One of the defining aspects of the first generation of AI chatbots is their absence of a persistent memory – each conversation effectively starts from scratch. However, OpenAI’s ChatGPT is set to change this norm by retaining conversational memory, according to a recent report by The Verge. While this holds exciting prospects for seamless interaction, it also carries inherent risks.
Currently under testing, the new “custom instructions” feature, accessible only by opt-in beta to ChatGPT Plus subscribers, enables users to establish unique parameters that are preserved across conversations. For instance, users could instruct the system that they teach third-grade students, and henceforth, all responses provided by the bot would be tailored for a young audience. Alternatively, users could specify the size of their family, and the bot would then offer accurate recipe measurements in subsequent interactions.
This feature is expected to be platform-wide, meaning third-party apps incorporating ChatGPT should eventually access it. This could prove particularly beneficial for mobile users, where the need to repeatedly provide the same information can be an inconvenience. However, OpenAI emphasizes that this feature is intended to streamline queries rather than form the basis of an all-encompassing AI personal assistant akin to Scarlett Johansson’s character in ‘Her.’
Naturally, the introduction of this feature raises significant privacy concerns, warranting its beta launch to allow for further fine-tuning. Moreover, adding another layer of instructions could potentially complicate queries, increasing the chances of bots generating unfounded responses. Given its beta status, users should not expect perfection at this stage.
The ‘custom instructions’ settings tab will adhere to the same guidelines as the bot itself, thereby ensuring that inappropriate requests or instructions containing personal identifiers will be omitted. While this serves to protect user privacy, it also brings to the fore the paradox of creating advanced digital assistants – the need for such technology to access personal data against the backdrop of the tech industry’s questionable track record in handling such information.
The feature update is set to roll out today, though initially only available to paying ChatGPT subscribers. Furthermore, due to compliance considerations, it will not be accessible in the UK and EU at launch, although OpenAI aims to expand its availability to these regions soon.
