Just caught something wild that's been circulating. A few weeks back, Karpathy dropped this interactive database analyzing which US jobs are most vulnerable to AI replacement, and it absolutely blew up. The man literally scored 342 different occupations on a 0-10 exposure scale. Then he yanked the entire project offline within hours. Too late though—the internet already had screenshots.

Here's what the data showed and why people freaked out. The average exposure score across all US jobs landed at 4.9 out of 10. But here's the thing: roughly 60 million jobs scored 7 or higher on that vulnerability scale. That's 42% of the workforce sitting in high-risk territory, representing about $3.7 trillion in annual salaries. Yeah, you read that right.

The pattern is actually pretty clear once you look at it. Anything screen-dependent is basically done. Software developers hitting 9/10, financial analysts at 9/10, data scientists at 9/10. Even lawyers scored 8/10. But the one that caught everyone's attention was medical transcriptionist salary data—these roles hit a perfect 10/10 vulnerability score. Medical transcriptionist salary ranges around $35,000-$45,000 annually, and the job is basically pure information processing. AI doesn't even need to think twice about that one.

Meanwhile, the safest positions? Plumbers, electricians, HVAC technicians, construction workers. Physical labor in unpredictable environments. Hinton actually suggested going into the trades, which is kind of hilarious given how many people spent years climbing the corporate ladder. Musk's take was even more blunt: "In the future, all jobs will become optional."

What really matters though is the pattern underneath. It's not random. Jobs requiring a bachelor's degree got hit harder. Higher salaries didn't protect anyone—if the work is information processing, you're exposed. Medical transcriptionist salary levels showed this perfectly: decent pay but completely automatable. Meanwhile, cleaners, food service workers, hairdressers, personal care workers? All in the green zone.

But here's where it gets more interesting. Harvard Business School actually did the real research on this. They pulled actual job posting data from 2019 through March 2025 and tracked what actually happened in the labor market. Turns out the story is more nuanced than "everyone's getting replaced."

Yes, hiring for the top 25% most automatable positions dropped 17% per company per quarter since ChatGPT launched. The finance and tech sectors got hit first. Clerical work, payroll positions, and yes, medical transcriptionist salary-range jobs are being systematically phased out. That part's real.

But simultaneously, hiring for positions with high AI-complementarity potential jumped 22% per company per quarter. Microbiologists, senior financial analysts, clinical specialists. These roles have something in common: AI can handle parts of the work, but the judgment, intuition, and human decision-making remain essential. The work didn't disappear—it transformed.

Here's the brutal part though. In jobs that are getting automated, companies aren't just replacing people with AI. They're "hollowing out" the roles. AI skill demand dropped 24% in these positions because there's just less work to do. The remaining tasks became simpler, more standardized. Entry-level positions are basically vanishing.

Meanwhile, in complementary roles, AI skill demand surged 15%. These jobs got more complex, not less. You need to understand AI tools, supervise outputs, integrate human-machine workflows. The skill floor actually went up.

This creates a weird dynamic. The traditional career ladder where someone starts in a basic data entry or reporting role, learns the ropes, and gradually becomes irreplaceable? That first rung is disappearing. The entry point is narrowing while the expertise gap is widening.

So what does this actually mean? It's not a straightforward massacre. It's a restructuring. Pure information carriers—people whose entire job is moving data around and following standardized processes—are getting displaced. But people who can make judgment calls in gray areas, who understand context, who can collaborate with AI rather than compete with it? They're becoming more valuable.

The real question for anyone still working is straightforward: What percentage of your actual work can AI do? If that number makes you uncomfortable, waiting isn't an option.