OpenAI has unveiled a major update to ChatGPT that allows the AI chatbot to remember all past conversations with a user.
The new feature allows ChatGPT to tailor answers based on previous interactions, bringing context to text, voice and image-generation for the first time.
OpenAI boss Sam Altman said the introduction of memory aims to deliver an entirely personalised AI assistant for every ChatGPT user.
“We have greatly improved memory in ChatGPT – it can now reference all your past conversations,” he wrote in a series of posts to X.
“This is a surprisingly great feature imo, and it points at something we are excited about: AI systems that get to know you over your life, and become extremely useful and personalised.”
The feature is currently only available for paid tiers of ChatGPT, with Mr Altman say it would roll out immediately for Pro users, “and soon for Plus users”.
OpenAI said there were currently no plans to introduce it for non-paying users, though premium features typically make their way into the free version as the platform evolves.
Subscribers in the UK, EU, Iceland, Liechtenstein, Norway, and Switzerland will also have to wait for the new memory feature, as OpenAI seeks to meet regulatory requirements.
Mr Altman said there will be the possibility to opt out of the new memory feature with ChatGPT, while users will also be able to chose a “Temporary Chat” option if they do not want a particular interaction saved.
Google recently launched a similar feature for its Gemini AI chatbot, just days after cyber security researchers discovered a way to corrupt its long-term memory by injecting false memories.
This demonstrated the potential to trick the AI into permanently “remembering” false information or instructions, which would impact the factual accuracy for future interactions. Google has since mitigated this type of attack.
