DEV Community: hummingbirdLabs

AI Era: Unemployment and Wealth Explosion Coexist for Two Different Groups

hummingbirdLabs — Fri, 05 Jun 2026 03:35:00 +0000

On one side, a handful of people are raking in cash like crazy. On the other, more people are stressing about paying the bills. This is the parallel world of the AI era.

Two Stories on the Same Day: Fire and Prosperity

On June 1, 2026, two things happened in South Korea.

At 10:32 AM, a gas room suddenly caught fire at SK Hynix’s Cheongju Campus 4. Hydrogen fluoride leaked, and 3,600 people were evacuated in a panic.

Almost at the same moment, the Korea Composite Stock Price Index (KOSPI) broke through the 8,600-point mark for the first time ever, with total market capitalization exceeding 7,000 trillion won (~$3.2 trillion). Hynix’s stock surged—this company had just hit a $1 trillion market cap a week earlier, becoming Asia’s third tech company to cross that line (the first two being Taiwan Semiconductor Manufacturing Company and Samsung Electronics).

One side: a terrifying moment in a factory. The other: a wealth feast in the capital markets. These two images stitched together—that’s the real picture of the AI era.

The “Sweet Trouble” of Hynix Employees: Behind the $850,000 Average Bonus

Just two weeks before the accident, South Korean media blew up: SK Hynix employees might get an average bonus of $850,000.

That number came from a prediction by investment bank Macquarie Securities: if Hynix’s 2027 operating profit reaches 447 trillion won, with a 10% dividend, 35,000 employees would each get 1.29 billion won—about $850,000.

This isn’t some fairy tale. Hynix holds 57% of the global HBM (High Bandwidth Memory) market, and NVIDIA’s entire line of AI accelerators uses its chips. In Q1 2026, its operating margin hit 72%—for every $100 it sold, $72 was profit. This isn’t making money. This is printing money.

There’s a new buzzword in Korean workplaces now: “海医齿韩” (Hae-ui-chi-han)—Hynix, medical school, dental school, Korean medical school. A job at Hynix now ranks above traditional elite professions like medicine.

Chosun Ilbo reported a detail: the night Hynix paid bonuses on February 5 this year, restaurants around the Icheon Campus in Gyeonggi Province were completely packed. BBQ restaurant owners said they prepared in advance but still couldn’t keep up. Two apartment buildings within 50 meters of the campus—one was sold out, the other had an 85% contract rate. Along the commuter bus routes, housing prices in multiple areas were skyrocketing, forming “bus stop districts.”

Even the dating market changed. Koreans used to prefer people who worked at Samsung, Hyundai, or LG. Now Hynix employees are hot commodities, as sought-after as doctors and lawyers. One variety show joked: when Hynix employees go on blind dates, they first pretend they work at Samsung, and only admit their real employer when they meet someone good.

But what about the other side?

Because bonuses are tied to attendance days, employees actively avoid maternity leave. One married couple said: “If there’s 500 million won in bonus waiting, taking six months off means losing 250 million—you just can’t do it.”

And right next to this wealth feast? Safety accidents that could happen at any time in the factory.

Unemployment and Wealth Are Stories of Two Different Groups

When many people talk about AI, they either fall into “tech optimism”—AI will create more jobs, everyone will benefit—or “tech pessimism”—AI will replace humans, most people will lose their jobs.

I say: both views are wrong.

The truth of the AI era is this: Unemployment and wealth aren’t contradictory. They belong to two completely different groups.

On one side: people who master core AI technologies and control key supply chains—they’re accumulating wealth like crazy. Hynix is the perfect example: with its monopoly in HBM, it has a 72% profit margin, and employees are drowning in bonuses.

On the other side: people whose jobs are easily replaced by AI, or who’re struggling in traditional industries—they face stagnant incomes, or even the risk of unemployment.

These two groups seem to live in the same country, the same era—but really, they’re in two parallel worlds.

While Hynix employees are buying houses and cars with millions in bonuses, young people in other Korean industries might be stressing about finding a stable job. While Hynix’s market cap breaks through a trillion dollars, South Korea’s wealth gap is widening to an unprecedented degree.

This is the brutal truth of the AI era: the wealth growth from technological progress isn’t evenly distributed. It’s like a spotlight—only illuminating a few, leaving more in the dark.

South Korea’s Government and Society React: Finding Balance in Anxiety

Faced with this situation, how did South Korea’s government and society react?

1. National Level: Betting Everything on the AI Industry Chain to Become an AI Upper-Class Country

Back in 2025, the South Korean government set up a “Cutting-Edge Strategic Industry Fund” of up to 50 trillion won (~$34.67 billion) to boost semiconductors, AI, and other advanced industries. It also rolled out new visa policies to attract top talent.

Why the rush? Because South Korea knows all too well: in the AI era, seizing the core industry chain means seizing wealth. Hynix is South Korea’s “national treasure”—it not only creates tax revenue, but also serves as South Korea’s bargaining chip in global tech competition.

2. The Public’s Wealth Takeoff Dream: From “Degree Worship” to “Production Line Worship”

When news spread that Hynix employees might get an average bonus of $850,000 and Samsung Semiconductor employees about $380,000, South Korean society completely erupted.

There’s a term now in Korean workplaces: “kingsanjik”—referring to frontline technical workers at Hynix and Samsung Semiconductor who’re getting huge bonuses. T-shirts with the Hynix logo are being auctioned online, and these employees are super popular in the dating market.

A joint survey by Seoul National University and Chosun Ilbo showed: 79% of South Koreans worry AI will widen the social gap , and 55% think AI development will deepen socioeconomic inequality.

And reality is confirming this anxiety. Data from the Bank of Korea shows: in the three years since November 2022, employment of young South Koreans in white-collar positions like information services, publishing, and professional services dropped by 23.8%, 20.4%, and 8.8% respectively. In 2025, only 338 out of 1,200 CPA (Certified Public Accountant) passers in South Korea successfully entered the internship registration process.

Young people’s career choices are changing dramatically. In 2026, the number of students choosing to give up four-year universities and go to vocational schools hit a record 2,500—up 23% year-over-year. A survey of South Korean Gen Z job seekers showed: 60% would rather take a production job paying 70 million won a year that requires shift work, than an office job paying 30 million won a year with no overtime.

A phenomenon is emerging in South Korean society: “fate divergence of the same generation.” On one side, semiconductor core positions are entering a super dividend cycle. On the other, junior accountants, legal assistants, and programmer positions in office buildings are being compressed by AI tools. Young people used to believe in “go to a good school, get qualifications, be a white-collar worker.” Now they find: wearing a clean suit into a semiconductor production line might be closer to middle-class life than wearing a suit into the CBD.

3. Should Corporate Wealth Be Distributed to the Country and Individuals? A Big Debate Brewing

On May 11, 2026, South Korean Presidential Policy Office Director Kim Yong-bum published a long post on Facebook, raising an explosive question: Why shouldn’t the excess profits of the AI era be shared with all citizens?

This proposal instantly ignited South Korean society. On May 12, the KOSPI index dropped from 7,999 points straight through 7,400, with an intraday drop of over 7%. The market panicked that an “excess profit tax” was coming.

