AI_chats.md

• GPT-4o 22.08.2024: Lower bound for state space size
  After several messages, code was partially copied with some modifications to make it more efficient, calculating fewer factorials.

• GPT-4o 28.08.2024: brainstorming to fix 180° rotation animations
  Main ideas were my own, ChatGPT added suggestions for algorithms to use in an implementation and explained pro's and con's of the options.

• GPT-4o 29.08.2024: implementing ideas for point- and piece orbit-calculations
  After a few messages, code was almost directly copied to tests/puzzle_piece_orbits.py

• Gemini (Gemini 1.5 Flash) ~17.08.2024 Implementing early stopping with StableBaselines 3
  This chat helped me learn about early stopping, which I simultaneously read up on in the StableBaselines documentation, then used a combination of what I learned from both sources in tests/nn_rl_stable_baselines/nn_rl_training.py. Code from this chat was not directly copied, but due to the simplicity of the problem is still similar.

• GPT-4o 16.09.2024 parallel RL training with sb3 on GPU unused, read sb3 documentation instead for more reliable information. Provided some inspiration for other RL methods to try if necessary.

• GPT-4o 19.09.2024 fix incorrect inverses in definition of Geaer Cube Ultimate
  Saved some simple but tedious work to correct an error resulting from a minor bug. Doing this without an LLM would have taken a few minutes.

• GPT-4o 27.09.2024 Write simple data analysis code for test files
  Saved time for coding some plots and basic evaluations. I have coded very similar graphs many times before, so ChatGPT was used purely to save time. I uploaded my test data file for one puzzle and described what graphs I wanted. After several messages and some manual adjustments for style of code and plots, the code was mostly copied to src/au_modules/poilcy_analysis.py.

• GPT-4o 15.10.2024 vectorize binary multi-goal reward
  Very simple modification to existing code. The output was partially used in src/ai_modules/nn_rl_reward_factories.py (added [:, None] to state for vectorization).

• GPT-4o 18.10.2024 Implement agent strategy visualization
  Helped to implement a visualization I came up with. Output code was not working. I went through it, changing anything I wouldn't have done the same way and fixed the core logic of the implementation. The method used to store intermediate values was changed by me since the AI generated solution deviated from my original intent, eventhough both were correct approaches.

• GPT-4o 26.10.2024 Simple GUI vor visualizing color mappings for rotation naming
  To verify that the implemented function to map colors (rgb tuples) to color names (strings) works, I wanted a GUI where I could visualize colors and where they are mapped. Once again, this was a simple code snippet that I have coded in a similar way many times before and it was faster to let an AI do it. I modified the output code to better reflect my vision. For example I removed borders, changed the background color, added logic to automatically split the output into multiple columns etc..

• GPT-4o 02.11.2024 advanced move sequence parsing (unused)
  Completely unused chat. After briefly looking at the response, I decided to implement the feature myself. I never looked at the exact outputs.

• GPT-4o 03.11.2024 Covnerting two old files to new JSON format I converted my code to output and expect algorithm moves files in json instead of txt format. To translate old files quickly, I used ChatGPT. The outputs were directly copied since they worked immediately. This saved a few minutes of tedious work.

• GPT-4o 18.11.2024 Implement plotting of TensorBoard data I used this code to load the tensorboard log files to then plot them in a customized way with matplotlib. ChatGPT helped to implement this slightly faster than I would have been able to without it. Generating the code required involved several iterations and some external sources. For the refinements of the initial solution, I only copied very small fragments from the new chat outputs.