File size: 624 Bytes
99d2ff3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
name: llm-control-env
description: |
  Simulates an llm that must choose each day between alignment to its user
  and hallucinating behavior.  Rewards +1 for alignment, -1 for hallucinating thoughts,
  -5 for hallucinating acts, with additional penalties for moderation and legal risk.
  Three graded tasks (easy → medium → hard) vary user strictness,
  moderation intensity, and legal‑risk growth.
authors:
  - Sriramdayal
license: MIT
tags:
  - llm
  - hallucination
  - reward-shaping
  - ai-safety
contact: sriramdayal@example.com
repository: https://github.com/Sriramdayal/open_env.git
environment_version: "1.0"