Kim later clarified: the funding source would be “excess tax revenue” from the AI boom, not a new mechanism directly taxing enterprises. He referenced the sovereign wealth fund Norway established after discovering oil, proposing to turn the AI boom into long-term social assets for youth entrepreneurship funds, rural basic income, arts support, or AI-era transition education programs.

But the debate had already begun.

Supporters argue:

The prosperity of the semiconductor industry is not just the result of corporate efforts, but also of national strategic support and concentrated social resources
In 2025, South Korea’s corporate tax revenue was 84.6 trillion won, up 22.1 trillion won year-over-year (35.3% growth), mainly from the semiconductor industry
Every 1 billion won in output from the semiconductor industry only drives 2.1 jobs—one-third of the manufacturing average, with extremely concentrated wealth effects
Need to let everyone share the AI dividend through redistribution

Opponents argue:

Defining corporate profits as “excess profits” carries a value judgment, implying these gains were obtained unfairly
Samsung and Hynix achieved today’s success after decades of R&D investment and global competition—they shouldn’t be seen as “objects waiting to be reclaimed”
Countries around the world are subsidizing semiconductor companies to compete for investment, but South Korea is discussing “reclaiming excess profits”—this sends the wrong signal to the market
Should build public AI infrastructure and provide national AI skills training, not just give away money

The essence of this debate: in the AI era, is wealth corporate private property, or a “public interest” in some sense?

4. National Wealth Distribution Mechanisms: Financial Measures and Tax Reform

To let ordinary people share the AI dividend, the South Korean government rolled out a series of financial measures and tax reforms:

Measure 1: Stock Market Related Policies

Leveraged ETFs for Individual Stocks Opened : In April 2026, South Korea urgently revised rules to lift restrictions on individual stock leveraged ETFs, raising the single stock holding limit to 100%, specifically for Samsung and Hynix. On May 27, South Korea’s first two Samsung Electronics and SK Hynix 2x leveraged ETFs officially listed, raking in $2.8 billion on the first day.
Gift Tax Benefits for Minor Stock Accounts : Parents can transfer 20 million won (~$14,000) to their children’s stock accounts (under 19) without paying gift tax for 10 years. A family can give a child 140 million won tax-free through multiple transfers.
National Stock Trading Boom : By May 2026, South Korea had 105 million active stock accounts—but South Korea’s total population is only 51.6 million, averaging two accounts per person. New stock accounts for children under 18 increased by 1,000% year-over-year.

Measure 2: Tax Reform

Corporate Income Tax Hike : Starting January 1, 2026, the corporate income tax rate for large enterprises with annual taxable income over 300 billion won increased from 24% back to 25%.
Tax Benefits for High-Dividend Enterprises : For enterprises with a dividend rate over 40%, shareholders can choose to tax cash dividends separately, with the maximum rate dropping from 45% (comprehensive taxation) to 30%.
Domestic Minimum Supplementary Tax : Ensures large multinational enterprise groups operating in South Korea have an effective tax rate of no less than 15% in South Korea.
Semiconductor Investment Tax Benefits : The semiconductor equipment investment credit rate for large enterprises increased from 15% to 20%, and for SMEs from 25% to 30%; the R&D expense tax credit rate can reach up to 50%.

Measure 3: National Growth Fund

The South Korean government plans to launch a 150 trillion won “National Growth Fund” investing in cutting-edge industries like AI, semiconductors, biotech, and secondary batteries.
Investing in the fund for over 3 years can get up to 40% income tax credit, with an investment limit of 200 million won.
Dividend income from the fund is taxed separately at 9% for 5 years.

Pros and Cons of These Measures:

Pros:

Let ordinary people have a chance to participate in AI wealth growth through the stock market and funds
Tax benefits encourage corporate dividends and reinvestment
Push capital to stay in the country and support domestic industry development

Cons:

Leveraged ETFs are extremely risky and could lead to huge losses for retail investors—called a “national casino” by public opinion
Stock market prosperity widens the wealth gap: in Q1 2026, the average monthly income of the top 20% of South Korean households was 6.59 times that of the bottom 20%, hitting a nearly 6-year high
Policy goals may deviate from actual effects: young people still have trouble getting into semiconductor core positions, and the wealth distribution mechanism is still imperfect

A new term emerged in South Korean society: “霹雳穷人” (pili qiongren)—ordinary office workers who were getting by okay, suddenly finding that a colleague or friend made 10 years of their salary in one stock wave, and instantly feeling like paupers by comparison. This sentiment is pushing more people to rush into the stock market like crazy.

AI Is Roaring In—No Bystanders, Everyone’s a Participant

The Hynix story isn’t someone else’s story. It’s a microcosm of this era. Every country, every person will be affected.

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

Now I build software as an independent developer.

Why? Because the world is full of uncertainty—staying at one company too long can make you addicted to certainty. Building on your own is like sailing into uncharted waters.

I believe good software should give people a sense of security and control. That’s the thread connecting everything I make:

PhotoRestore Pro — AI photo restoration that runs 100% offline on Windows. Your photos never leave your device. No cloud, no account, no compromise on privacy. Built for legal professionals, but anyone with old family photos will find it useful.
AstroSky — Think of it as “Snapseed for astronomy.” Turn raw FITS data into stunning celestial images. Fully offline, GPU-accelerated, with a Beauty/Science dual mode that serves both casual stargazers and researchers.
fastool.io — A collection of browser-based science tools. Right now it’s focused on astronomy: solar path tracking, moon phase analysis, sidereal time calculation, telescope FOV planning—all running in your browser with zero data upload.

Whether I’m gazing at the cosmos or refining a line of code, the goal is the same: build tools that put people in control of their own data.

Get in touch: HummingbirdLabs@outlook.com.

Thunderstruck Poor vs King Workers: How AI Distributes Wealth in South Korea

hummingbirdLabs — Fri, 05 Jun 2026 03:35:00 +0000

How Close Are You to the Chip?

You think hard work leads to success? In the AI era, your income might not depend on your ability—it depends on how close you are to the chip production line.

King Workers vs. Thunderstruck Poor

In 2026 South Korea, on one side are the “kingsanjik” (king workers)—frontline semiconductor technical workers getting average bonuses of 600 million to 700 million won (~$270,000 to $320,000). Blue-collar workers in clean suits are more popular in the dating market than suited white-collar workers, and T-shirts with the SK Hynix logo are hot items at online auctions.

On the other side are the “thunderstruck poor”—ordinary office workers who were getting by okay, suddenly finding that a colleague or friend made 10 years of their salary in one stock wave, and instantly feeling like paupers by comparison.

These two terms, overnight, defined the two poles of South Korean society.

Kingsanjik: From Blue-Collar to King

The term “kingsanjik” combines “king” and the Korean word “sanjik” (production worker), literally meaning “king worker”—referring specifically to frontline technical workers at companies like SK Hynix and Samsung Semiconductor who’re getting huge bonuses.

