OpenAI trained o1 and o3 to ‘think’ about its safety policy https://t.co/JDj7t23pWw https://t.co/QLQIzv9yfm on https://twitter.com/AcerboLivio/status/1870925782832292093