The AI Era: What Should We Do to Survive? (Reflections on Founding StreamPDF)

Introduction

AI is called the fourth industrial revolution. Whose life did it take?

Revolutions kill people.

The first industrial revolution began around the 1760s and lasted until about the 1840s. It carried humanity from an agrarian society into industrial civilization and opened the age of steam. Who got hurt this time? The proletariat — farm laborers and artisans. For the first time, humanity held efficient means of production independent of any individual.

The second industrial revolution began around the 1870s and lasted until the early 20th century (roughly 1870-1914). It carried humanity from the age of steam into the age of electricity. Who got hurt this time? The proletariat, and small business owners. Electricity further increased the efficiency of the means of production.

The third industrial revolution began around the 1940s and lasted until the early 21st century (roughly 1940-2010). It carried humanity from the age of electricity into the information age. Who got hurt this time? The proletariat, industrial blue-collar workers, and companies running on outdated, inefficient technology. Information exchange dramatically increased the efficiency of the means of production.

The Fourth Industrial Revolution: Whose Life Did It Take?

AI, the fourth industrial revolution, was born on the internet and is developing at extreme speed. Within the internet industry, programmers, designers, and data analysts were among the first to get hurt.

Since ChatGPT launched on November 30, 2022, the tech layoff wave has not stopped. According to Layoffs.fyi, in just a few years more than 650,000 tech workers have been laid off across more than 2,600 recorded events:

• 2023: 263,000 layoffs, the highest annual total on record. Google cut 12,000, Microsoft 10,000, Amazon 27,000 in total, Meta 21,000 across 2022-2023, and Salesforce 8,000. The same year, IBM announced a hiring freeze on roughly 7,800 back-office roles that AI could handle.
• 2024: 152,000 layoffs — about 780 people losing their jobs every day. Intel cut 15,000, Tesla 14,000, and Cisco more than 10,000. The same year, Duolingo replaced 10% of its contract translators with AI.
• 2025: the first year AI was openly cited as the reason for layoffs. Intel cut another 22,000, Microsoft cut 9,000 in July, and Amazon cut 14,000 in October, writing in an internal memo that "AI is the most transformative technology since the invention of the Internet." The same year, Salesforce's AI agents replaced about 4,000 human support agents, and Chegg, whose students were flocking to ChatGPT, was forced to cut 22% of its staff.
• 2026: the cuts accelerate. In March, Oracle announced a one-time layoff of up to 30,000 people — the largest in its history. Dell cut 11,000. In April, Meta cut 8,000, explicitly attributing it to "AI efficiency." Block's founder Jack Dorsey cut nearly 40% of his workforce, saying AI can now take on more tasks. As of October, nearly 200,000 tech workers have been laid off in 2026 — more than 1,000 per day, the fastest pace of any year on record.

What is more intriguing: the companies cutting their own people are precisely the ones posting record revenue and record stock prices. After its layoffs, Meta's stock rose more than 150% in 2023. Wall Street told every CEO, with real money: layoffs + an AI narrative = a rising stock. Layoffs are no longer a signal of crisis; they are a medal for embracing AI.

This is not just a layoff wave. It is a revolution, and the cut positions are not coming back.

The difference from the previous three revolutions: each of those had the human brain as the backstop. Human intelligence was irreplaceable in the first three industrial revolutions. This fourth one wants to replace human intelligence itself.

Fortunately, machines can only replace intelligence — they cannot take responsibility. You cannot ask a machine to be accountable. And, more ironic still, even though this machine cannot take responsibility, every time you ask it to work, you still pay for every answer, in tokens.

What Does the Future of AI Look Like?

Nobody knows. I believe even OpenAI, Anthropic, and Google cannot predict where AI will end up. But one thing is certain: the workers — the proletariat — will get hurt again in this wave.

If AI truly replaces human intelligence and society moves toward communism, UBI will become the new social foundation. But then where do citizens' rights come from? Cut off from social labor, how do we account for a person's rights? Should a human being enslaved by AI have social rights at all?

And a question about the present: do people brainwashed by tittytainment like TikTok deserve social rights?

EU announces plan to restrict social media access for under-15s

Hidden inside this grim topic is one lucky fact: AI can replace human intelligence, but it cannot consume society's productive output.

