I'm excited about ChatGPT's memory upgrade - but I'm quickly seeing a downside (www.zdnet.com)

OpenAI has rolled out an upgrade to ChatGPT's memory capabilities, enhancing its ability to maintain a profile based on users' past interactions. This new functionality, known as Dreaming V3, allows ChatGPT to analyze and utilize historical chat data to provide more personalized responses. As of 2026, the success rate for recalling factual tasks has improved dramatically, jumping from 41% to 82%, and user preference adherence has increased from 31% to 71%. However, concerns are mounting within the AI/ML community about potential inaccuracies and the implications of retaining outdated or irrelevant information. Users have reported instances where the AI incorrectly synthesizes details, leading to confusion and distrust. The significance of this upgrade lies in its attempt to create more contextually aware AI interactions, yet it raises serious questions about privacy, data retention, and the potential for biased responses. Users cannot fully erase past memories, even when turning off the memory feature, which could lead to a skewed representation of one's identity as perceived by the AI. The fact that ChatGPT’s memory encompasses detailed profiling means it could inadvertently misinterpret user queries based on prior conversations that do not reflect current interests or contexts. This development underscores the delicate balance between enhancing AI capabilities and ensuring user trust and safety.