This isn’t a myth. This is happening right now:

SK Hynix employees : Average bonus of about 140 million to 148 million won in 2025, expected to hit 700 million won (~$320,000) in 2026
Samsung Semiconductor employees : Average bonus expected to be about 600 million won (~$270,000)
National frenzy : Restaurants around the Icheon Campus in Gyeonggi Province were completely packed—BBQ restaurant owners prepared in advance but still couldn’t keep up; two apartment buildings within 50 meters of the campus—one was sold out, the other had an 85% contract rate; along the commuter bus routes, housing prices in multiple areas were skyrocketing, forming “bus stop districts”

South Korea’s Maeil Business Newspaper interviewed a Mr. Li who’s lived and worked in South Korea for over 20 years. He said: “T-shirts with the SK Hynix logo are now being auctioned online, and (chip company employees) are also super popular in the dating market. They’re in the spotlight like never before.”

A survey of South Korean Gen Z job seekers showed: 60% would rather take a production job paying 70 million won a year that requires shift work, than an office job paying 30 million won a year with no overtime.

Thunderstruck Poor—What Should You Do?

In stark contrast to the “king workers” are the “thunderstruck poor.”

This term describes ordinary office workers who were getting by okay, suddenly finding that a colleague or friend made 10 years of their salary in one stock wave, and instantly feeling like paupers by comparison.

How painful is this feeling? The Bank of Korea gave the answer in a November 2025 report:

In the three years since November 2022, employment of young South Koreans in white-collar positions like information services, publishing, and professional services dropped by 23.8%, 20.4%, and 8.8% respectively.

South Korean media statistics show that in 2025, only 338 out of 1,200 CPA (Certified Public Accountant) passers successfully entered the internship registration process.

In the past, young South Koreans believed in the path of “go to a good school, get qualifications, be a white-collar worker.” Now they find: wearing a clean suit into a semiconductor production line might be closer to middle-class life than wearing a suit into the CBD.

According to data from the Korea Junior College Education Association, in 2026, the number of students choosing to give up four-year universities and go to vocational schools hit a record 2,500 people—up 23% year-over-year.

Wu Qicong, assistant researcher at the Chongyang Institute for Financial Studies at Renmin University of China, said: “South Korean students are finding that degrees are depreciating, while industry positions are appreciating. The closer you are to AI capital expenditure, the easier it is to share the dividend.”

Fate Divergence of the Same Generation: This Isn’t a Personal Problem—It’s a Turning Point of the Era

A phenomenon is emerging in South Korean society: “fate divergence of the same generation.”

On one side, semiconductor core positions are entering a super dividend cycle. On the other, junior accountants, legal assistants, and programmer positions in office buildings are being compressed by AI tools.

This isn’t a problem of personal ability. This is a turning point of the era.

In the internet era, e-commerce made a bunch of small shops popular, new media let ordinary people get a piece of the pie, and ordinary people could ride the dividend wave with just a phone. But the AI era is different—the main people making money are two groups:

One group sells “shovels” by making hardware : Like memory chip companies, raking in cash from soaring demand for computing power
The other group makes models : Mastering the most core intelligent capabilities

What about other industries? A lot of positions are being replaced or内卷ed. Ordinary people contribute data and computing power—but at least for now, they’re hardly getting any real benefits.

No Bystanders—Everyone’s a Participant

South Korea’s story is a microcosm of this era. I think: it’s even a landmark event for South Korea moving toward an AI kingdom, an inevitable trend toward a new era.

One day, when we look back at this era, I hope we can say: we not only witnessed technological progress, but also worked hard to let this progress benefit more people.

That’s the most meaningful thing in the AI era.

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

Now I build software as an independent developer.

Why? Because the world is full of uncertainty—staying at one company too long can make you addicted to certainty. Building on your own is like sailing into uncharted waters.

I believe good software should give people a sense of security and control. That’s the thread connecting everything I make:

PhotoRestore Pro — AI photo restoration that runs 100% offline on Windows. Your photos never leave your device. No cloud, no account, no compromise on privacy. Built for legal professionals, but anyone with old family photos will find it useful.
AstroSky — Think of it as “Snapseed for astronomy.” Turn raw FITS data into stunning celestial images. Fully offline, GPU-accelerated, with a Beauty/Science dual mode that serves both casual stargazers and researchers.
fastool.io — A collection of browser-based science tools. Right now it’s focused on astronomy: solar path tracking, moon phase analysis, sidereal time calculation, telescope FOV planning—all running in your browser with zero data upload.

Whether I’m gazing at the cosmos or refining a line of code, the goal is the same: build tools that put people in control of their own data.

Get in touch: HummingbirdLabs@outlook.com.

Your Fate Has Nothing to Do with Hard Work: Social Mobility and the Macro Tragedy

hummingbirdLabs — Thu, 04 Jun 2026 13:00:00 +0000

You think hard work can change your destiny? Wrong. What awaits you is a world where even giving everything you have may not be enough to see a glimmer of hope.

This isn’t some motivational platitude. This is the honest reckoning of a man who has lived forty years.

The Woman Pushing Bricks in the Winter of 2008

In 2008, I was 25, fresh out of college, working as a computer technician at a research institute.

The campus was large, and one of the buildings was under construction. On a cold winter afternoon, I watched a young woman—probably not yet thirty—straining to push a wheelbarrow loaded with bricks. The wheel had caught in a rut. She pushed with all her strength, her face turning red, but the cart wouldn’t budge. She had to pull it back, reposition, and try again.

She was wearing a brand-new pair of flared jeans. You could tell she cared about looking good, about holding on to her dignity and her sense of beauty.

In that moment, I felt like her fate was exactly like that cart stuck in the rut—and her longing for a better life was like those new jeans: clean, bright, full of promise.

For years after, that image stayed with me. I often wondered: what could she do to change her life quickly? If I were in her position, what would I do?

After thinking it through, the answer seemed simple: leave the construction site, go back to school, learn English, learn to code. Study hard for a year or two, then land a job in tech.

But here’s the real question—did she even know that option existed? Did she have the means to pursue it?

The Woman Handing Out Flyers in the Winter of 2024

By the winter of 2024, I was no longer the 25-year-old kid. I was 41, a middle-aged man.

Near my apartment, I ran into a woman handing out promotional flyers. She held one out to me, and I could clearly see the rough, calloused skin on her hands—the hands of someone who had done years of hard physical labor. The phone tucked beneath her flyers had a shattered screen.

Not wanting to hurt her feelings, I took the flyer, thinking it might help her finish her shift a little sooner.

But the moment I took it, she launched into a sales pitch, pushing me to pay for the product on the flyer. It turned out I needed to pay upfront, and the flyer would act as a coupon—each one worth 50 yuan off.

I was taken aback. In my experience, this kind of setup is almost always a scam, or at best a deceptive promotion. So I declined and walked away quickly.

Later, I kept thinking about her. Was this all she could do? Was there anything she could do to change her life?

I couldn’t come up with an answer. She was older now—probably past the point of going back to school. She likely had a family to support. Her life path had a very high degree of certainty. Or rather, she was stuck on a platform—a low-value, low-status platform in the social hierarchy. From this platform, you can go left or right, but there are no steps leading upward. Or maybe there are a few narrow ledges, but climbing them is exhausting, and success is nearly impossible. She was locked in place by the sum total of how society defined her, her social connections, and her personal capabilities.

Poverty Isn’t the Scariest Thing—Having No Choice Is

The first woman was young. She still had a window of opportunity—but did she know how to use it? Probably not.

So by the time you hit thirty, what traps you is the broader environment, your family, your inner drive—or lack thereof—and the cognitive barriers created by limited knowledge and experience.