The end point of all production is consumption, and the economy is a closed loop: people work, earn wages, and spend those wages buying back what they produce. In the first three industrial revolutions, machines replaced muscle; the displaced moved to new jobs, kept earning, kept spending — the loop never broke. This time, if AI replaces intelligence itself, the displaced will struggle to find the next job. With no wages, who consumes what AI produces? AI does not eat, does not travel, does not buy a house.

This is why UBI is being discussed so seriously: not as a utopian welfare fantasy, but as a life-support option for the loop itself. But UBI has to be paid out of the wealth AI creates — so who holds the power to distribute it?

I believe AI will inevitably enter the physical world and replace a huge number of workers. When UBI really arrives, are we the ones being enslaved by AI?

What Should We Actually Do to Survive?

Is AI's thinking the same thing as human thinking? In my perception, yes. The human brain relies on hundreds of millions of neuronal synapses to decide the next action and the next sentence. At its core, it is an inference engine. The biggest difference between humans and AI today is memory: a human brain can retain and distill decades — even a century — of experience and knowledge; AI cannot yet.

But AI is overwhelmingly strong: even with its memory mechanism limited, it was born already holding a vast amount of knowledge — something no human can do. A human needs over a decade of training after birth to become a functioning adult.

I believe that one day, AI's memory will improve dramatically too.

But can we trust what AI produces? Ask yourself: when you write code today, do you review every line AI generates? Most people don't. Not because you trust AI, but because the sheer volume of AI output has exceeded the boundary of human review capacity. Fair enough. On the internet this is acceptable — the web produces countless pages every day, some useful, some not. But the physical world's tolerance for error is far lower than the internet's. That is exactly why the internet became the field where AI exploded.

So in the physical world, what is left for humans?

Back to the earlier conclusion: AI can generate everything, but it cannot take responsibility. And the concrete form of taking responsibility is review. The real world is not the internet; its fault tolerance is extremely low. Behind every bridge drawing, drug formula, financial statement, and legal document, there must stand a person who can be held accountable (because chopping off an AI's head accomplishes nothing). When AI monopolizes generation, what humans can and must do is verify it and sign off on its results. I believe reviewing AI's behavior and output is the only work left for the working class: we shift from producers to reviewers.

But review is far harder than generation. AI's mistakes tend to look perfectly correct; only someone who knows more than it does can catch its errors. That means review demands professional skill and long training — something not everyone can achieve.

The other path is to go deep into a field, so deep that AI needs you. AI's knowledge is average: broad but shallow. It shines where data is dense; in the sparse depths of a specialized field, it fails confidently. Take the field I know, PDF: the spec runs nearly a thousand pages, real-world malformed documents are endlessly weird, and AI-written parsing code that looks flawless collapses on edge cases — while someone who has read the PDF.js source spots the problem at a glance.

This path runs in two directions: inward, use AI as a research assistant, exploring the depths of a field through long conversations; outward, package what you learn into tools and hand them to other AIs. At that point you are no longer AI's competitor — you are its eyes and hands. Or you become the reviewer for AI in that field, returning to the path above.

Conclusion

Frankly, AI is terrifying for ordinary people: a human needs over a decade of schooling to become a functioning adult, while AI is born with all of it. In the first three industrial revolutions, humanity held everything together with its intelligence; this time, intelligence itself is what the revolution has come for.

Revolutions kill people. But after every revolution settles, the survivors stand on new means of production. This time, I hope the ones standing on top are us who learned to enslave AI — not us who were enslaved by it.

About me: founder of StreamPDF, working deep in the PDF domain and the AI domain. This article was created with AI assistance, reviewed by me, and I remain responsible for its content.

I am currently reading the PDF.js source in depth. If you are interested, check out the series:

• PDF.js Deep Dive 1: The Architecture of PDF.js
• PDF.js Deep Dive 2: getDocument & getPage
• PDF.js Deep Dive 3: getViewport & render
• PDF.js Deep Dive 4: MessageHandler
• PDF.js Deep Dive 5: PDFFetchStream and PDFWorkerStream
• PDF.js Deep Dive 6: PDFManager.initDocument