New study shows why simulated reasoning AI models don't yet live up to billing arstechnica.com 3 points by pseudolus 20 hours ago