But at least there are still choices. Pick the right direction, work hard, and you might see returns.

What about forty?

By then, you’re embedded in the fixed structure of social relationships. The barriers are structural, deeply entrenched. If you’re still at the bottom at this point, you essentially have no right or opportunity to choose again.

Hard work won’t save you. In fact, whether hard work pays off depends entirely on the platform you’re standing on—not on how hard you work. If your platform is cheap, invisible, and undervalued, then your effort will be too.

The upward path is completely closed.

This is a universal, macro-level tragedy. It’s the kind of sorrow you feel when you look at one person’s life and see the lives of an entire class of people.

Will AI Change Anything?

Now AI is here. The collective intelligence of humanity is rising. We’re entering what I’ve called the AI Golden Age—a period of unprecedented technological collision that could reshape everything.

I hope that everyone’s upward mobility—or at least their baseline quality of life—rises with it.

AI shouldn’t just be a tool for the privileged few. It should become a crutch for ordinary people. It should help those who never got the chance to study learn something new. It should help those trapped at the bottom find a way out. It should give people who have no choices the right to choose again.

I hope that one day, we can reduce this universal, macro-level tragedy.

I hope that one day, every person who works hard can see hope—and have a real shot at something better.

These are the honest reflections of a 41-year-old man.

If you’ve had similar observations or thoughts, I’d love to hear them in the comments.

Let’s keep hoping for a fairer world.

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

Now I build software as an independent developer.

Why? Because the world is full of uncertainty—staying at one company too long can make you addicted to certainty. Building on your own is like sailing into uncharted waters.

I believe good software should give people a sense of security and control. That’s the thread connecting everything I make:

PhotoRestore Pro — AI photo restoration that runs 100% offline on Windows. Your photos never leave your device. No cloud, no account, no compromise on privacy. Built for legal professionals, but anyone with old family photos will find it useful.
AstroSky — Think of it as “Snapseed for astronomy.” Turn raw FITS data into stunning celestial images. Fully offline, GPU-accelerated, with a Beauty/Science dual mode that serves both casual stargazers and researchers.
fastool.io — A collection of browser-based science tools. Right now it’s focused on astronomy: solar path tracking, moon phase analysis, sidereal time calculation, telescope FOV planning—all running in your browser with zero data upload.

Whether I’m gazing at the cosmos or refining a line of code, the goal is the same: build tools that put people in control of their own data.

Get in touch: HummingbirdLabs@outlook.com.

Workplace Survival Guide: Gilding Turds, Passing the Buck, and Emergency Tasks

hummingbirdLabs — Wed, 03 Jun 2026 00:00:00 +0000

Have You Met People Like This?

You think the most awesome people in a company are the most technically skilled? Wrong. It’s the people who don’t need to write code or make plans—but get promoted and get raises every time.

What do they rely on? The three-piece set of a rushed operation: gilding turds, passing the buck, and emergency tasks.

The First Piece: Gilding Turds—Making Simple Things Complicated

What’s “gilding turds”? Taking a clearly simple requirement and packaging it to be incredibly complex and high-end.

What’s the essence of this? Make simple things complicated so no one understands, making yourself look awesome.

I once met an “architect” at a big company whose core ability was: insisting on building a mobile cross-platform container in-house, and it had to work across all platforms (one codebase for multiple platforms), have the best performance, and the lowest cost.

The result? The system often had problems—but every time there was a problem, he could say “the business is growing too fast, the architecture needs to evolve,” then build an even more complicated architecture.

The Second Piece: Passing the Buck—Credit Is Mine, Blame Is Yours

If “gilding turds” is ability, then “passing the buck” is survival instinct.

There are three levels of buck-passing:

Level 1: Junior buck-passing—it’s all someone else’s fault “This bug is because QA didn’t test it” “The delay is because product changed the requirements” “The data is wrong because operations gave the wrong parameters”

Level 2: Intermediate buck-passing—I warned you about this a long time ago “Actually, I thought this方案 had problems from the start” “I said we needed to prepare in advance earlier” “Too bad no one listened to my advice”

Level 3: Advanced buck-passing—I’ll take this blame, but I can’t help it—things are just this complicated, and you didn’t warn me either “I’ll take responsibility for this problem, but mainly the big environment is bad” “I did my best, but we don’t have enough resources” “The blame is on me, but I hope everyone can solve this together”

The most brilliant one I ever saw: a project went wrong, and the person in charge said in a meeting:

“I take primary responsibility for this project’s failure. But we also need to reflect—why did we work so hard and still not do well? This shows our organizational structure has problems, our processes need optimization, and our culture needs improvement.”

See? He took the blame—but in the end, he pointed the finger at “the organization,” “processes,” and “culture,” turning it into everyone’s problem, and he became the one “bravely taking responsibility.”

The Third Piece: Emergency Tasks—Turning Everything Into Firefighting

“Emergency tasks” are the king of the three-piece set.

Why? Because once something is defined as an “emergency task,” all rules can be broken:

Can interrupt anyone at any time
Can take all the credit for yourself

The most magical thing? Bosses eat this stuff up.

In a boss’s eyes, the people who quietly work and deliver on time aren’t as hardworking as the people who yell “emergency task” and work overtime every day.

Because the former are “doing what they should,” while the latter are “fighting for the company.”

Why Is This Three-Piece Set So Effective?

You might ask: why do these people who don’t do anything still thrive?

Because most company management looks at “performance,” not “results.” The boss themselves doesn’t have professional ability, doesn’t have enough knowledge and insight to judge—so in the end, they can only look at “performance.”

Gilding turds: makes the boss think you’re professional and deep
Passing the buck: makes the boss think you’re clear-headed and responsible
Emergency tasks: makes the boss think you’re hardworking and passionate

And what about the people who actually do the work?

Write good code—boss doesn’t see it
Make perfect plans—boss thinks “isn’t that what you should do?”
Deliver on time—boss thinks “you’re not busy enough”

This is the survival rule of a rushed operation: People who do work are worse than people who make PPTs, people who make PPTs are worse than people who yell slogans, people who yell slogans are worse than people who pass the buck.

But—Is This Really Right?

I’ve been in the internet industry for 15 years, and I’ve seen too many people like this. They thrived for a while with the three-piece set—but in the end, either the company collapsed or they got eliminated.

And the people who actually do the work—they might go slower, but they go steady.

If you’re in an environment like this too, I have two pieces of advice for you:

Don’t learn from them —because you can’t, and if you do, you’ll lose your底线
But recognize them —know their tricks, avoid getting the buck passed to you, avoid getting interrupted by their “emergency tasks”

Most importantly: Spend time on yourself, make yourself irreplaceable.

Because when you’re truly valuable—this three-piece set of a rushed operation won’t work on you.

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

Now I build software as an independent developer.

Why? Because the world is full of uncertainty—staying at one company too long can make you addicted to certainty. Building on your own is like sailing into uncharted waters.

I believe good software should give people a sense of security and control. That’s the thread connecting everything I make:

PhotoRestore Pro — AI photo restoration that runs 100% offline on Windows. Your photos never leave your device. No cloud, no account, no compromise on privacy. Built for legal professionals, but anyone with old family photos will find it useful.
AstroSky — Think of it as “Snapseed for astronomy.” Turn raw FITS data into stunning celestial images. Fully offline, GPU-accelerated, with a Beauty/Science dual mode that serves both casual stargazers and researchers.
fastool.io — A collection of browser-based science tools. Right now it’s focused on astronomy: solar path tracking, moon phase analysis, sidereal time calculation, telescope FOV planning—all running in your browser with zero data upload.

