routine," more enjoyable.

Among the many new imaging features of the OPPO Reno16 Pro, the POP camera that can take instant photos and film shots in a second, and the ability to create collages of real-time photos, are undoubtedly the features that best embody the "Reno" spirit.

However, it would be too simplistic to define the OPPO Reno16 series solely as "fun".

Design and functionality determine whether a user will pick it up, but the user experience determines whether a user will want to keep using it. While more and more brands are busy teaching users to make trade-offs, the OPPO Reno16 series chooses to continue to amplify its advantages.

Even in the current climate of cost pressures in the industry, the OPPO Reno16 Pro still features a 200-megapixel main camera, maintaining the characteristics of the Reno series and, more importantly, continuing to serve Reno users who "love to take great photos"—they may not be keen on studying sensor size or pixels, but they definitely care about how good the photos they post on social media look.

Specifications can only remain at the press conference, while the success rate of photos can remain in the user's memory. Configurations that live on a PowerPoint presentation can never compare to photos that live in an album.

In addition, the OPPO Reno16 Pro adds a "gimbal" capability to this 200-megapixel main camera, extending the boundaries of Reno's imaging capabilities to equally youthful formats such as Vlog and live streaming.

These features and configurations are essentially core elements that continue to serve the "Reno" brand. Choosing to "invest" more in 2026 will inevitably lead to increased costs, but ultimately it will all serve the "Reno" brand, which already enjoys user trust.

After all, users are willing to pay for progress, but rarely applaud compromise.

Fun and easy to use, the OPPO Reno16 series remains a phone that is "pleasing at first sight and never gets old." While more and more phones are teaching users to accept regrets, those that strive to reduce regrets stand out.

But Reno is more than just a phone.

Reno is no longer just a phone

From the very beginning of the Reno brand, it has been synonymous with "high quality." This brand image is now expanding, and OPPO hopes it will carry more than just products—it also includes a more complete aesthetic expression and user recognition.

Along with the OPPO Reno16 series, the "Reno Land" brand ecosystem was also launched, which includes tablets, headphones, and the "electronic bar" OPPO Bubble.

The OPPO Bubble defies easy categorization into any traditional digital product category; it's more like a trendy accessory with a screen that attaches magnetically to the back of the phone.

Just like young people hang badges on their backpacks, put stickers on their water bottles, and apply them to their computers, OPPO Bubble is essentially a form of identity expression, a private space for self-expression.

When needed, it can also serve as a viewfinder for selfies with the rear camera, making full use of the 200-megapixel main camera and strobe flash.

Reno wasn't the kind of product line that just focused on maximizing specs. It needed a touch of unnecessary cuteness, some details to showcase, and something that young people would be willing to play with. OPPO Bubble isn't the core selling point of the OPPO Reno16 series, but it's like a glimpse of Reno's personality.

Whether it's the OPPO Reno16 series itself or Reno Land products, when you see them, you can immediately think of the Reno brand, the vitality, aesthetics, and emotional values ​​it represents, as well as a lifestyle that grows in sync with the times.

For hardcore digital enthusiasts, Reno's proposition might be a bit unfamiliar, but in the broader consumer sector, truly successful consumer brands ultimately sell more than just the product itself.

For example, Harley-Davidson motorcycles don't just sell expensive motorcycles; they sell a slightly "rebellious" spirit.

What we're selling is the ability to make a 43-year-old accountant put on a black leather jacket, ride a motorcycle through town, and instill fear in people.

The product is just a vehicle; the sense of belonging is the ultimate goal.

What kind of products would I be willing to spend money on in 2026?

The OPPO Reno16 series is not an isolated incident.

Before it, the OPPO Find N6 became another blockbuster foldable screen product with its seamless crease, and the imaging capabilities of the OPPO Find X9 Ultra and Find X9s Pro helped OPPO's sales continue to rise in overseas markets.

These figures precisely demonstrate that good products will find buyers in any price cycle.

Therefore, the question that the OPPO Reno16 series answers is: what should mid-to-high-end mobile phones do in the face of rising prices?

Consumers' replacement cycle has been extended to four years or even longer. Every time they spend money, it's like a vote. "Is it worth it?" has replaced "Do I want it?" as the key to purchasing decisions.

Any shortsightedness will erode years of accumulated reputation; any compromise will affect the quality of user experience; any concession will crack the brand's trust.

The problem isn't raising prices, but rather making customers feel cheated while raising prices.

Because everything that's cut will ultimately be recouped from the user's trust.

People are becoming more rational in their consumption, but that doesn't mean they only buy "cheap things." There are still long queues for luxury goods, concert tickets are still hard to come by, and high-end coffee hasn't disappeared.

People are simply starting to recalculate value, wanting to pay for tangible experiences, for lifestyles they identify with, and for attitudes and individuality. At the same time, they are increasingly unwilling to pay for "downgrades" or compromise.

For users, memory price increases will continue for another two years. Choosing a good phone to get through this difficult period is a practical and realistic decision.

Behind this is not the simple transaction logic of "you get what you pay for," nor is it a narrative built on parameters and prices; it is a longer-term "value commitment" between manufacturers and users.

In an era where everyone has to simplify, defending "price" is not so difficult, but defending "value" is not easy.

It is precisely this sense of trust in upholding values ​​that makes us willing to choose a brand and a product again, because we know that this time will not disappoint us either.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.