Karpathy urgently deleted its database! An AI-generated doomsday image went viral, threatening the jobs of 60 million white-collar workers.

Author: New Zhiyuan

The "workplace verdict" of the AI ​​era: Will 60 million people lose their jobs?

Last night, AI guru Karpathy launched a wildly popular project—karpathy.ai/jobs/—which provides an in-depth analysis of the extent to which AI is "eroding" employment.

He extracted 342 occupations from the U.S. Bureau of Labor Statistics (BLS) and assigned a score (0-10) to each job regarding the risk of being replaced by AI.

The results were alarming, with the industry-wide average exposure score reaching as high as 4.9.

In particular, all "screen-dependent" professions are in dire straits, as they are basically within the reach of AI—

• Software developers 9/10

• Medical Transcriptionist 10/10

• Lawyer 8/10

• 9/10 ordinary office worker

Statistics show that approximately 60 million jobs are in the high-risk range, meaning 42% of them have a risk score of 7 or higher, with a total annual salary of $3.7 trillion.

What jobs are the safest? The answer is cleaners, plumbers, and roofers; those involving complex physical labor become the safest havens.

Hinton once suggested: Go become a plumber.

Musk commented scathingly, "In the future, all jobs will become optional."

Some netizens have even compiled a video that gathers the opinions of AI experts on predicting unemployment.

The 60 million white-collar jobs across the US are truly in danger!

This project went viral online, but Karpathy deleted the post just minutes after it went live, and it's now showing a 404 error on GitHub.

Fortunately, AI influencer Josh Kale cloned the entire repository before it went offline.

As you can see, on the far left of the project homepage, all the key metrics are marked, including exposure and salary.

The Gemini Flash index of 342 occupations and 143 million jobs across the United States shows an average exposure level of 4.9 across all occupations.

Link: https://joshkale.github.io/jobs/

Of these, the jobs most affected (6-10) accounted for 42%, or 59.9 million; while the jobs least affected (0-1) accounted for only 4%, or 6.2 million.

Jobs with an annual salary of over $100,000 (scoring 6.7) are more likely to be replaced by AI; while those with an annual salary of less than $35,000 are least affected (scoring 3.4).

Moreover, professions requiring a bachelor's degree are most vulnerable to the impact of AI.

Overall, AI is precisely targeting jobs based on "information processing density".

White-collar clerical positions that rely on word processing, data analysis, coding, and standardized processes, regardless of their salary, are collectively facing a crisis.

Conversely, positions involving physical operations, complex interpersonal interactions, or requiring real-time on-site judgment remain in the safe zone.

White-collar worker massacre

On the right side of the homepage, similar professions are grouped together.

Let's start by compiling a list of job positions with an AI exposure index of 6 or higher.

The lower left corner area mainly features office and administrative positions, all of which scored above 7 points, including clerks and receptionists.

Moreover, their median annual salary generally fluctuates around 43,000 yuan, and the educational requirement is basically a high school diploma.

For example, the median annual salary for office clerk positions (9/10) is $43,630, with a job size of 2.6 million.

Financial Clerk (9/10), Median Annual Salary: $48,650, Job Size: $1.2 million.

The core responsibilities of these positions are mostly routine tasks such as data entry and document formatting. They are almost entirely digitized and routine, making them highly susceptible to the impact of AI automation.

The subcategories of "Business and Financial Operations" in the upper right corner are almost all in high demand.

These positions offer a median annual salary between $50,000 and $100,000 and require a bachelor's degree.

For example, financial analysts (9/10) have a median annual salary of $101,910 and a job size of 429,000.

This work is almost entirely "digitalized," including large-scale dataset processing, trend analysis, and report generation—precisely AI's forte.

Of course, computer-related jobs will also be significantly impacted by AI. After all, Dario Amodei predicted that AI would replace software engineers within the next 6-12 months.

As can be seen from the chart below, software engineers (9/10), computer systems analysts (8/10), and computer support specialists (8/10) are all in the high-risk range.

They hold a median annual salary of up to 130,000, yet they are among the most easily replaceable.

In addition, lawyers (8/10), data scientists (9/10), graphic designers (9/10), cashiers (7/10) and other jobs are also at high risk of being replaced by AI.

It is worth mentioning that medical transcribers have the highest risk of all positions.

Go become a plumber

Nowadays, the safest jobs are really those that involve "human interaction with physical entities".

In the interactive chart, it is clear that the large areas that are green are basically related to complex on-site environments and hands-on positions.

For example, in construction and professional construction jobs, the average exposure index is between 1 and 3, and these physical jobs must be done by humans.

Take plumbers, pipe fitters, and steam pipe fitters as examples. They only require a high school diploma, have a median salary of 62,970, and are the least likely to be laid off.

Its core work is "heavy physical labor", which not only requires quick hands and feet and strength, but also the ability to solve various emergencies in real time in complex and ever-changing environments such as narrow mezzanines or construction sites.

