Snake Bug-Fixing with Jev

Sonam R reports an experiment using Jev with Gemma-4B to fix bugs in a Snake game. The setup separated executive-function decisions from code generation.

View on X

Sonam R@aksssrao𝕏

We ran an experiment fixing bugs in a Snake game using Gemma-4B paired with Jev. The setup was built to isolate executive function from code generation: vanilla Gemma had to decide every action and write the patch, while Jev handled routing through a four-part evaluation checklist (next action, evidence threshold, loop detection, completion) and only invoked Gemma when an edit was required. Across

Sep 17, 2026X postsView on X

Vanilla Gemma had to choose every action and write patches. Jev routed actions through a four-part checklist—next action, evidence threshold, loop detection, and completion—and invoked Gemma only when an edit was required.

Also filed under Games & real time

Steer a 3D Box with Goals
The build steers a 3D box with plain-language goals. An LLM plans each instruction, while TypeSafe Jev makes move, turn, and jump decisions in real time.
Benchmarks Jev on Pokémon Red
The project benchmarks Jev, TypeSafe’s fast decision model, against Jev paired with a GPT-6 Sol planner for playing Pokémon Red.
Jev-Powered Rust Survival Game
Anthony Eckert says he has a Rust survival-game clone in pre-alpha, with an agent playing using Jev. He says crypto rewards are coming soon, with browser play and full agent support planned for tomorrow.
A Jev-Made Unity Build
Kevin Kern says the build was made with Jev, GPT Live, Blender, Unity, and more.