A feature I desperately want: the ability to curate / edit earlier messages in a chatbot thread, to better direct future output from the model.

I want to be able to prune dead ends or weird suppositions from the chatbot to keep the rest of the conversation from deteriorating.

This is, on a technical level, easy, since each time you get a new message from the LLM, it sends the whole past chat to the model–the model doesn't care if you tweaked what it ostensibly said in the past.

Why does this feature not exist in the main chatbots?

Because it's too niche a feature, too pro, too confusing.

ChatGPT is already hyper-scale, which means it needs to focus on the marginal user and put all of its effort into making it easier to use for low intent users.

The tyranny of the marginal user.