Skip to content

2048 × Jev

Instrumented 2048 web lab where every move is a Jev Choice over four directions, with probabilities, latency and cost shown live.

ProjectbetaMIT

Overview

2048 x Jev hands every move of the game 2048 to Jev and shows its probability distribution, confidence, latency and token cost in the UI. The author says the goal is not a fun game but making it visible and measurable how context design affects a decision model.

Each turn is a single Choice over the four directions with no heuristic fallback, so every move is attributable to Jev. A user-set confidence threshold can pause play for human review. Prompts and board rules are editable, backends are bring-your-own-key, and game archives can be imported and exported. It ships a free-trial backend on Vercel and a static GitHub Pages build that calls it.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich direction should the board move?up · down · left · right

How it works

  1. Serialize the board and rules into state
  2. Ask Jev one Choice over four directions
  3. If confidence is below the user threshold, pause for review
  4. Otherwise apply the move and log probabilities, latency and cost

Strengths

  • Every move attributable to the model (no fallback)
  • Live view of probabilities and cost for prompt experiments
  • Free-trial backend and bring-your-own-key options
  • 76 unit tests run in the start script

Caveats

  • Key sealing is explicitly not cryptographic protection
  • Docs primarily in Chinese with an English README

Getting started

git clone https://github.com/ARCJ137442/jev-2048.git && cd jev-2048 && npm install && ./start.sh

At a glance

Type
Project
Status
beta
Language
TypeScript
Runtime
Node server + browser
Has code
yes
Has tests
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Gaming: a 2048 board where every move is a Jev `Choice` over four directions with no heuristic fallback, gated by a user-set confidence threshold that pauses play for human review, with editable prompts and board rules, bring-your-own-key backends, and archive import/export.

    yibie/awesome-jev
  • This project integrates Jev to provide structured decisions for its workflow. See the repository for implementation details.

    logicrw/awesome-jev-projects
  • An instrumented 2048 web lab where every move is a Jev (TypeSafe AI System One) Choice, with no heuristic fallback | 用 Jev 决策模型驱动每一步的 2048 网页实验台,概率、置信度、延迟与成本全部摊开可见,且刻意不做启发式兜底.

    hellogumbo/awesome-jev

Frequently asked questions

Is there a fallback if Jev is unsure?

No heuristic fallback; a confidence threshold can pause play for a human instead.

Do I need a key to try it?

No, the hosted deployments include a limited free trial; local runs can use your own key.

What is it for?

Studying how context and prompt design change a decision model's choices.

Related projects

  • Bilingual turtle-soup (lateral-thinking) puzzle site where Jev acts as host, judging player questions and grading solutions.

    AppbetaTypeScript
  • Browser Minesweeper demo where Jev answers 'is this cell a mine?' for every frontier cell in one request, with an exact solver as fallback.

    ProjectprototypeJavaScript
  • Pandu: tiny local policy networks (2.9K-5K parameters) with calibrated uncertainty that fall back to a teacher when unsure.

    ProjectresearchPython
  • Pokémon Red on PyBoy where code owns route and arithmetic, Jev picks only at branches, and battle faint predictions are Brier-scored.

    DemoprototypePython
  • Twenty-questions game where Jev's confidence decides when to stop: it commits, hedges or refuses to guess.

    DemoprototypeTypeScript
  • Harness where Jev or Laya pick DOOM tactics and targets from legal options while a deterministic controller presses buttons.

    ModelresearchPython