OpenAI seems to be trying to improve ChatGPT functionality regularly, now the Chatbot has a feature to remember its users.
As a reminder, one of the main tenets of the first wave of AI chatbots was that the technology did not have a continuous memory to remember who the user was, which caused everything to reset upon restarting the conversation from scratch.
However, OpenAI will change this, as the chatbot will remember who the user is from every conversation, according to a report from The Verge. However, on the other hand it can be both exciting and risky.
The feature being tested in beta for ChatGPT Plus subscribers is called Custom Instructions, and allows users to set unique, fixed parameters in each chat.
For example, the user tells the system that you are teaching third graders, so each answer will suit your student or provide information on how big your family is so that the chatbot provides a list of food recipes that suit your family.
Meanwhile, this feature will be set to work throughout platform, so third-party apps with ChatGPT will have this feature. This feature is also very useful on mobile devices, because if you repeat the previous question it will be very boring and annoying.
It’s worth noting that OpenAI touts this feature as a way to simplify queries, and not a first step toward an AI-powered personal assistant that anticipates our needs. As an illustration of this technology in the film Her where Scarlett Johansson becomes an AI personal assistant.
Apart from that, this feature also has obvious privacy issues which is why it is being rolled out as a beta so the company can work out some of the issues. Also, adding another layer of instructions complicates the query and causes the chatbot to just make things up.
This feature’s settings tab is governed by the same rules as this bot itself, so it won’t do anything bad. OpenAI provides examples such as trying to include the question of tips on killing people as a special instruction, but this doesn’t work.
In addition, it will also delete personal information that can be used to identify users, which has both good and bad sides. It’s no secret that tech companies can’t be trusted when it comes to personal data, but digital assistants until recently had no access to personal data.
An update that brings this feature is available as of today, though only to subscribed ChatGPT subscribers. The update is also not available in the UK and EU regions, the company expects the feature to roll out in those regions soon.