The 2nd ChatGPT4PCG Competition

Character-like Level Generation for Science Birds

Prompt Engineering Examples

We provide a few example programs implemented using various prompt engineering techniques. These examples are offered to assist you in getting started with the competition. You can utilize these examples as a foundation to develop your own advanced prompt engineering techniques. The examples are accessible in a GitHub repository, which you can find here.

Tools

ChatGPT

  1. Open your preferred web browser (such as Chrome, Firefox, or Safari).
  2. Navigate to the OpenAI website at https://chat.openai.com/chat
  3. Start interact with the model.

Science Birds Evaluator

This is a modified version of Science Birds that supports blocks used in the competition. It provides a feature that evaluates the stability of a particular level and generates an image used for the similarity checking script.

You can download this version of Science Birds from https://github.com/chatgpt4pcg/modified-science-birds.

Installation Instruction

  1. Download Unity Hub
  2. Open a project in Unity Hub
  3. Download the required version of Unity (if needed)
  4. Open the project in Unity

For further installation steps, please refer to the Science Birds repository.

Online Converter

The converter is a tool that creates an XML file and an image from a series of drop_block() function string. The resulting XML file provides a structured representation of the functions while an image shows the visual representation of the functions. It is recommended to use the converter on a desktop device to ensure optimal usage and accurate generation of output files.

  1. Visit our online converter at https://chatgpt4pcg.github.io/tools/converter
  2. Paste the text into the first input box, ensuring that it is in the correct format
  3. Click the "Convert" button.
  4. You can either copy or download the XML result and test it in Science Birds. To download the image, click the "Download image" button

Note that the XML file must be placed in the Assets/StreamingAssets/Levels folder.

Please note that the image generated by this online tool is intended for illustrative purposes only and will not be used for evaluation. For the actual evaluation tool, we recommend checking out the modified version of Science Birds.

Online Code Extractor

The online code extractor extracts a code block from ChatGPT API responses, demonstrating the same algorithm used in the code extraction script.

Visit our online code extractor at https://chatgpt4pcg.github.io/tools/code.

Online Letter Classifier

Please note that the Online Letter Classifier embed in the tool page is also available at Hugging Face Spaces and only shows the top five predicted classes with associated percentage. For the full version of the model, please use the similarity checking script. This model was a fine-tuned version of Vision Transformer (ViT) trained on this dataset. More information on the dataset is also available on the same page.

Visit our online Letter Classifier at https://chatgpt4pcg.github.io/tools/classifier.

Offline Scripts and Packages

Code extraction script - https://github.com/chatgpt4pcg/code-extraction-script

Text-to-XML conversion script - https://github.com/chatgpt4pcg/text-to-xml-converter-script

Similarity checking script - https://github.com/chatgpt4pcg/similarity-checking-script

Diversity checking script - https://github.com/chatgpt4pcg/diversity-checking-script

Scoring and ranking script - https://github.com/chatgpt4pcg/scoring-and-ranking-script

chatgpt4pcg npm package - https://github.com/chatgpt4pcg/chatgpt4pcg

chatgpt4pcg Python package - https://github.com/chatgpt4pcg/chatgpt4pcg-python