Hello OpenAI team and community,
I would like to propose a feature that could significantly improve ChatGPT’s performance in long conversations: user-controlled message virtualization and automatic hiding of older messages.
1. The problem: Long conversations become increasingly difficult to use
As a conversation grows, the ChatGPT web interface can become slow, laggy, or even temporarily freeze. I recently experienced this problem myself while using ChatGPT.
This is not only an issue for users with older or low-end computers. Even powerful devices have finite resources, and continuously rendering an ever-growing conversation can create performance problems over time.
Long conversations are often extremely valuable. Users may keep months of learning, research, programming, project development, personal workflows, and important contextual information in a single thread.
Being forced to abandon such a conversation because the interface becomes too slow is not a satisfactory long-term solution.
2. Why existing workarounds are insufficient
Starting a new chat can disrupt continuity. The original conversation may contain detailed context, previous decisions, corrections, preferences, and a naturally developed working relationship with the assistant. Projects, Memory, and Custom Instructions can help preserve some information, but they are not necessarily equivalent to continuing the exact same conversation.
Clearing the cache, disabling extensions, using incognito mode, or repeatedly refreshing the page may help diagnose certain performance problems, but these measures do not necessarily address the underlying cost of rendering a very long conversation.
Similarly, loading messages in smaller batches is useful, but it is not necessarily the same as limiting how many messages remain rendered in the page at once.
3. My proposed solution: User-controlled message virtualization
I suggest implementing a system that separates the complete stored conversation history from the messages currently rendered in the browser.
For example:
-
Choose the visible message window: Allow users to display only the latest 5, 10, 20, or another chosen number of messages.
-
Automatically hide older messages: As new messages arrive, older messages should be removed from the active rendered interface rather than remaining indefinitely in the page.
-
Preserve the complete conversation: Older messages must remain saved and accessible. Hiding them must not delete them.
-
Restore older messages on demand: Users should be able to click “Show Previous Messages” or scroll upward to retrieve earlier parts of the conversation.
-
Remember the user’s preference: After refreshing or reopening a conversation, ChatGPT should restore the selected display limit rather than rendering the entire history immediately.
-
Keep AI context intact: Hiding messages from the visible interface should not automatically remove the information the model needs to maintain conversational continuity. The conversation’s context-management system can continue operating independently of the browser’s rendering strategy.
Ideally, this feature would be optional, with a simple control that allows users to choose how much of the conversation remains visible.
4. Why this could be more effective
The central idea is simple: a conversation can remain complete without every message being rendered on the screen simultaneously.
Virtualization and on-demand rendering are established techniques used in web applications to manage large collections of content. Applying these techniques comprehensively to long ChatGPT conversations could reduce unnecessary rendering work, improve scrolling and typing responsiveness, and potentially reduce memory consumption.
The actual performance improvement would depend on the implementation and on the source of each slowdown, but this approach directly addresses a major potential source of interface overhead.
5. The long-term goal
I do not want users to have to choose between a fast interface and a valuable, long-running conversation.
I want users to be able to maintain one continuous workspace for their learning, research, and projects without being forced to start over, repeatedly clean up their browser, or reorganize their work across multiple chats simply because the conversation has grown.
My suggestion is not to delete old messages or sacrifice the assistant’s context. It is to give users control over what the interface displays while preserving their complete conversation history.
I would appreciate feedback from the community on whether this approach is already fully implemented in ChatGPT and whether further improvements to user-controlled message virtualization could be considered.
Thank you for taking the time to read this suggestion. I hope it contributes to making ChatGPT more responsive, accessible, and reliable for users across all devices.