Andrej Karpathy, co-founder of OpenAI, briefly released an interactive visualisation on March 15 that explored how artificial intelligence might affect different jobs across the United States. The project analysed 342 occupations listed by the U.S. Bureau of Labor Statistics before Karpathy removed the site just a few hours after it went live, citing widespread misunderstanding of the data.
Even though the original page was taken down, copies of the tool and archived versions of the site continue to circulate online. The visualisation quickly attracted attention across social media, with many users sharing and debating the results.
Karpathy later explained that the project was simply a quick experiment built in about two hours on a Saturday morning. He described it as a small “vibe-coded” project intended to help people explore labour data visually. The code and dataset were made public so others could experiment with the information or build their own visualisations. However, after noticing that the results were being widely misinterpreted, he decided to remove the site.
The tool examined roughly 143 million jobs in the U.S. economy and assigned each occupation an AI exposure score ranging from 0 to 10. A higher score indicated that tasks within that profession might be more affected by AI tools. Across all jobs, the weighted average exposure score came out to around 4.9.
According to the analysis, about 42% of jobs, representing roughly 59.9 million positions with combined annual wages of around $3.7 trillion, received scores of 7 or higher. These scores suggested that a significant portion of tasks in those roles could potentially be influenced by AI technologies.
The project also highlighted a clear difference between income levels. Occupations paying more than $100,000 per year averaged an exposure score of 6.7, while jobs earning less than $35,000 annually averaged 3.4.
Many digital-heavy white-collar roles ranked among the most exposed. Jobs such as software developers, data analysts, accountants, lawyers, financial professionals, writers, and medical transcriptionists typically scored between 8 and 10. Meanwhile, more physically oriented occupations like roofers, janitors, and home health aides appeared near the bottom of the scale with scores close to one or below.
Karpathy emphasised that the scores were not predictions about job losses. The visualisation and its GitHub documentation included a disclaimer explaining that the ratings were rough estimates based mainly on how digital a job’s tasks are. Real-world outcomes, he noted, depend on many other factors such as market demand, regulations, and social preferences.
Despite those warnings, the map quickly spread online, with many people interpreting it as a forecast of which jobs might disappear due to AI.
The discussion also drew comments from Elon Musk, CEO of Tesla and SpaceX. Musk suggested that advances in AI could eventually make work optional, adding that widespread automation might lead to a future with universal high income rather than economic disruption.
Also Read: Equinix Introduces Distributed AI Hub to Simplify and Secure Enterprise AI Infrastructure
