One Pager 📄

The Idea 💡

I miss Arthur Morgan. I want to be able to interact with Arthur from the afterlife and ask questions in the hopes that I will receive contextual answers to questions like I was asking Arthur himself.

Origin 🌱

I enjoy working towards achieving platinum trophies on the PlayStation in my, limited, free time. Being a dad, husband and full time Engineering Manager means I am often occupied with the aforementioned, leaving me with about a golden hour of gaming a day, at most. At the time of writing I have accrued 11 Platinum trophies, some of which where a grind or took skill (in my humble opinion) to achieve. Of those, here are some of the platinum trophies I am most proud of achieving;

• The Witcher 3
• Sekiro
• Bloodborne
• Elden Ring
• Demon’s Souls
• Tony Hawks Pro Skater 1 + 2

The satisfaction and the sense of achievement and completion I get when I eventually do achieve the platinum trophy is what keeps me coming back. I am always looking for the next platinum trophy that I can aim for.

Read Dead Redemption 2 🤠

After completing the Read Dead Redemption 2 story in 2023 and considering myself “done” with this game and I stopped playing. I have seen countless videos, read numerous articles and looked at the trophy list to determine that achieving the platinum trophy in RDR2 is no easy feat. At the start of the year in 2025, I decided to jump back on the horse. Literally. As I had already completed the main story I continued from my save in the epilogue and started hunting as John Marston. I was having fun, but something was niggling in the back of my mind. I missed playing as Arthur Morgan.

LLM Research 🔬

As an Engineering Manager working in a Software Engineering department I have an obligation to research new technologies and keep up to date with the latest trends. The hot topic at the moment, of course, is AI and LLM’s.

I naively set out to understand how LLM’s work in the hopes that I would be able to create my own, simple LLM. Who was I kidding? This was a mammoth task and is something I hope to achieve, one day, but lets park that goal, for now.

During the research phase of wanting to create an LLM from scratch I came across the idea of retraining an existing model. This sounded much more achievable for my smoother brain so I decided to investigate this. After looking into this in detail, I broke it down into these steps;

• Step 1: Find a subject that an existing LLM has limited knowledge about
• Step 2: Curate training data to “fill these gaps”
• Step 3: Train an existing model with this new data
• Step 4: Produce the new model
• Step 5: Use the new model

With the above checklist in place, I just needed to figure out what I wanted to retrain the model on and then I could get started.

Merging the Two Ideas 💡💡

I knew I wanted to retrain a model so I could understand what the process involved and I knew I missed playing as Arthur Morgan. What if I could merge the two and output a retrained model that would let me interact with a version of Arthur Morgan with all of the context of the game and the premise that he would be speaking from the afterlife. The question is, does the LLM already know about Arthur Morgan and the context of the game? I decided to investigate this further. Lets just ask!

💬 Question: do you have the context of all the story dialog and the journal entries from Arthurs perspective?

🤖 Answer: I don’t have complete verbatim transcripts of every dialogue exchange or a perfect reproduction of all journal entries from the game. My knowledge about Red Dead Redemption 2 comes from my general training data rather than from having the full game script or journal text at my disposal. I can provide summaries of key dialogues, describe the general content of Arthur’s journal entries at different points in the story, and discuss the major themes and character developments reflected in both. However, I wouldn’t be able to quote entire conversations word-for-word or reproduce every journal entry exactly as it appears in the game. Arthur’s journal is particularly significant as it provides insight into his inner thoughts and reflections that aren’t always expressed in dialogue. It contains his observations about the changing world, sketches of landscapes and people he encounters, and his evolving perspective on Dutch, the gang, and his own life choices. If you’re interested in specific conversations or journal entries from particular points in the story, I can share what I know about them in general terms. Would that be helpful?

TLDR; I was researching LLM’s and wanted to attempt to retrain a model. I was also playing the RDR2 epilogue, in my down time, and I missed playing as Arthur. I decided to merge these two streams into one and attempted to retrain a model with contextual data from the game. I then instructed the model to act as Arthur from afterlife.