we hit 1,300 AI skills assessments and the data is starting to get interesting

building aisa.to -- an AI that assesses people's AI skills through conversation instead of multiple choice.

we just crossed 1,300 completed assessments and the patterns in the data keep surprising me. the average score is 52/100, which honestly tracks. but the breakdown is where it gets interesting: safety practices scored lowest at 45/100, and AI fundamentals (5.0/10) came in weaker than prompting (5.6/10).

people are decent at using AI tools but have almost no systematic approach to checking whether the output is actually correct. that verification gap is probably the single biggest risk in how organizations are adopting AI right now.

the other thing I didn't expect: 62% of people score below Proficient. not because they're bad at AI, but because the bar for actual proficiency is higher than most people realize. knowing how to prompt is table stakes -- the gap is in evaluation, safety, and understanding the limitations.

we published the full data breakdown at aisa.to/state-of-ai-fluency if anyone wants to dig into the methodology. also just shipped a new piece on the 11 different ways to actually measure AI skills (aisa.to/blog) because most of the existing approaches are basically glorified quizzes.

next milestone: 2,000 assessments. trying to get enough data to do proper cohort analysis by industry.