What it does
The intentionally comic experiment is slow and costly because every new symbol requires another model call.
Playable demos & bots
jevchat
Make Jev “chat” by sampling its probability distribution over the next character or token, one request at a time.
Bookmark: jevchat
Keep this in your collection.
Leave a noteWhat would you try with this?
: jevchat
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
A developer can adapt this Python project to build an experimental chat tool. The code asks the TypeSafe service to guess the next letter of a reply. It connects to TypeSafe using an access key. You could borrow this approach to explore how an AI builds sentences.
The software asks the outside AI service for one letter at a time. This requires a separate request for every single character in a message. Because of this, generating a full reply is very slow. The creator notes the cost is highly impractical and the results are meant for fun.