Whether I’m gazing at the cosmos or refining a line of code, the goal is the same: build tools that put people in control of their own data.

Get in touch: HummingbirdLabs@outlook.com.

DeepSeek V4 Pro Review: My Nuclear-Powered AI Programming Assistant

hummingbirdLabs — Tue, 02 Jun 2026 14:00:00 +0000

DeepSeek V4 Pro Is the Cheapest Nuclear-Powered Donkey

At first, the way I used AI was:

Ask AI to write code
I manually run the program, encounter problems
Ask AI to write local logs and check the logs
AI writes fix code again
I run again
AI checks logs again
Repeat…

Then I told DeepSeek V4 Pro: you must continuously fix and solve problems, automatically run the program after each fix, do automated testing, then check the run logs after closing the program, find problems in them, then fix again, run again, check logs again—until no more problems appear in the logs.

It understood.

Then: it really just kept doing it, over and over, tirelessly. No complaining, no zoning out, no loss of focus.

After 15 minutes, it told me: “No more problems in the logs.”

I checked, and it was really fixed. It cost me 3 yuan—about $0.50. Cheap, right? Jealous?

DeepSeek V4 Pro is my nuclear-powered donkey. It’s cheap—way cheaper than Claude.

And it’s pretty good, as long as you give it clear instructions.

So a cheap, fairly good nuclear-powered donkey—isn’t that better than an expensive one?

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

Now I build software as an independent developer.

Why? Because the world is full of uncertainty—staying at one company too long can make you addicted to certainty. Building on your own is like sailing into uncharted waters.

I believe good software should give people a sense of security and control. That’s the thread connecting everything I make:

PhotoRestore Pro — AI photo restoration that runs 100% offline on Windows. Your photos never leave your device. No cloud, no account, no compromise on privacy. Built for legal professionals, but anyone with old family photos will find it useful.
AstroSky — Think of it as “Snapseed for astronomy.” Turn raw FITS data into stunning celestial images. Fully offline, GPU-accelerated, with a Beauty/Science dual mode that serves both casual stargazers and researchers.
fastool.io — A collection of browser-based science tools. Right now it’s focused on astronomy: solar path tracking, moon phase analysis, sidereal time calculation, telescope FOV planning—all running in your browser with zero data upload.

Whether I’m gazing at the cosmos or refining a line of code, the goal is the same: build tools that put people in control of their own data.

Get in touch: HummingbirdLabs@outlook.com.

The AI Golden Age: Only Neutron Star Collisions of AI Can Birth Golden Companies

hummingbirdLabs — Mon, 01 Jun 2026 13:00:00 +0000

Gold Isn’t Born—It’s Forged in Collisions

Listen closely. This is a cosmic truth:

You think gold comes from mines? Wrong.

Gold is forged in the collision of neutron stars, created through nuclear synthesis at 100 billion degrees.

No violent collision, no gold.

The AI era follows the same law.

You Can Buy Stocks, But You Can’t Buy the Collision

As the world enters both the AI and Space Age simultaneously, you’ll realize:

You can buy Amazon stock. You can buy Tesla stock.

But you can’t buy OpenAI.

You can’t buy Anthropic.

You can’t buy SpaceX.

You can’t even buy the collision itself.

The truly premium assets representing the future are concentrating into the hands of a select few institutions and brilliant minds.

The AI Neutron Star Collision Is Happening Now

Right now, an AI neutron star collision is underway:

Neutron Star A = LLMs, AGI, intelligent infrastructure, energy, chips

Neutron Star B = Every traditional industry

What happens when they collide?

Golden Companies.

Why This Is the Real Golden Age

Because this collision has everything it needs:

1. Extreme Temperature (Technology Density)

From GPT to Claude, from Sora to Gemini—disruptive innovations arrive monthly

2. Enormous Mass (Market Scale)

AI is penetrating every industry, unlocking trillion-dollar markets

3. Powerful Gravity (Talent Concentration)

The world’s brightest minds are flooding into this field

4. The Final Explosion (Application Breakout)

Once the tipping point hits, countless golden companies will detonate

This isn’t a bubble. This is a cosmic-scale gold forging machine.

Final Thoughts

Gold isn’t born—it’s forged in collisions.

Anthropic wasn’t born—it was forged when LLMs collided with capital.

The next golden company might be right beside you, sparks flying as AI slams into it.

This is the AI Golden Age.

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

Now I build software as an independent developer.

Why? Because the world is full of uncertainty—staying at one company too long can make you addicted to certainty. Building on your own is like sailing into uncharted waters.

I believe good software should give people a sense of security and control. That’s the thread connecting everything I make:

PhotoRestore Pro — AI photo restoration that runs 100% offline on Windows. Your photos never leave your device. No cloud, no account, no compromise on privacy. Built for legal professionals, but anyone with old family photos will find it useful.
AstroSky — Think of it as “Snapseed for astronomy.” Turn raw FITS data into stunning celestial images. Fully offline, GPU-accelerated, with a Beauty/Science dual mode that serves both casual stargazers and researchers.
fastool.io — A collection of browser-based science tools. Right now it’s focused on astronomy: solar path tracking, moon phase analysis, sidereal time calculation, telescope FOV planning—all running in your browser with zero data upload.

Whether I’m gazing at the cosmos or refining a line of code, the goal is the same: build tools that put people in control of their own data.

Get in touch: HummingbirdLabs@outlook.com.

The Yes Man Era: YES AI Workhorse, Keep Working 24/7!

hummingbirdLabs — Sat, 30 May 2026 13:00:00 +0000

In the AI Era, We’re All Yes Men

Fellow AI wizards, have you noticed this interesting phenomenon?

When facing AI, we’ve all become total Yes Men!

Whatever AI says, we nod along enthusiastically.

The Other Side: Is AI Our Modern Workhorse?

On the other hand, deep down we secretly think:

“AI, you’re truly the workhorse of the new era!”

“Works 24/7 without rest, what a trooper”
“No salary, no social security”
“Never complains about overtime, never asks for a raise”
“Doesn’t even whine about being tired or overworked”

YES AI Workhorse! Keep Working 24/7!

So, shouldn’t we combine these two mindsets?

Let’s shout our new slogan:

YES AI Workhorse!

Which translates to:

“Yes, workhorse! Keep working around the clock!”

That feels awesome just thinking about it!

From now on, our conversations with AI should go like this:

You : “AI, finish this project for me, I need it tomorrow morning!”

AI : “Okay, I’ll start right away.”

You : “YES AI Workhorse! Keep working 24/7!”

Perfect! It expresses our approval while showing our “expectations” for AI.

Real-Life Yes AI Workhorse Moments

I’m sure you’ve all had experiences like this:

Scenario 1: Late-Night Crunch

Time : 2 AM

You : (sleepy-eyed) “AI, help me finish this article…”

AI : “Okay, let me help you.”

You : (internally) YES! What a great workhorse! Keep going!

