AI now grades its own understanding of what you know — and gets it right 83 to 97 percent of the time

Researchers built an AI system that listens to a person answer questions and instantly figures out whether they're a novice, intermediate, advanced, or expert in that domain. The system works by having another AI conduct the interview, scoring each response in real time rather than waiting until the end, and comparing its assessment against what people say about themselves.

This is a small change with obvious downstream uses: customer service chatbots that switch to simpler language when they detect confusion, tutoring systems that adjust difficulty on the fly, recruitment tools that assess candidate expertise during a conversation rather than after a test. The accuracy matters because it's measurable and specific. Right now, most systems either ask users directly ('What's your skill level?' — people lie) or use crude proxies like education level or job title. This one learns from what someone actually says. That's a threshold: once you can measure someone's expertise continuously and accurately during an interaction, you can adapt to it continuously. The flip side is less visible but real: every conversation now generates an expertise profile that a system can use, store, and act on.