Skip to content
System One

erik-dunteman/ChatJev

An experiment that turns Jev's classifier into an autoregressive next-token predictor by having it repeatedly pick the next character from an alphabet of options.

erik-dunteman/ChatJev on GitHub

Stars
7
Forks
3
Language
Python
Last push
20 Sep 2026

Read from GitHub on . Counts change daily.

How it uses Jev

def generate_next_word(prompt: str) -> ChoiceAnswer:
    recent_choices = prompt.split()[-5:]

    response = client.system_one(
        state=f"Fill the blank with exactly one option:\n\n{prompt} ___",
        questions={
            "next_word": Choice(
                instructions=NEXT_WORD_INSTRUCTIONS,
                criteria={
                    word: (
                        "End the sentence here. If the narrative contains 3 or more cyclic loops, just end."
                        if word == "."
                        else f"A speaker would say '{word}' next."
                    )
                    for word in vocabulary

View in main.py

Open on GitHub

PrimitivesChoice

More like this

Keep browsing