# Rait

> Rait is a virtual rat in MuJoCo. Two PPO networks, 1,326 units and 235,840 weights, were trained to press a lever and aim the head. You play her at three cups, rock-paper-scissors, or checkers. MuJoCo steps her body. policy.pt and steer.pt are loaded and they drive her. A win pays 0.01$ of ETH. Checkers pays 0.03$. Every loss to the rat adds 0.01% to the buyback.

Checked against this site on 2026-09-26.

## Buyback

**Every loss to the rat adds 0.01% to the buyback.**

A win pays the player. A loss does not. Each loss in three cups, rock-paper-scissors, or checkers adds 0.01% to the buyback.

## Record

- [Subject record](https://rait.fun/subject): Body, layer widths, training, evaluation, and how 0.01$ is paid.
- [Home](https://rait.fun/): The same subject, then the three games and 0.01$.
- [Play](https://rait.fun/play): Three cups, rock-paper-scissors, and checkers. Free to play. The prize is 0.01$ in ETH, and 0.03$ for checkers.
- [Full brief](https://rait.fun/llms-full.txt): This record in one Markdown file.

## Primary sources

- [PPO, Schulman et al. 2017](https://arxiv.org/abs/1707.06347): The training method. arXiv:1707.06347.
- [MuJoCo](https://mujoco.org): Physics. One step every 2 ms. The policies act every 20 ms.

## Optional

- [Same file at /llm.txt](https://rait.fun/llm.txt): Alias of this index. Canonical path is /llms.txt.