JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Implementation report

Building an open AI tool that chooses answers instead of typing

The author built a free version of Jev, an AI tool that chooses from options rather than writing an answer. This guide explains how the project works and how it was used to play a video game.

Original by Stephen BlumIntegrations

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits

“We Rebuilt Jev's API on an Open Model and Used It to Play Doom” by Stephen Blum. Read the original source.

This expanded guide is an AI-narrated adaptation prepared by JevMade. It expands the source’s essential ideas, examples and caveats in JevMade’s own words and is not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

Our summary

The author built an open version of Jev, an AI tool that chooses from options rather than writing an answer. People use this design to help AI assistants make quick decisions, like sorting incoming messages or picking a direction in a video game, without waiting for the software to type.

Instead of generating words, the software reads the instructions once and calculates the mathematical probability for specific letters, like A, B, or C. It speeds up this process by saving the repeated parts of the instructions in memory, so it only has to read the new information each time.

This project is useful for developers who want to test fast AI decision-making on their own computers. The author only copied the original tool's interface, not its special training. In these tests, the video game demo still ran slower than real time on a single computer.

Key takeaways

  1. The AI tool chooses an allowed letter based on mathematical probability instead of writing out a full response.
  2. Standard software code handles the game map, while the AI only chooses the next local action.
  3. Saving repeated instructions in memory cut the time it took to make a single choice in these tests.

This project only copies the original tool's interface, not its training methods or performance claims. The results come from a small set of authored tests and one computer, and the game demo ran slower than real time.

Blocks.ai blog · Source reviewed

Read the original guide Opens the author’s site in a new tab.