The new default ChatGPT model cuts hallucinations by 52.5% on high-stakes topics and introduces memory sources that show users exactly what context shaped their responses.
