Decoding SAFI: What's Next for AI and the Job Market?

Large Language Models (LLMs) are poised to redefine the labor landscape. The Skill Automation Feasibility Index (SAFI) offers a concrete measure of this potential shift. SAFI evaluates four leading LLMs, LLaMA 3.3 70B, Mistral Large, Qwen 2.5 72B, and Gemini 2.5 Flash, across 263 tasks representing 35 skills from the U.S. Department of Labor's O*NET framework.

High Stakes, High Scores

Visualize this: Mathematics and programming emerge as the most susceptible skills to automation with SAFI scores of 73.2 and 71.8, respectively. On the flip side, active listening and reading comprehension are far less vulnerable, scoring 42.2 and 45.5. The disparity signals a shift, are we looking at a future where tech prowess reigns supreme?

The Inversion Dilemma

A curious phenomenon surfaces: a capability-demand inversion. Skills in high demand for AI-exposed roles are precisely those that LLMs underperform at, according to SAFI benchmarks. This suggests a misalignment between current AI capabilities and job market needs. What happens when the machines excel at what we least expect?

Augmentation Over Automation

Numbers in context: 78.7% of AI interactions analyzed are augmentative rather than purely automating tasks. This points to a future where AI enhances human abilities rather than replacing them entirely. With all four models converging to similar skill profiles, there's a clear indication that text-based automation is more about skill dependency than model variance.

While SAFI measures only text-based skill representations, it provides an essential snapshot. The data, all open-sourced, foreshadows a workplace where AI augments skillsets. But which skills will flourish, and which will fade? The chart tells the story, and it's just beginning to unfold.