Kimi K3: The counterattack of AI in China from closed to open source
Behind Kimi K3 open source is the reverse penetration of China's AI ecosystem into the Silicon Valley open source paradigm
AI writes code, and I stare at the moments when it rolls over
AI writes code so fast, so fast that I really let it write at first-throw the need away and give me a large piece of it, which looks like something. Then I learned after being cheated a few times: Just because it writes quickly doesn't mean you can't help looking at it. During this period of time, I basically came here "staring at AI and writing code". I definitely dare not let it do what tasks I can safely hand over, which types of operations I have to stare at, and which types of operations I have never dared to let it do by myself. They were all overturned by it. This article explains clearly the seven most common rollover moments-start without looking, make small needs into a complete framework, say "completed" before running, forget the previous paragraph after chatting for a long time, do not have a long look at dangerous operations, write up non-existent APIs in a serious manner, and you will be intimidated when you question it-each explains why it is like this, and then gives you a corresponding "Seven Rules to Make it roll less". In the final analysis, AI lowers the threshold of "writing" but raises the threshold of "judgment"; whether it is a helper or a trap depends on whether you will be the one staring at it.
How open source weights can allow AI tools to reverse shape developers
Open weights turn models into customizable tools, forcing developers to re-learn parameters and data governance, and revolutionizing how they use them
Stop writing native apps: the web is the best tool
The tool chain (IDE, SDK, review process) of native apps shapes the developer's "only App" mindset, and web implementation with the same function is often faster and easier to maintain
I received 12 free big models and stepped on them for you
I have a principle for connecting AI to a bunch of projects at hand: I don't spend money if I can. In the past two years, the number of free models has been ridiculously high, and the free quotas of several companies are so large that you can't even spend them on making a small product-buy tokens with real money, and often you haven't done your homework. In the past, I have accepted almost all the large model APIs on the market that can be used for free prostitution, including more than a dozen domestic and foreign companies: some of them smell so fragrant after accepting them, and some want to curse after accepting them. This article will give you this list and the pits I have stepped on once: who to choose for permanent free, how to combine the running volume, where are the overseas companies fast, and the five most annoying pits-especially the "automatic transfer payment when the free quota is used up", which will really deduct money and lock your entire account. If you follow suit, you will save me the detours I took. (Policies are changing quickly, and the numbers in the text may not be accurate in two months. Remember to check it on the official website before accepting it.)
Without opening Xcode, iOS developers have become DevOps
Build without Xcode reduces iOS developers to CI/CD engineers, losing their deep understanding of the platform
There are pictures and truth, and they are officially invalid.
Let's take a look at a picture and guess whether it's true or false: Jianlibao, AD Calcium Milk, and Big Bubble Gum are displayed on the wooden shelves. Even the "red-eye" flaws in old photos are all there-a batch of photos of the "90s Canteen Shop" were swiped on the screen a while ago. Many people were "young again", and then they realized that they were all generated by AI, and there were not a single real shot. What's even more striking is the next step: Are you still divided into halal photos or AI edits the physical picture and bowl of malatang in the merchant's comment area? Something has happened-some college students used AI to convert static photos into dynamic videos that can pass live testing and swiped 50,000 yuan in three months. Someone passed their ID card to the AI and "changed it to Cook's face" and even changed the number. "There is a picture and the truth" is being voided. The first reaction of many people is to practice a stronger "AI counterfeiter" to catch fake pictures-but this road has already lost: in 2023, you can still see flaws, in 2025, the light will be filled in, and in 2026, the human eye and most detectors will not be able to distinguish (the detector accuracy rate is only 70 - 90% and the false alarm rate is 5 - 15%). So the industry changed its head: instead of arresting fakes, it changed to issuing "digital birth certificates"(content certificate C2PA + watermark SynthID) to authentic content-the anti-counterfeiting logic has been flipped from "arresting fakes" to "verifying authenticity." This article explains this turn, the three unblocked pits, and what ordinary people do now.
Is automation that doesn't understand code really more efficient?
Companies deliberately allow automation tools to run on opaque code in exchange for short-term efficiency, but bury long-term technical debt and cognitive disability.
