Great review by Carl Hendrick!

Here are my notes

Does creativity require slow, deliberate thinking? Mostly not. Two-thirds of correct solutions came from fast intuition, not deliberation. What predicted creative success was well-structured knowledge networks, not thinking time.

Can AI generate reliable personalised feedback at scale? Yes, with caveats. Across ~600 learners studying statistics, practice with even basic correct/incorrect feedback outperformed video lectures by ~0.5 SD in 24% less time. But to generalise to new problems, learners needed explanatory feedback and sufficient prior knowledge. Takeaway: practice works, but shallow feedback only gets you shallow gains.

Can AI feedback build teacher confidence as well as a human mentor? Not quite. Both AI and human feedback improved self-efficacy, but teacher feedback produced larger gains in student engagement efficacy and was rated more credible.

Students know interleaving works better, and still choose the easy option. Even when reframing made students believe that the harder mixed-practice strategy was more effective, they still defaulted to easier blocked practice. Belief doesn't change behaviour. If you want to use effective strategies, build them into your system rather than relying on willpower in the moment.

Does slower reading mean deeper comprehension? No. Eye-tracking showed that students who reread everything indiscriminately performed worse. What actually predicted comprehension was targeted rereading. In other words: going back to the specific sentence that didn't make sense. "Read more carefully" is bad advice. "Stop, question, recheck that specific part" is better.

AI feedback is getting surprisingly good, but with guardrails. GPT-4o generated personalised feedback at 90–98% accuracy with zero hallucinations in a statistics learning study. But it required serious prompt engineering and worked best paired with human-designed visual explanations. The realistic near-term model is AI handling adaptive text while humans design the scaffolding; something I’ll also explore in tomorrow’s post.