Sasha Rush commited on
Commit
aadb98e
1 Parent(s): dffbd1c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +9 -2
README.md CHANGED
@@ -1,14 +1,21 @@
1
  # GPTWorld - Prompt Golf
2
 
 
3
  <a href="https://colab.research.google.com/github/srush/GPTWorld-Challenge/blob/main/GPT4_game.ipynb" target="_parent"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab\"/></a>
 
4
 
5
  GPTWorld is an educational environment to learn about Prompting.
6
  It consists of a grid-world environment to test the ability of language models to follow instructions in a grounded environment.
 
7
 
 
8
  ![download (1)](https://user-images.githubusercontent.com/35882/234447369-6a4ca94d-5bb8-4c8e-a34d-a1ff0614bf7d.gif)
 
9
 
10
- The goal of the puzzle is to construct a prompt that can get GPT-4 to solve a complex game.
11
 
 
12
 
 
13
 
14
- <img width=500px src="https://user-images.githubusercontent.com/35882/234614176-789fd6a8-bc7c-4f42-889b-81b3879cca36.png"/>
 
1
  # GPTWorld - Prompt Golf
2
 
3
+ <p align="center">
4
  <a href="https://colab.research.google.com/github/srush/GPTWorld-Challenge/blob/main/GPT4_game.ipynb" target="_parent"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab\"/></a>
5
+ </p>
6
 
7
  GPTWorld is an educational environment to learn about Prompting.
8
  It consists of a grid-world environment to test the ability of language models to follow instructions in a grounded environment.
9
+ You instruct the model to generate code to play the game.
10
 
11
+ <p align="center">
12
  ![download (1)](https://user-images.githubusercontent.com/35882/234447369-6a4ca94d-5bb8-4c8e-a34d-a1ff0614bf7d.gif)
13
+ </p>
14
 
15
+ The goal of the puzzle is to construct a prompt that can get GPT-4 to solve a complex game. You can watch in real-time as GPT plays the game.
16
 
17
+ <img width=500px src="https://user-images.githubusercontent.com/35882/234614176-789fd6a8-bc7c-4f42-889b-81b3879cca36.png"/>
18
 
19
+ ## Leaderboard
20
 
21
+ I was able to solve the big puzzle using roughly 2000 tokens (input/output). Can you do better?