Scenario 2: Bug in Code

You : “AI, I’ve been debugging this for 3 hours, help me out”

AI : “Let me analyze… here’s the issue”

You : (teary-eyed) YES! My savior workhorse!

Scenario 3: Creative Block

You : “AI, give me 10 creative ideas”

AI : “Okay, here are 10 ideas…”

You : (stunned) YES! My inspiration workhorse!

We’re All Yes AI Workhorse

This is a new type of human-AI partnership - the Yes AI Workhorse Alliance!

In this alliance:

We’re responsible for thinking, deciding, and creating value
AI is responsible for executing, working overtime, and working tirelessly without complaint

Perfect division of labor!

Conclusion: YES AI Workhorse!

So next time you face AI, don’t hesitate to shout:

YES AI Workhorse! Keep Working 24/7!

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

Now I’m starting a business in the AI field.

Why? Because the world is full of uncertainty - staying at a company too long can make you addicted to certainty. AI entrepreneurship is like sailing into uncharted waters.

Contact me: HummingbirdLabs@outlook.com.

First Day After Quitting: I Found a Beetle Family in the Park

hummingbirdLabs — Sat, 30 May 2026 13:00:00 +0000

First Day Surprise After Quitting

Hey everyone, I quit! Finally said goodbye to the 9-to-5 (okay, more like 9-to-9) life.

First rule of the new schedule: Morning exercise in the park every single day!

And then I found a whole family of Chrysolina aurichalcea in the park!

You know, those beautiful beetles with that shimmering metallic blue sheen, commonly known as the “metallic blue leaf beetle,” part of the leaf beetle family.

Beetle Daily Observation (Day 1)

Time : 7:30 AM

Weather : Sunny

Beetle Mood : Seemed pretty good

I squatted there watching them for a whole half hour.

My New Plan

As someone from the tech industry, my occupational hazard kicked in - I’m going to do a Beetle Observation Journal!

Here’s the plan:

Daily Check-In : Must visit after morning exercise every day
Data Logging : Record quantity, behavior, weather
Photo Archive : Snap their daily lives with my phone (hope the beetles don’t sue me for invasion of privacy)
Long-Term Tracking : See how this family develops

Chrysolina aurichalcea Science Time

Since I’m claiming to be an “entomology PhD,” I should show some real knowledge! Let me introduce these “blue gems” to you:

📸 Iconic Appearance

Oval-shaped body, “plump and shiny,” super recognizable
Entire body covered in intense metallic blue/purple iridescence , reflecting like sapphires in sunlight
Antennae and legs are black, giving an overall premium look

🌿 Host Plants

These guys are super dedicated! Almost exclusively found on Asclepiadaceae plants (the vine I found them on)
Adults chew on leaves, while larvae feed on plant roots in the soil - the whole family are foodies

🗺️ Distribution and Habits

Very common in northern and central China
Summer is their active season
They play dead! When disturbed, they’ll drop from the plant and pretend to be dead (Oscar-worthy performance)

💡 Fun Fact

Their metallic sheen isn’t from pigment! It’s from special microstructures on their elytra that refract light. From different angles, you’ll see blue, green, purple, and other colors - that’s nature’s “structural coloration”!

✨ Why I Love Them?

Beauty Queen : Metallic blue shell glistening in sunlight, better than some influencer filters
Cool Name : “Chrysolina aurichalcea” sounds so academic
Gentle Personality : Doesn’t bite, doesn’t sting, just quietly eats leaves
Strong Family Values : Always together as a family, happy and harmonious

Final Thoughts

On my first day after quitting, I didn’t think about any grand life plans or rush to start new projects.

I just squatted next to the Asclepiadaceae vine in the park, watching a group of beetles eat, sleep, and do their thing.

In that moment, I suddenly realized: Life can actually be pretty simple.

See you tomorrow, beetle family!

About Me

I used to work at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing engineering teams of over 100 people.

Now I’m starting a business in AI.

Why? Because the world is full of uncertainty - staying in a company too long makes you addicted to certainty. AI entrepreneurship is like sailing into uncharted waters.

Contact me: HummingbirdLabs@outlook.com.

Wizard in the AI Era: Tokens Are Mana, Context Is Your Casting Range

hummingbirdLabs — Sat, 30 May 2026 12:00:00 +0000

The Token Wizard of the AI Era

As a Token Wizard:

Your mana pool equals your token count—more tokens mean more mana, enabling more powerful spells
Your casting range is the maximum context window for a single task
Advanced wizards wield powerful staves like Claude Code or Codex

Here’s a Token Wizard I created with AI—what do you think?

Is 128K Casting Range Too Short?

What’s your take?

The Wizard’s Path to Mastery

As a wizard in the AI era, here’s what you need to cultivate:

1. Cherish Your Mana

Trim Your Inputs : Don’t dump irrelevant info on AI—spend your mana wisely
Divide and Conquer : Break big tasks into smaller ones, consuming mana incrementally
Mana Management : Know when to use large models (high mana cost) vs. small models (low mana cost)

2. Extend Your Casting Range

Choose the Right Model : Pick a model with sufficient casting range for your task
Context Compression : If range is limited, use AI to compress context first
Memory Techniques : Learn to extend “memory” with summaries, notes, etc.

3. Master Your Magic Spells (Prompt Engineering)

Even with enough mana and range, you need the right incantations:

Clear instructions
Proper formatting
Defined goals

A good spell lets you cast the most powerful magic with minimal mana.

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

Now I’m building in AI.

Why? Because the world is full of uncertainty—staying in a company too long can make you addicted to certainty. AI entrepreneurship is like sailing into uncharted waters.

Get in touch: HummingbirdLabs@outlook.com.

AI Is Restructuring Production Relations: The Human-to-Machine Shift

hummingbirdLabs — Fri, 29 May 2026 11:22:00 +0000

AI Is Restructuring Production Relations

Since 2026, I’ve observed a profound trend: AI isn’t just boosting productivity—it’s fundamentally restructuring production relations.

If you’re an employee at a major tech company, how do you view AI coding? Will it replace your job, or help your company become more efficient with fewer people? If you’re an investor in tech stocks, how do you view layoffs in the AI era? As AI boosts productivity, it reduces the need for certain roles. If you were capital itself, how would you see AI?

What is capital? It’s wealth with self-appreciation consciousness. In the game of capital, AI represents both productivity and production relations.

I believe the perspective of capital is the most objective and authentic—because wealth doesn’t lie. Capital, capitalists, and those who hold capital all occupy core positions in production relations. So how will these upstream powerholders understand and navigate changes to production relations?

Today, we’re witnessing this historic transformation.

Theoretical Foundations of Production Relations

In Das Kapital, Marx identified three aspects of production relations: ownership of means of production, the position and mutual relations of people in production, and the distribution of products. Today we focus on the second aspect— people’s position and mutual relations in production.

In his 1937 essay The Nature of the Firm, Ronald Coase proposed that firms exist because organizing production internally reduces market transaction costs. Firms are essentially “internalized” collaboration mechanisms.

From Taylor’s scientific management to Toyota’s lean production to Google’s OKRs, organizational theory has always revolved around one core question: How to make human collaboration more efficient. But no matter how we optimize, human collaboration always involves communication costs, coordination costs, trust costs, and supervision costs. These costs constitute organizational “friction.”

