According to @karpathy, earlier predictions that computer vision would quickly eliminate radiology jobs have not materialized, with the field growing rather than shrinking. Source: @karpathy on X, Sep 25, 2025. According to @karpathy, the reasons include narrow benchmarks that miss real-world complexity, the multifaceted scope of radiology beyond image recognition, deployment frictions across regulation, insurance and liability, and institutional inertia. Source: @karpathy on X, Sep 25, 2025. According to @karpathy, Jevons paradox applies as AI tools speed up radiologists, increasing total demand for reads rather than reducing it. Source: @karpathy on X, Sep 25, 2025. According to @karpathy, AI is likely to be adopted first as a tool that shifts work toward monitoring and supervision, while jobs composed of short, rote, independent, closed, and forgiving tasks are more likely to change sooner. Source: @karpathy on X, Sep 25, 2025. For traders, this framing highlights gradual AI integration and expanding workloads in regulated, high-risk domains, a narrative relevant to AI-linked equities and AI-themed crypto projects tied to compute utilization. Source: @karpathy on X, Sep 25, 2025. Full post reference is the Works in Progress article shared by @karpathy. Source: @karpathy on X, Sep 25, 2025.