Problem
Your conversation quality with LLMs significantly drops the longer you have it.
They poison their own context. Maybe you can call it context rot, where as context grows and especially if it grows with lots of distractions and dead ends, the output quality falls off rapidly. Even with good context the rot will start to become apparent around 100k tokens (with Gemini 2.5).
Quote taken from Workaccount2 on Hacker News
Solution
ChatGPT web interface allows you to edit any message from your history, which essentially enables you to "fork" the conversation. This effectively drops an entire branch of the conversation you had, allowing you to remove unnecessary context from your chat.
You can always browse your other branches by hovering over the message that was edited and caused another conversation branch.
You can do multiple branches at any point of your conversations.
If you see that the conversation is going to a dead end or you have identified additional important context, just go back to the previous message and edit it by adding those useful bits of information.
Benefits
You fix your “context rot” problem, and your conversations are the peak of possible quality.
You keep whole conversation in a single place without need starting from scratch
Responses are faster
Limitations
Unfortunately, this seems to be available from the ChatGPT web interface. MacOS app or mobile app allow you to do edits but don’t allow you to browse alternative branches of conversation.
Shared conversations don’t include other branches. The conversation fork being shared is the one currently selected during sharing.