I want to hear other people’s thoughts using long AI chats vs making a new chat for every edit/change.
I’ve had a mix of experiences with this but the longer chats seem to do better than the new chat for each edit, unless the edit has nothing to do with my previous chat.
For example, if i’m fixing a problem or adding a new feature , keeping the same chat open until it’s done seems to work better for me but ends up in very long chats which lag/crash cursor (but hopefully fixed in 1.4.5?). I have to explain to the model about my code and what i want it to do because i’m doing things it conceptually doesn’t understand (i.e. not in the training data). So this is where the longer chat seems to work wonders.
However, if in the long chat I want to edit something else that i haven’t talked about in the chat it seems to not be as good, so starting a new chat in that instance is better. I would like to have multiple chat tabs open with different models (which i believe they just added as a feature) and not have it lag out the whole program (which again i hope is fixed now but i haven’t tested yet). This way i could keep the long chats for the certain feature it is working on or separate chats by feature.
I have yet to be able to start a new chat and have it understand the concepts of my codebase, no matter how much documentation and rules and logic workflows, diagrams, etc i create, it just does not understand in a new chat without me having to correct it over and over. I’ve resorted to making a single rules file with everything that it makes mistakes on or doesn’t know in new chats to try and limit this, it seems to be helping. This is all a work in progress black box kind of stuff as I don’t think even the model makers know the magic words to say to the model to get it to work how someone wants.
Glad to hear your thoughts on this.
Thanks