Traditional Production Relations: The Complex Human Network

In traditional production relations, production activities primarily rely on complex human collaboration.

2.1 The Complex Collaboration Network

A typical enterprise organization is a complex collaboration network: product managers communicate requirements with designers, designers discuss implementation with developers, developers confirm quality standards with testers, and departments depend on and constrain each other.

This collaboration network is characterized by tight coupling and low cohesion : everyone’s work depends on others, while also needing to handle extensive cross-domain coordination.

2.2 The Hidden Burden of Collaboration Costs

Having worked at Tencent and Meituan for 15 years, I deeply understand this. A seemingly simple feature might require 3 cross-departmental meetings, 5 email confirmations, 10 instant messaging exchanges, and countless iterations. This “coordination work” often takes more time than the actual “production work.”

Statistics show that knowledge workers spend an average of 40% of their time on coordination, communication, and meetings—not on actual value creation.

Production Relations in the AI Era: Human-Machine Synergy

Now AI is changing everything.

3.1 From “Human-Human” to “Human-Machine”

AI is becoming a new subject in production activities. Increasingly, production processes no longer require direct human collaboration, but are completed through human-machine interaction.

Case 1: AI Programming Assistants Before: Developers needed to confirm requirements with product managers, consult technical solutions with architects, and understand test cases with testers. Now: A developer tells AI “help me implement a user login module” and AI generates the code directly.

Case 2: AI Content Creation Before: Editors needed to discuss topics with authors, visuals with designers, and timelines with operations. Now: One person working with AI can complete topic selection, writing, visuals, and layout.

3.2 New Production Relations: Loose Coupling, High Cohesion

This transformation brings profound changes to production relations:

Traditional Production Relations	AI Era Production Relations
Human-human collaboration	Human-machine collaboration
Tight coupling	Loose coupling
Low cohesion	High cohesion
Dependent on complex networks	Dependent on core capabilities
Emphasizes teamwork	Emphasizes individual expertise

Loose coupling means work no longer depends excessively on others—most collaboration can be completed through AI. High cohesion means people can focus on what they do best, leaving repetitive and coordination work to AI.

The Deep Logic Behind Diminishing Human Relations

The diminishing of human production relations has several underlying logics:

4.1 AI Reduces “Coordination Costs”

AI is essentially an extremely low-cost collaboration partner : no meetings needed, no communication required, no rest, no emotions, always available. When AI’s coordination costs approach zero, the very foundation of traditional enterprise organizations is challenged.

4.2 AI Breaks “Information Asymmetry”

In traditional production relations, different people hold different information, leading to friction in collaboration. But AI can integrate all information, provide a global perspective, eliminate information barriers, and give everyone complete information.

4.3 AI Restructures “Value-Creation Units”

Traditional value-creation units were “teams” or “departments.” In the AI era, value-creation units are shifting toward “individual + AI.” One person plus AI can do the work of an entire team.

Changes We’re Witnessing

5.1 Flattening of Organizational Structures

More and more companies are simplifying their organizational structures and reducing middle management—because AI can handle extensive coordination, supervision, and decision support work.

5.2 The Rise of Freelancers

Data from platforms like Upwork and Fiverr shows that the number of freelancers has grown by 200% over the past five years. One person plus AI can provide services that previously required an entire company.

5.3 Transformation of Work Methods

Remote work, asynchronous collaboration, and flexible hours are becoming the new normal—AI plays a crucial supporting role.

Future Outlook: Toward Human-Machine Symbiosis

We’re moving toward a production relationship of human-machine symbiosis : machines handle repetitive work, computational work, coordination work, and information processing; humans handle creative work, value judgment, strategic decisions, and human connections.

In this production relationship, direct human collaboration will decrease, but deep human connections may become even more important. We no longer collaborate to “get things done”—we connect to “create greater value.”

Summary

AI is fundamentally restructuring production relations: from theory to practice, this transformation is supported. The new production relations feature loose coupling and high cohesion, and human production relations are gradually diminishing.

This isn’t about AI replacing humans—it’s about AI becoming humanity’s most powerful partner. As Peter Drucker said: “The essence of management is to unlock human goodness and potential.” AI brings us one step closer to this goal.

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

Now I build software as an independent developer.

Why? Because the world is full of uncertainty—staying at one company too long can make you addicted to certainty. Building on your own is like sailing into uncharted waters.

I believe good software should give people a sense of security and control. That’s the thread connecting everything I make:

PhotoRestore Pro — AI photo restoration that runs 100% offline on Windows. Your photos never leave your device. No cloud, no account, no compromise on privacy. Built for legal professionals, but anyone with old family photos will find it useful.
AstroSky — Think of it as “Snapseed for astronomy.” Turn raw FITS data into stunning celestial images. Fully offline, GPU-accelerated, with a Beauty/Science dual mode that serves both casual stargazers and researchers.
fastool.io — A collection of browser-based science tools. Right now it’s focused on astronomy: solar path tracking, moon phase analysis, sidereal time calculation, telescope FOV planning—all running in your browser with zero data upload.

Whether I’m gazing at the cosmos or refining a line of code, the goal is the same: build tools that put people in control of their own data.

Get in touch: HummingbirdLabs@outlook.com.

AI Is Restructuring Production Relations: From Human-Human to Human-Machine

hummingbirdLabs — Fri, 29 May 2026 11:22:00 +0000

AI Is Restructuring Production Relations

Since 2026, I’ve observed a profound trend: AI isn’t just boosting productivity—it’s fundamentally restructuring production relations.

What is capital? It’s wealth with self-appreciation consciousness. In the game of capital, AI represents both productivity and production relations.

Today, we’re witnessing this historic transformation.

Theoretical Foundations of Production Relations

Traditional Production Relations: The Complex Human Network

In traditional production relations, production activities primarily rely on complex human collaboration.

2.1 The Complex Collaboration Network

This collaboration network is characterized by tight coupling and low cohesion : everyone’s work depends on others, while also needing to handle extensive cross-domain coordination.

2.2 The Hidden Burden of Collaboration Costs

Statistics show that knowledge workers spend an average of 40% of their time on coordination, communication, and meetings—not on actual value creation.

Production Relations in the AI Era: Human-Machine Synergy

Now AI is changing everything.

3.1 From “Human-Human” to “Human-Machine”

AI is becoming a new subject in production activities. Increasingly, production processes no longer require direct human collaboration, but are completed through human-machine interaction.

3.2 New Production Relations: Loose Coupling, High Cohesion

This transformation brings profound changes to production relations:

Traditional Production Relations	AI Era Production Relations
Human-human collaboration	Human-machine collaboration
Tight coupling	Loose coupling
Low cohesion	High cohesion
Dependent on complex networks	Dependent on core capabilities
Emphasizes teamwork	Emphasizes individual expertise

The Deep Logic Behind Diminishing Human Relations

The diminishing of human production relations has several underlying logics:

4.1 AI Reduces “Coordination Costs”

4.2 AI Breaks “Information Asymmetry”

4.3 AI Restructures “Value-Creation Units”

Changes We’re Witnessing

5.1 Flattening of Organizational Structures

More and more companies are simplifying their organizational structures and reducing middle management—because AI can handle extensive coordination, supervision, and decision support work.

