I’ve been using ai studio for quite a while now, and long conversations are what is great with the site.
Yet sometimes I run out of tokens, to fix this I would download the conversation file from ai studio folder in Google drive and convert it to a text file. I would post this to a new chat with same instructions and settings and continue without much hassle. (It would only amount to a maximum of a 20-30 thousand tokens even on a 1mil chat before when I removed images and files)
It would save like this as long lines for each block.
Here is the link for the file having both earlier save format(initial) and new save format(later).
What this does is that these files are almost the same token length as the original chats which is not helpful. And another thing I noticed when I tried to only get the chat from another seperate chat, when a chat gets too wordy or filled with long text, the size actually increases even more(120k chat came to 510k tokens from file) this is so problematic, as I often use multiple bots as a system that requires each other’s context.
Is this change intentional and is there no fix for my use case??
I just reset the conversation as the AI gets too confused if there is a long conversion that it has to analyze. By resetting it, it has a clean slate again. I tell the AI to put important details in a file “instructions.txt”, which it will read and maintain during the conversations, allowing me to edit things if need be.
This tends to give more reliable results.
Ok so I was actually asking about transfer of full chat for context to another bot. That was how I tried to use it.
How do you get such txt files from ai studio?? I only know code block can be downloaded.
And where is that instructions file put into the chat?? Is it put at the start, but later it forgets this and uses only the current context for that.
I even have a hard time making it follow the custom instructions first before the chat context…
Such text files are generated by the AI as you just tell the AI to generate this text file as part of the project. I generally tell the AI to generate a ‘public’ folder and an ‘instructions.txt’ file inside that folder. The it creates it and you can later edit it or tell the AI to modify it.
This ‘public’ folder is useful when you deploy the App to your own web server, as you need to call ‘npm’ to deploy it to your server. And npm generates a /dist folder inside the project and copies all files from public to dist. Nginx would be set with the dist folder as root, so all files in the public folder also becomes available on the website as public data. On my own site, that would be example (using example as value as the moderators don’t like posting of real sites).
In my conversions, I just have to refer to this instructions file for commands that I provided in the past that matter, allowing me to clean up the chat before it gets too long.
You can download your entire chat as MarkDown-file with a single click with my userscript. See the green custom download button at the right in the toolbar:
By resetting it, it has a clean slate again. I tell the AI to put important details in a file “instructions.txt”, which it will read and maintain during the conversations, allowing me to edit things if need be.