Research-backed, actionable articles on AI at work, endurance training and how people learn. Every claim has a source.
A 38-trial meta-analysis puts the gain at 1.7% on one-minute efforts, and a 40-minute run trial found nothing. The dose, the timing and the gut test.
Read the article →Nominal groups out-produce brainstorming teams on count and on quality. Eight studies on production blocking, the hybrid structure, and a session plan.
Read the article →A 10-minute nap lifted alertness for 155 minutes and a 45-minute one added 9% to shuttle-run distance. Ten studies on nap length, timing and sleep inertia.
Read the article →Benchmarks said models reason worse in French. Google found most of that gap was a broken test set. The part that survives changes how you write prompts.
Read the article →Two preregistered replications flattened the famous pen-versus-keyboard effect. A 2024 meta-analysis of 24 studies found a smaller one that hinges on review.
Read the article →Creatine kinase rose 340% after a first downhill run and 63% after the second. Eight studies on eccentric preconditioning, the dose, and the timing.
Read the article →Two rounds of self-review dropped GPT-4 from 95.5% to 89.0% on GSM8K. Eight studies on why models cannot grade their own reasoning, and the external checks that do work.
Read the article →Preparing to teach lifts recall on its own. The gain survives a week only when the explanation comes from memory. Four experiments, two meta-analyses, one protocol.
Read the article →Beetroot juice bought club cyclists 2.8% over 4 km, and nothing at all at 72 mL/kg/min. The dose, the 2.5-hour window and the mouthwash mistake.
Read the article →Escalation keeps 30% to 40% of IS projects alive. The evidence on what stops them: kill criteria set in advance, a threshold you did not write, and a fresh reviewer.
Read the article →