This week Anthropic (the company behind Claude) released a new report that measured not just if people use AI, but how well they collaborate with it.

Two insights I find interesting:

• The strongest indicator of AI skill is iteration. People who use claude with back-and-forth conversations show double the "fluency behaviors" (like checking for missing context) compared to people who use it like Google (one-shot requests).

• When people ask AI to generate finished artifacts (code, full documents, apps), they become less critical. While people who use Claude become very specific about formatting, they are statistically less likely to fact-check the AI or question its reasoning.

Implications for lifelong learners:

• Never accept the first explanation or summary Claude gives you. Ask follow-up questions, ask it to explain its logic, or ask it to rephrase concepts you didn't understand.

• When Claude creates something that looks done (e.g., a clean Python script or a formatted history essay), be even more critical about it. Actively look for what might be missing or hallucinated. Do not equate formatting with accuracy.