Study shows why simulated reasoning AI models don't yet live up to their billing arstechnica.com 5 points by furcyd 7 hours ago