ChatGPT gains basic memory feature with new update

Artificial intelligence research institute OpenAI has introduced a new update to its popular app ChatGPT, adding the ability to set custom prompts for all chats.

As reported by OpenAI, this new feature brings a kind of memory to AI systems, essentially allowing them to recall some context in their interaction with the user.

Called “Custom Instructions,” this update allows users to set preferences that affect how ChatGPT responds in conversations. This feature is in beta, initially available to Plus plan users, with a broader rollout planned for the coming weeks.

This effort was born out of feedback from users who complained about starting each conversation from the beginning. Having engaged with users around the world, OpenAI now focuses on steerability in adjusting models to the diverse contexts and needs of individuals.

This new feature ensures that every conversation takes user preferences and requirements into account, eliminating the need for repeated instructions.

Benefits of Custom Instructions

For example, a data scientist working on a complex machine learning project can tell ChatGPT to remember that they are working with a particular dataset and need help cleaning the data and predicting the outcome.

Authors using ChatGPT to generate plot ideas for their novels can now specify that they are writing a sci-fi book set in a post-apocalyptic world, along with character and logline details. The model remembers these details and all subsequent suggestions are tailored to this genre and setting.

Additionally, crypto traders may want to discuss the latest trends in Bitcoin prices. Users can set the AI to remember their interest in Bitcoin, and future market analysis generated by ChatGPT will revolve around Bitcoin, eliminating the need to specify their interest every time.

Plugin description

Despite being in beta stage, there are already indications that the Custom Instructions feature can enhance plugin usage and provide contextual information that can be applied to different plugins. However, OpenAI warns that ChatGPT may not always be able to interpret these instructions correctly at this time.

By adding custom instructions related to the plugin, people interested in analyzing data such as charts and price movements can have the plugin retrieve the data without specifically asking for it at each prompt.

Privacy issues are also addressed. OpenAI takes steps to prevent the storage of user instructions that violate our Usage Policy. A common way to “jailbreak” ChatGPT and turn off its security features is probably part of the intended focus of this strategy. Additionally, steps are taken to remove personal identifiers in custom instructions to improve model performance.

The integration of the basic “memory” function is a remarkable step forward, but the true potential of this function will only become apparent as users continue to explore and operate this newly enhanced tool.

Poor performance

Despite the positive progress made by OpenAI, concerns have been raised about ChatGPT’s capabilities. According to recent information, Research papers from UC Berkeley and Stanford UniversitySome users claim that the underlying technology of ChatGPT Plus, GPT-4, is significantly underperforming.

Researchers compared GPT-4 to its predecessor, GPT-3.5, and found that performance declined over time. For example, when asking whether 17077 is prime, ChatGPT’s accuracy dropped by 95.2%.