Source: New Intelligence Yuan (New智元)
Editors: Taozi, Haokun
The "workplace verdict" of the AI era: Are 60 million people about to lose their jobs?
Last night, AI luminary Andrej Karpathy launched a viral project at karpathy.ai/jobs/, providing an in-depth review of the extent to which AI is "eroding" employment.
He extracted data on 342 occupations from the U.S. Bureau of Labor Statistics (BLS) and assigned an AI substitution risk score (0-10) to each role.
The results were alarming: the average exposure score across all industries reached a staggering 4.9.
Especially, "screen-dependent" professions are in full crisis mode, falling almost entirely within AI's range:
- Software Developers: 9/10
- Medical Transcriptionists: 10/10
- Lawyers: 8/10
- General Office Clerks: 9/10
Statistics show that approximately 60 million jobs are in the high-risk zone, meaning 42% have a risk score above 7. The total annual salary sum for these roles amounts to $3.7 trillion.
So, which jobs are the safest? The answer lies with janitors, plumbers, and roofers. Occupations involving complex physical labor have become the safest havens.
Hinton once suggested: Go become a plumber.
In response, Elon Musk commented sharply, "In the future, all work will be optional."
Some netizens even compiled a video collecting predictions of unemployment from major AI figures.
60 Million White-Collar Jobs in the US Really in Danger!
Although this project went viral across the internet, Karpathy deleted the post just minutes after launch, and the link now leads to a GitHub 404 error.
Fortunately, AI influencer Josh Kale cloned the entire repository before it went offline.
As seen on the project's homepage, the leftmost section labels all key metrics, including Exposure and Salary.
Out of 342 occupations and 143 million jobs in the US, scored by Gemini Flash, the average exposure level for all professions reached a high of 4.9.
Link: https://joshkale.github.io/jobs/
Among them, roles most affected (scores 6-10) account for 42%, totaling 59.9 million jobs. Those least affected (scores 0-1) make up only 4%, amounting to just 6.2 million jobs.
Jobs with annual salaries exceeding $100,000 (scoring 6.7) are more easily replaced by AI, while those earning under $35,000 are least affected (scoring 3.4).
Furthermore, professions requiring a bachelor's degree are the most vulnerable to AI impact.
Overall, the trend indicates that AI is precisely targeting jobs based on "information processing density."
White-collar clerical roles relying on text processing, data analysis, coding, and standardized processes are collectively flashing "red," regardless of their high salaries.
Conversely, jobs involving physical operations, complex human interaction, or requiring on-the-spot judgment remain in the safety zone.
The Great Slaughter of White-Collar Jobs
On the interactive area of the homepage, occupations with similar natures are grouped closely together.
Let's first tally the jobs with an AI exposure index above 6.
In the bottom-left region, we mainly find office and administrative roles, all scoring above 7, including clerks and receptionists.
Moreover, their median annual salary hovers around $43,000, generally requiring only a high school diploma.
For instance, Office Clerks (9/10) have a median annual salary of $43,630 and a workforce size of 2.6 million.
Financial Clerks (9/10) have a median annual salary of $48,650 and a workforce size of 1.2 million.
The core duties of these roles mostly involve routine tasks, data entry, and document formatting. Having become almost entirely digitalized and routine, they are extremely susceptible to AI automation.
In the top-right,细分 roles under "Business and Financial Operations" are almost entirely in the red zone.
These positions have median annual salaries between $50,000 and $100,000 and typically require a bachelor's degree.
For example, Financial Analysts (9/10) have a median annual salary of $101,910 and a workforce size of 429,000.
The content of this work is almost "fully digitalized," including processing large datasets, trend analysis, and report generation—areas where AI excels.
Of course, computer-related jobs are also significantly impacted by AI. After all, Dario Amodei once predicted that AI would replace software engineers within 6 to 12 months.
As seen in the chart below, Software Engineers (9/10), Computer Systems Analysts (8/10), and Computer Support Specialists (8/10) are all in the high-risk zone.
Despite holding median salaries as high as $130,000, they are among the most easily replaceable groups.
Additionally, roles such as Lawyers (8/10), Data Scientists (9/10), Graphic Designers (9/10), and Cashiers (7/10) all face high risks of AI substitution.
Notably, Medical Transcriptionists represent the highest risk among all occupations.
Go Become a Plumber
Nowadays, the safest careers are truly those where "hands interact with physical entities."
In the interactive chart, it is clear that the large green areas basically relate to complex on-site environments and hands-on operational roles.
As shown below, construction and specialized trade occupations have average exposure indices between 1 and 3; these physical jobs must be done by humans.
Take Plumbers, Pipefitters, and Steamfitters, for example. Requiring only a high school diploma with a median salary of $62,970, they are the least likely to be eliminated.
Their core work involves "heavy physical labor," requiring not only dexterity and strength but also the ability to solve various unexpected situations in real-time within narrow interlayers or changing construction sites.
AI simply cannot perform these core installation and repair tasks yet.
Similarly, food service professions, including chefs, waiters, bartenders, and food processors, also remain in the safety zone.