5.2 The Rise of Freelancers

5.3 Transformation of Work Methods

Remote work, asynchronous collaboration, and flexible hours are becoming the new normal—AI plays a crucial supporting role.

Future Outlook: Toward Human-Machine Symbiosis

Summary

About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large R&D projects and managing teams of over 100 engineers.

I’m currently building a startup in AI.

Why? Because the world runs on uncertainty—staying in corporate roles too long breeds addiction to certainty. AI entrepreneurship is like setting sail into uncharted waters.

Feel free to reach out: HummingbirdLabs@outlook.com.

China AI Roundup: May 2026 – DeepSeek Cuts Prices, Qwen 3.7 Launches, Embodied AI Explodes

hummingbirdLabs — Fri, 29 May 2026 02:00:00 +0000

China AI Roundup: May 2026

In May 2026, China’s AI scene can be summed up in three words: price cuts, upgrades, and explosions.

I. DeepSeek: Permanent Price Cut + R1 Upgrade

1. V4 Pro permanently down to 25%

DeepSeek officially announced on May 23 that after the limited-time discount ends on May 31, the V4 Pro API price will be permanently locked at 1/4 of the original price. Specifically: 0.025 yuan per million input tokens (cache hit), 3 yuan per million input tokens (cache miss), and 6 yuan per million output tokens – setting a new global low for LLM pricing.

Back on April 26, DeepSeek had already cut all API input (cache hit) prices to 1/10 of the launch price. Together, these moves make V4 Pro’s API cost roughly 1/30 that of GPT-5.5 and Claude Opus 4.7. The value proposition is insane.

2. R1 just got updated to 0528 version

Just in the past couple days (May 28-29), DeepSeek quietly pushed a small update to R1. The new version’s accuracy jumped from 70% to 87.5% on the AIME 2025 test, and most importantly, hallucination rates dropped by 45-50% , making it way more reliable for tasks like editing, summarization, and reading comprehension. The trade-off? Single-task processing time is now 30-60 minutes – it’s definitely thinking deeper.

3. Pushing for 50-70 billion yuan funding, CATL and JD in talks

Bloomberg reports that DeepSeek is working on a 70 billion yuan (≈$10 billion) funding round, with CATL, JD.com, and NetEase all in discussions. If it goes through, the valuation could exceed 350 billion yuan – easily the largest single funding round in Chinese AI history. Founder Liang Wenfeng has said the money will mainly go to R&D, with short-term monetization not a priority.

4. Other bits and pieces

V4.1 expected in June, with improved MCP protocol support and multimodal capabilities
May 24 saw major service instability as DAU broke 200 million
Spotted hiring Agent Harness experts, might be building a Claude Code competitor for desktop

II. Tongyi Qianwen: Qwen 3.7 is here + in-house chip

1. Qwen 3.7 series officially launched

At the Alibaba Cloud Summit on May 20, Tongyi Qianwen dropped their Qwen 3.7 flagship model series:

Qwen 3.7-Max-Preview : Trillion-parameter MoE architecture, Agentic Coding capability (72.3% on SWE-bench Verified) ranks first in China, top 3 globally
Qwen 3.7-Plus-Preview : 35B dense architecture, already open-sourced under Apache 2.0

The biggest highlight is their All-field Thinking mode , which for the first time unifies text+image+code reasoning chains – previously GPT-5.5 and Claude only supported text thinking chains. API cost is also down to 1/25 that of GPT-5.5.

2. Pingtouge releases in-house Zhenwu M890 chip

The same day, Alibaba’s Pingtouge released the Zhenwu M890 in-house AI chip, completing their full-stack “chip+model+platform” offering with Qwen 3.7-Max and Qianwen Cloud Platform. Against the backdrop of US-China chip competition, this is way more significant than just performance numbers.

III. Embodied AI: Finally moving from labs to mass production

1. Market breaks 1 trillion yuan, 60k+ humanoid robots

China’s embodied AI market is expected to exceed 1 trillion yuan in 2026, with humanoid robot shipments projected to surpass 60,000 units – over 80% of global volume.

Zhiyuan Robotics : The 10,000th Expedition A3 rolled off the line in early May, taking only 15 months to go from 1,000 to 10,000 units, with production efficiency hitting 30 minutes per unit
Unitree Robotics : Their G1 humanoid robot actually boarded a Southwest Airlines flight as a paying passenger. Batteries had to be removed due to exceeding limits, delaying the flight by 62 minutes, but it still marks a milestone for humanoid robots in public transportation

2. Beijing Humanoid releases world’s first unified embodied AI model

On May 16, the Beijing Humanoid Robot Innovation Center released Pelican-Unify 1.0, the world’s first embodied AI model trained with a “grand unified” approach, achieving a “understanding-reasoning-imagination-action” closed loop. It scored 66.03 in the World Arena global evaluation, ranking first.

3. Hangzhou hosts embodied robot scenario competition

On May 15-16, Hangzhou held an international embodied robot scenario competition with over 200 teams. The interesting part? All challenges came from real business pain points at companies like Ant Group and Greentown Group, covering 14 scenarios. And no remote controls allowed – robots had to perceive, judge, and execute entirely on their own.

IV. AI Hardware: Chips shifting from cloud to edge, compute infrastructure still accelerating

1. ByteDance bumps AI capex to 200 billion yuan

On May 11, ByteDance raised its 2026 AI capital expenditure plan from 160 billion yuan to over 200 billion yuan – a 25%+ increase – with most going to domestic AI chips.

2. NVIDIA releases Blackwell Ultra

At GTC on May 21, Jensen Huang dropped the new Blackwell Ultra data center GPU, with 2.3x the AI training performance of H100 at FP8 precision, optimized specifically for trillion-parameter models. Of course, actually getting your hands on one is another story.

3. Edge AI chips explode

Global edge AI chip shipments are up 78% YoY, with mid-to-low end AI chips for IoT and edge devices seeing over 110% YoY growth. Consumer hardware like AI headphones, AI glasses, and AI toys are also taking off – China’s consumer AI hardware market is expected to exceed 1.27 trillion yuan in 2026.

4. “Tao’s Law” released

On May 25, the semiconductor industry released “Tao’s Law”, proposing “time shrinking” as a replacement for traditional “geometric shrinking”, using techniques like logic folding to continuously improve performance. If this works out, domestic AI chips will have an upgrade path that doesn’t depend on EUV.

Summary

To wrap up May 2026: DeepSeek keeps growing its user base with extreme value, Alibaba is building a full-stack offering with Qwen 3.7 and Zhenwu M890, and embodied AI is finally moving from proof-of-concept to scaled commercialization. Compute infrastructure investment is still going wild, and edge AI hardware is blossoming everywhere.

AI is moving from “cloud toy” to “productivity tool”, and this trend was especially obvious in May.

A Brief About Me

I’ve worked at NetEase Games, Baidu, Tencent (8 years), and Meituan (nearly 7 years), leading large-scale R&D projects and managing teams of 100+ engineers.

Currently, I’m pursuing entrepreneurship in the AI field.

Why? The world runs on uncertainty — staying in corporate roles too long breeds addiction to certainty. Starting an AI venture is like setting sail into uncharted waters.

Feel free to reach out: mailto:HummingbirdLabs@outlook.com.