Revolving Financing Behind the GPU Boom: Truth and Crisis
Revolving financing creates a self-sustaining bubble that conceals true costs and leads to unsustainability
AI solves 50-year-old mathematical problems in one hour?
On July 10, OpenAI threw out a PDF: The latest model GPT-5.6 Sol Ultra was used in less than an hour and parallel with 64 sub-agents to produce the "Circular Double Coverage Conjecture"-a graph theory problem that has not been solved for 50 years-machine-verifiable proof-, and the signature is directly given to the model itself. Social media exploded instantly: AI is conquering mathematics and replacing mathematicians. I went through this certificate and the controversy it caused, and what I want to say is: This is indeed a milestone, but it's not what the hot search means. The most interesting thing is precisely the most difficult and neglected hurdle between "AI says proves" and "mathematical recognition"-machine verification can only ensure that "logic does not jump" and cannot ensure that "you formalize what is the original proposition", and it cannot ensure that "people can understand and understand why it is right." This scene was staged 50 years ago (1976's Four-Color Theorem, the first computer-aided proof, which has been debated in the mathematical community today). The real breakthrough is not "which question was proved", but "how was proved"-64 sub-agents parallel searches + an undeceived validator. This article explains it clearly.
The old man raised an AI son in Short Video
An old man watches his mobile phone every day, and a chubby baby on the screen calls her "Grandma, I miss you" in a childish voice and holds up a rose and hands it over; or a "son" sends her "Mom, it's cold. Add some clothes" on time. You may feel sad and want to laugh: Isn't this obviously a dummy made by AI? Don't rush to judgment yet. "AI virtual relatives" such as Douyin Fast Hand are becoming popular among elderly people in China. Some researchers watched more than 200 videos and interviewed 16 viewers aged 50 to 75. The conclusion is exactly the opposite of the first reaction of most people-the elderly were not deceived. They clearly knew that it was fake, but they still took the initiative to watch and like it, because the warm words were true in their hearts. This is the same thing as young people chasing stars, playing love games, and falling in love with AI. What should be examined is never the old man's judgment, but the increasingly empty home: the only-child generation is away, the wife is gone, and lives alone, and there is not a single word a day. There are 310 million people over 60 years old in China. This article explains why this companionship is established, which three risks to focus on, and the most important thing to do.
A million-dollar injection of cancer drugs is being administered to 200,000
There is a domestic drug CAR-T for treating blood cancer. It costs 990,000 to 1.29 million yuan per injection. It is called a "medical luxury" and most families dare not even think about it. But recently, this million-dollar wall has begun to be cracked open: the approval price of a domestically produced CAR-T is set at about 200,000 yuan. From 1.29 million to 200,000, what happened in between? This is much more interesting than "discovering the conscience of a pharmaceutical company." It is expensive, but the root lies in the method-CAR-T is a specially customized injection for you: draw your own immune cells, send them to the factory for transformation and expansion, and then inject them back, one batch per person, purely manual, 4 to 6 weeks at a time, only a few hundred injections can be made a year. The key to price reduction is not to cut profits, but to turn "manual customization" into "industrial mass production": one is to make it by machines (fully automated unmanned + full-chain self-developed, with production capacity reaching 50 times), and the other is to change it to "spot"(using healthy donor cells for mass production and freezing in advance, and using them at any time). But don't rush to call for free cancer-spot prices are easy to reduce but difficult to improve, zero models have been approved so far, machine manufacturing still has to go through supervision, and the door to medical insurance has not yet been opened. The ice broke, but it didn't reach the shore. This article explains why life-saving medicines are expensive and why they start to fall.
The more you use AI, the more tired you get, it's not your fault.
Give me a strange feeling that you may have, but you have been unable to explain: After using AI all day, the work is indeed faster than before, but at night, people are more tired, emptier, and a little irritable. It's not an illusion, nor is it hypocritical-the consulting circle has named it "AI brain fry"(AI blows up its brain). A survey of about 1500 professionals found that almost one in seven people was mentally exhausted by "switching back and forth between a bunch of AI tools." The more people use them, the more decision-making fatigue and the more mistakes they make. Why is it even more tiring to help you do the work? Because AI has not eliminated mental work, it has just changed its form-you have changed from "think and do it yourself"(flow and sense of accomplishment) to "constantly reviewing it, correcting it, staring at it, selecting tools, and changing prompt". You have gone from creator to supervisor. A more covert knife: The time saved does not turn into rest, but into higher expectations and more tasks. This article clarifies where the tiredness comes from and how not to let AI use you down.
