What it does
The author presents it as an illuminating bad idea: one proper Choice answers better and cheaper. The experiment maps where repeated classification stops behaving like language generation.
Apps & data pipelines
let-jev-speak
Coaxes sentences from a classifier by choosing a domain, then asking Jev to pick each next word from bounded vocabularies.
Bookmark: let-jev-speak
Keep this in your collection.
Leave a noteWhat would you try with this?
: let-jev-speak
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
To try this experiment, start by collecting a few simple questions you want to test. A developer can connect the tool using an access key from TypeSafe. They can run it on a computer or connect it to a chat bot.
The software matches your question to a built-in topic pack, then picks each answer word one by one. This is an entertaining experiment rather than a practical writing tool. Because it asks for every single word separately, generating a reply takes many steps and costs more.