• Together with the group you worked with, finish the negotiating game sketch you started in class, turning it into an polished interactive experience. (I thought the way Jiapei embedded UI elements and graphics was quite successful. You could also consider using sound or simple animation.)
• Document some of the most funny, deep, meaningful or surprising dialog snippets that you encountered playing this game.
• Watch the trailer for Papers, Please, an award-winning indie puzzle game by Lucas Pope (no LLMs or AI involved!) some people called a “dystopian political sim”. (We didn’t get to watch this in class - wanted to share such scenario as example for where LLMs could be considered in more darker, critical contexts.)
• Come up with 3 questions about anything relating to week 6’s content?
• Assemble a dataset that you want to use to fine-tune an LLM with. This could take one of the following forms:
  • (Most straight-forward) A list of URLs of YouTube Videos. (Next week I’ll show how to download the audio from YouTube and use an AI model to transcribe each of the videos to extract the text.)
  • (Medium straight-forward) Put together a text file (no formatting) of text to train from. Should be at least 3000+ lines of text.
  • (Least straight-forward) Put together a list of question-response pairs you want to train from. (e.g. “What is your favorite food?” ⇒ “Japanese”) Best in JSON-format - e.g. [{”prompt”: “…”, “completion”: “…”}, …]. Should be at least 3000 pairs.
• Share some notes about what could make this dataset interesting or meaningful. What do you hope the fine-tuned model will be able to produce with it?
• (If anyone does any experiment with the Agent sketches, post those too - but this doesn’t count for the grade ;)