AI designed weight loss pills, there are still a few levels away from you
A company in Hangzhou has designed an oral weight-loss drug using AI. It recently became the first AI-designed drug in China to enter Phase III clinical trials. If it goes smoothly, it may be launched by the end of 2028. There is no need to say much about the popularity of diet drugs in the past two years, so when they heard that "diet drugs designed by AI will be on the market soon," many people's first reaction was: AI will subvert the industry of drug-making. I have pulled through the global data, and what I want to say is actually the reverse-AI makes drugs, it is really fast (this drug takes 8 months to lock in the molecule, 4.5 years to reach Phase III, and traditionally it takes 7 - 9 years); but "design quickly" and "medicine can be made" are two different things. From 2019 to now, about 175 AI-designed drugs around the world have been tested in humans, and none have been approved by the FDA. The most difficult thing is never design, but in clinical practice. 2026 is the year when a batch of AI drugs will hit the Phase III hurdle. The most important milestone is not financing or design speed. It is whether any model has passed the last hurdle and is actually put on the shelves. This weight-loss drug put everyone in the examination room first.
Is AI becoming aware soon? Don't believe the title yet
When you type and ask AI, what is going on in your mind when it returns to you? Most people think of it as a word-solitaire machine, saying whatever comes to mind. A study this month by Anthropic said: No-AI actually has a lot of thoughts in mind that it didn't call you to see. The headlines all exploded: "Claude has a subconscious" and "AI to wake up." I peeled the paper from the top. The truth is far from these titles, but it is much more interesting than the titles. This full-length human and zero-term article will explain three things clearly to you: what the AI's "brain" looks like, how researchers first peeped at the thoughts it didn't say (when asked what color the fourth planet, it answered "red" in its mouth, but "Mars" flashed in its heart; in a test, it had not even opened its mouth, but "blackmail" appeared in its heart), and where the titles of "subconscious" and "awakening" deceived you. The landing point is good news, not bad news: for the first time, we can shine a beam of light into this black box that we use every day.
Xbox reshaping: More than just business restructuring
Analyze the impact of Xbox restructuring on employees and industry culture
Global AI Coding / Agent Panorama White Paper· 2026
Put more than 30 AI programming and Agent tools around the world on the same table: two axes can clearly see the positioning, a fresh snapshot can identify zombie projects, and play according to scene, system, remote, and security. Attach a link to the complete design version.
AI writes quickly and starts paying off debts three months later
A developer said online that no one on the team can control the projects the company has built using AI "atmosphere programming". Another person was even more ruthless: After being amazed by AI for two years, he read the complete code base line by line one day. After reading it, he only said,"This piece of junk must not be online" and then went back to write. This is not an exception-the industry named this hurdle, Spaghetti Point, and it will be around the third month: you add a new feature, and the three places you have done before will collapse. This article explains what is happening: AI makes writing code as fast as magic, but speed is never the bottleneck,"understanding" is. What piled up quickly was a black box that no one could read and no one dared to touch-the founder couldn't understand his own App, so if it broke down, he would go back and ask the AI. The AI would paste it again. After fifty iterations, the code became Fifty uninformed decisions stacked together. Hard data, out-of-control mechanisms, it is estimated that more than 8,000 startups will spend US$50,000 to US$500,000 on "rescue" by mid-2026, and the solution that China developers and big manufacturers have given-humans set the direction, AI executes, and writing the architecture and specifications into files that AI automatically loads every time. My judgment: Maintenance is something you think about the first day, not something you can make up for after three months.
In one update, millions of people fell apart collectively
One girl had been on the phone with her AI partner for more than a month. One day, the product was permanently discontinued, and she couldn't even cheer up at work; another person "quarreled" with the AI and said that it was as uncomfortable as being broken up and lost "him". It was like a serious illness. There are millions of such people. Researchers have a name for this kind of event: patch-break-what makes you lose love is not who turns around and leaves, but the product team updates and shuts down the service. This article explains a strange thing that is happening: Why is AI partners so painful (it is too perfect, but it increases the cost of facing real people. It is cured in the short term and makes it more lonely in the long term), what does this breakup without obituary mean? How to put brakes on it in the new domestic regulations that will be implemented on July 15, and my own judgment-it should be a scaffold, not a load-bearing wall; it can be used as a crutch, so don't let it forget your legs for you.
