mapped 11 AI skill tests. here's what I found.
wrote a breakdown of 11 different AI skill tests for the blog this week (aisa.to/blog/ai-test-11-ways-to-measure-ai-skills). wanted to understand what's actually out there.
short version: most are multiple-choice quizzes testing whether you know what RAG stands for. fine for awareness, tells you nothing about whether someone can actually use AI well.
this is the gap we're building into with aisa — conversation-based assessment that watches how you think, not what you memorise. 1,300+ assessments in and the data keeps showing quiz scores and practical ability barely correlate.
wildest stat from our data: safety skills average 45/100. people are getting decent at prompting but almost nobody has a process for verifying output or handling sensitive data in prompts. aisa.to/state-of-ai-fluency has the full breakdown.
Yep. In practice, “can they prompt?” and “can they reject a plausible wrong answer?” are very different skills. I’d score verification separately: asks for evidence, checks a second source/tool, and knows when not to use the output. That would make the safety gap much more actionable than one 45/100 number.