In addition, barbers, animal care workers, janitors, personal healthcare providers, and material movers in transportation are less impacted by AI.
In short, the value of Hinton's statement continues to rise.
The Internet Explodes; Karpathy Responds
Last night, once this chart appeared, it quickly went viral online, with many predicting that white-collar workers were in for trouble.
Half a month ago, Anthropic also released a report titled "The Impact of AI on the Labor Market: New Metrics and Early Evidence."
Similar to Karpathy's data, the report pointed out that currently, AI coverage for computer programmer tasks is as high as 75%.
Following closely are customer service representatives, data entry clerks, and medical records specialists; these are the "disaster zones" of AI impact.
In contrast, about 30% of occupations are basically unaffected, such as chefs, lifeguards, and dishwashers, because these jobs require significant human physical collaboration.
However, the current actual adoption rate of AI represents only a small fraction of the theoretical capabilities of AI tools.
Precisely because this chart triggered massive panic on social media, Karpathy subsequently deleted the data in an emergency.
He explained, "This was just a hobby project I whipped up over the weekend in 2 hours 'by feel,' and it has been over-interpreted by everyone."
Harvard Confirms: AI Isn't Just "Killing" Jobs
The panic is real, but it is not the whole picture.
Harvard Business School professor Suraj Srinivasan, along with researchers from the Hong Kong University of Science and Technology and Ohio State University, published a major working paper titled "Substitute or Complement? The Impact of Generative AI on the Labor Market," offering a harder and more complex answer.
Paper Link: https://www.hbs.edu/ris/Publication%20Files/25-039_05fbec84-1f23-459b-8410-e3cd7ab6c88a.pdf
The research team utilized a dataset covering almost all online job postings across the US, tracking real supply and demand changes for each posting from 2019 to March 2025.
First, looking at the substitution aspect.
After the release of ChatGPT, hiring for the batch of jobs with the highest automation potential (top 25%) decreased by an average of 95 positions per quarter per company, a drop of 17%.
The finance and technology sectors were hit first. Roles like document processors, payroll clerks, medical transcriptionists, and telemarketers—these "screen-brick-laying"工种—are being systematically cleared out by AI.
Then, looking at the augmentation aspect.
During the same period, hiring for the batch of jobs with the highest augmentation potential (top 25%) increased by an average of 80 positions per quarter per company, a rise of 22%.
Microbiologists, financial analysts, and clinical neuropsychologists share a common trait: part of their work can be accelerated by AI, while another part still relies on human experience, intuition, and social skills to manage.
Behind these two sets of numbers lies a precise quantification method.
The research team used GPT-4o to evaluate over 19,000 specific tasks across more than 900 occupations one by one. Based on whether AI could reduce task completion time by more than half, they classified them into four levels: "No Exposure," "Direct Exposure," "Application Exposure," and "Image Exposure." Combining this with the importance weight of each task within a job, they calculated an "Automation Score" and an "Augmentation Score" for each profession.
The differentiation at the skill level is even more startling.
In highly automated jobs, demand for AI-related skills plummeted by 24%, and total skill requirements contracted simultaneously, with the frequency of new skills emerging continuing to decline.
These jobs are being "hollowed out." As AI takes over most structured tasks, the remaining work becomes simpler and more standardized, and corporate demand for humans decreases.
Conversely, in jobs with high augmentation potential, the trend completely reverses. Demand for AI-related skills grew by 15%, and both total skill requirements and the number of new skills are climbing.
These jobs are becoming more complex. Employees must not only know how to use AI tools but also possess the ability to supervise AI outputs and integrate human-machine collaboration processes. Taking the financial industry as an example, investment managers and analysts use AI to process massive amounts of market data, but the final judgment and decision-making remain in human hands.
AI did not indiscriminately slash all white-collar jobs. It is more like a "career restructuring": pure information porters are eliminated, while those who can collaborate with AI are becoming more valuable.
How Much Time is Left in the Window?
Karpathy deleted the post, but the data cannot be erased. Harvard's paper is calmer, but its conclusion is equally unsparing.
Whether you look at Gemini Flash's scorecard or the empirical research covering the entire US hiring market, the pointer leads to the same fact: AI's restructuring of white-collar jobs is already happening.
It is just not a one-size-fits-all massacre, but a divergence.
What is being cut are jobs where the work content can be fully described and processes can be standardized and dismantled.
What remains, or even becomes more valuable, are roles that require making judgments in gray areas, building trust between people, and making final decisions based on AI outputs.
This divergence brings a cruel consequence.
In the past, the first step on the white-collar career ladder was often standardized entry-level work: data entry, report writing,初级 coding, and basic analysis.
Young people started here, doing repetitive work, slowly accumulating experience and judgment, and eventually growing into irreplaceable individuals.
Now, AI is pulling away this first step.
The entrance has narrowed, but the reward at the finish line is even greater.
For everyone still in the workforce, there is truly only one question that needs answering.
What percentage of your work is something AI cannot do?
If the answer makes you uneasy, the time to act is not tomorrow, but now.
References:
https://x.com/_kaitodev/status/2032927164883153402?s=20