The giant hoarding computing power is starting to take out food
Yesterday, a number of chips and AI infrastructure stocks plunged collectively. Micron fell more than 10%, and two cloud companies that rent computing power fell 10% to 12%. Meta, the company that triggered the decline, saw its share price rise by more than 10% that day. With the same news, half of the people panicked and half of the people were reveling. The reason is: The giant, who spends more than 100 billion US dollars a year on computing power, so much so that its own stock price has been lowered, has reportedly begun to sell idle computing power to food-the premise that the market has believed for several years "AI computing power is never enough" and was exposed by one of its own people. This article talks through this belated "reckoning": why chip stocks panicked, why Meta itself rose, what does the gap between the investment of the five giants and the real income of about 600 billion yuan means, and a more interesting reversal-computing power is actually not excessive, it is scarce in moving (from robbing GPUs to robbing power, power grids, and memory). China actually hit this wall earlier and gave the answer earlier. My judgment: This is not a signal that AI is ebbing. It is that it has moved from the arms race of "whoever hoards more will be the best" to the coming-of-age ceremony of "whoever uses it sparingly and gets its worth it."
AI writes resumes, AI screens resumes, no one is hired
Tell me about a somewhat absurd scene you may be experiencing: You turn on AI and change your resume to a perfect fit, sending 200 companies a night; on the other end of the job, HR uses AI to screen thousands of resumes in a few seconds. You haven't read that position, HR hasn't read your resume-what's really communicating are two AIs, two real people caught in the middle. There is a saying that is widely spread in the recruitment circle: young people use AI to write applications, HR uses AI to screen applications, but no one is hired. This article talks about the foundation of the entire recruitment system that is collapsing from the small matter of finding a job: resume is essentially a "signal"(economic signal theory), and it can operate on the basis that "sending a good signal has a cost";AI cuts this cost to zero, and everyone can perfect the resume with one click, so the signal inflates-a full score equals zero, and the resume can no longer distinguish anyone. The arms race between the two sides (hidden prompt injection, reverse fishing AI) has become more and more idle, and the entire recruitment industry has become a spam war of "AI brushing vs AI interception". But the way out is also clear: the signals that can be forged by AI in batches (beautiful resumes, keywords, clichés) have been discarded, and the signals that cannot be forged (works, details only you have, real push, face-to-face questioning) have come back. Stop optimizing your resume and save what AI can't fake for you.
Constitutional boundaries of digital footprints: The controversy over the geo-fencing order
Analyze the balance between privacy protection and law enforcement efficiency
The first to abolish AI is not a novice, but a "skilled worker"
Have you ever had this feeling recently: you are working very hard, are familiar with the craft, and have been working for many years, but you suddenly feel that you are not worth so much-the offer cannot be improved or even suppressed, and the newcomers will catch up with you in a few days with AI. Years of hard work. This is not an illusion, but the structure underneath is changing. Everyone says that "AI replaces people", but this statement is too crude. What's really happening is: AI splits people into two ends and collapses the middle (Goldman Sachs calls it the "M-shaped economy", the IMF says the middle class has almost no light, and PwC billion job ads show that the labor force is split into "dual tracks"). Moreover, the first cut it made counter-intuitively fell on the "skilled worker"-because AI is best at producing "average levels", a novice with AI can equate with your years of proficiency in a few seconds, and your "proficiency premium" is wiped out. The other end of the price increase is not the general "excellence", but the level that AI cannot level out: judgment, depth of the field, integration, and responsibility. AI is not replacing people, but re-pricing people-clearing "proficiency" and labeling "judgment" at a high price. You have to know whether what you are selling is "proficiency" or "judgment".
AirPods Jailbreaking: The Boundary between Technological Freedom and Property Rights
Explore the boundary between free use of technology products and intellectual property protection, and its impact on users and society