Welcome
Welcome to the AI Generated Lore Project! Here, we made it our goal to focus on
working with ai to generate new information, mainly items, that could function as
actual items in the Dark Souls universe. This would require them to contain
lore-accurate information that aligns with how the world of Dark Souls works. On
this site, you'll find different pages that explain how our pilgrimmage turned out,
and what we found. There will be visualizations, documentation, discussions,
problems we faced, and anything else that can accurately explain the process of how
we (somewhat) trained an ai model to produce information based on the list of items
found in the first Dark Souls video game. Take a look around!
We have a variety of pages showcasing what we did! Our Documentation and Examples
pages show what we got and how we got there, while the Visualization page helps
explain how the information is being used! Lastly, our Problems page gives a lok
into some of the many challenges that we encountered, and briefly explains how we
overcame them.
Research Questions
- Can Chat GPT generate items that fit into the Dark Souls lore?
- Can we fine-tune an ai model with only information from the first Dark Souls
game and get it to output new items?
- Can we see how ai takes information and uses it to generate a response based on
a prompt?
These are just a few questions that we were asking as we began working with the
OpenAI system. We wanted to take a look at how ai works within generating new
information, and potentially using it in a new and unique way. The project initially
focused on the use of ChatGPT and how we could give it our information in a
copy-paste format, and it essentially worked. However, bigger ideas came into play
and the thought of training a model using python became a central point in achieving
what we want. Over time, this changed how we looked at ai, and raised more questions
about how it functions.
Conclusions
Throughout our experiences working with OpenAI and the AI models, we learned a lot.
The potential of artificial intelligence is great, and using it effectively is a
challenging task. During the process, we learned a lot about Python and how we can
use it as a tool. We went more in-depth into different branches of coding and text
analysis in general by making use of programs like spaCy and PANDAS, along with
becoming more familiar with different file types such as JSON files, SVG, and more.
The use of all of these tools allowed us to reflect on how we interacted with the ai
model and learned about ai as a whole. We mainly came to the conclusion that ai are
very particular and that it takes a lot to train one and make it useful. We were
able to produce an ai model based off of an existing model, and that proved to be
difficult. However, we were able to speculate and analyze how ai models work, and
how they think; that is that they operate off of hypotheticals and a plethora of
examples. We also discovered how the ai uses fine-tuned JSON files to look at the
examples and base its thought process off of the provided information. From our
experiments, we found that this works. Essentially, we've been able to take a dive
into the brain of an ai and understand how similar words and tokens were able to be
used as associations that can be made into viable results in the output of a prompt.
We still get to use the playground after we worked on everything, so we believe we really made some
progress in understanding how to use ai.
