Small decisions. Interesting possibilities.Submit contentSubmit
JevGPT preview

JevGPT

An experiment that makes a non-generative decision model assemble a chat reply one word at a time.

Added to Jevfast

How it uses Jev

Jev receives the conversation transcript as state. At each step, bucketed Choices propose vocabulary words, a final Choice runs off the bucket finalists, and a Noul estimates whether the reply is complete. Local code applies repetition penalties and decodes the selected word until completion is allowed or the word limit is reached.

What Jev decides

Reply completion checkExample answers · not a recorded Jev response · Source ↗
Question 1 · complete
YOUR APP
INSTRUCTION

Is the reply complete?

STATE

Conversation transcript, latest user message, and the partial reply generated so far.

JEV · NOUL
YesNo

App workflow

  1. Build transcript state

    The client combines conversation history, the latest user message, and the partial reply.

  2. Propose vocabulary candidates

    Code shuffles the vocabulary into bounded Choice buckets and asks in parallel, alongside a completion Noul.

  3. Run off finalists

    The two leading candidates from each bucket compete in a final Choice.

  4. Decode and continue

    The client adjusts the returned distribution with repetition penalties, appends a word, and repeats until the stop rule or word limit.

Why it is interesting

It tests the boundary between typed decisions and text generation: a chain of word classifications can produce readable fragments, while bucket comparisons, stop conditions, and repetition control expose the costs and failure modes.