AI still can't handle those core installation and repair tasks.

Similarly, food service professions, including chefs, waiters/waitresses, bartenders, and food processors, are also in the safe zone.

In addition, professions such as hairdressers, animal care workers, cleaners, medical personal care workers, and material handling workers are less affected by AI.

In conclusion, Hinton's statement is becoming increasingly valuable.

The entire internet erupted in outrage, and Karpathy responded.

Last night, this chart quickly went viral online, with many predicting that white-collar workers were about to suffer.

Two weeks 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 indicates that AI currently covers 75% of the tasks assigned to computer programmers.

Following closely behind are customer service representatives, data entry clerks, and medical record specialists—all of whom are among the hardest hit by AI.

In contrast, about 30% of occupations remain largely unaffected, such as chefs, lifeguards, and dishwashers, because these jobs require a great deal of human physical labor.

However, the actual adoption rate of AI currently only accounts for a small fraction of the theoretically feasible capabilities of AI tools.

Because this image caused widespread panic on social media, Karpathy subsequently deleted the data.

He explained, "This is just a hobby project that I spent two hours on the weekend 'by feeling' writing code, and it has been over-interpreted by everyone."

Harvard confirms: AI is not just "killing" jobs.

The panic is real, but it is not the whole picture.

Harvard Business School professor Suraj Srinivasan, in collaboration with researchers from the Hong Kong University of Science and Technology and Ohio State University, published a groundbreaking working paper titled "Substitution or Complementarity? The Impact of Generative AI on the Labor Market," offering a more robust and complex answer.

Paper link: https://www.hbs.edu/ris/Publication%20Files/25-039_05fbec84-1f23-459b-8410-e3cd7ab6c88a.pdf

The research team directly pulled up a dataset covering almost all online job postings across the United States, and tracked the actual supply and demand changes for each job from 2019 to March 2025.

Let's look at the alternative side first.

Since the release of ChatGPT, the number of jobs hired for the top 25% of positions with the highest automation potential has decreased by an average of 95 per company per quarter, a drop of 17%.

The financial and technology industries are bearing the brunt, with "screen-based" jobs such as clerical workers, payroll clerks, medical transcribers, and telemarketers being systematically phased out by AI.

Now let's look at the reinforced side.

During the same period, the recruitment volume of the top 25% of positions with the highest potential increased by an average of 80 positions per company per quarter, representing a growth rate of 22%.

Microbiologists, financial analysts, and clinical neuropsychologists share a common characteristic: some of their work can be automated by AI, while others must rely on human experience, intuition, and social skills.

Behind these two sets of figures lies a sophisticated quantitative method.

The research team used GPT-4o to evaluate more than 19,000 specific tasks across more than 900 occupations. Based on whether AI could reduce task completion time by more than half, the tasks were divided into four levels: "no exposure," "direct exposure," "application exposure," and "image exposure." The team then combined the importance weight of each task within the job to calculate the "automation score" and "enhancement score" for each occupation.

The skill-based differentiation is even more alarming.

In highly automated jobs, the demand for AI-related skills has plummeted by 24%, and the overall skill requirements have also shrunk accordingly, with the frequency of new skills emerging continuing to decline.

These positions are being "emptied out." As AI takes over most structured tasks, the remaining work becomes simpler and more standardized, and companies require less and less from human workers.

However, in high-potential roles, the trend is completely reversed. Demand for AI-related skills has increased by 15%, with both total skill requirements and the number of new skills being upgraded.

These roles have become more complex. Employees not only need to be proficient in using AI tools, but also need to have the ability to supervise AI output and integrate human-machine collaborative processes. Taking the financial industry as an example, investment managers and analysts use AI to process massive amounts of market data, but the final judgments and decisions still rest with humans.

AI hasn't treated all white-collar workers the same. It's more like a "career restructuring," where pure information carriers are eliminated, while those who can collaborate with AI become more valuable.

How much time is left in the window of opportunity?

Karpathy deleted the post, but the data couldn't be removed. The Harvard paper was more level-headed, but its conclusions were equally scathing.

Whether you look at Gemini Flash's rating system or empirical studies covering the entire US job market, they all point to the same fact: AI is already reshaping white-collar jobs.

However, it was not a one-size-fits-all massacre, but a process of division.

The positions that were cut were those whose job content could be fully described and whose processes could be broken down into standardized steps.

Those positions that remain and even become more valuable are those that require making judgments in ambiguous areas, building trust between people, and making final decisions based on AI output.

This division has a cruel consequence.

In the past, the first step in the career ladder for white-collar workers was often a standardized entry-level job, such as data entry, report writing, basic coding, and basic analysis.

Young people start here, doing repetitive work, gradually accumulating experience and judgment, and eventually growing into irreplaceable individuals.

Now, AI is removing 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 only one question that truly needs to be answered.

What percentage of your work is beyond the capabilities of AI?

If the answer makes you uneasy, then the time to act is not tomorrow, but now.