Karpathy jobs repo returns with AI treemap

The interesting part is not the viral heatmap, but the workflow behind it: Karpathy turned public labor data into a reusable prompt-driven analysis stack. That makes this feel less like a one-off infographic and more like a compact blueprint for AI-assisted research tools.

• –The repo cleanly separates scraping, parsing, scoring, and site generation, which makes it unusually hackable for anyone who wants to repurpose the pipeline.
• –The “Digital AI Exposure” layer is intentionally framed as a rough LLM estimate, not a rigorous forecast of which jobs disappear next.
• –Because the scoring prompt is swappable, the same system can be reused for other lenses like humanoid robotics, climate impact, or offshoring risk.
• –It is a good example of where LLMs shine today: fast synthesis over messy public data, with humans still responsible for interpretation and caveats.