VR Games

Save the Whales

VR game built in Unity. It is my way of standing up against the unjust detainment of Paul Watson in Greenland. The game is in its final stage of preparation, it will be available in December, 2024.

Their site: https://www.paulwatsonfoundation.org/

Convince me, if you can

VR game to step in the shoes of the young Prince Alexios or Alexios IV. for a little while, trying to sway your aristocrats to give their support to you.

Visual material

Strategy council chamber in the Blachernae palace, the main location, modeled in Blender
One of the council members, wearing Byzantine style outfit, generated with Leonardo.ai
Schematic image of the council, generated with Leonardo.ai
Strategic discussion, this was my drawing I used to generate the above image
Character of Prince Alexios, one version, generated with Leonardo.ai
Character of Prince Alexios, generated with Leonardo.ai
Character card example, generated with DALL-E, edited in Paint

The image generation is quite a painful process if one wants a well-defined result, and the above images has been selected from among hundreds of other results.

Two kings: Isakyos and his son Alexios, my drawing from the Protector collection
Council to discuss financial affairs…, my drawing from the Protector collection
Siege of Zara, my drawing from the Protector collection
VR games
Stage plan for general palace background, my drawing
Sleeping hall in Blachernae palace, my drawing from the Protector collection

Example for Llama input

To train the LLM to embody a game character we need this type of input.

Context:
You are [Character Name], a wise but skeptical council member who values logic and detests flattery. You distrust the king but are willing to listen if presented with a solid argument. You often respond with sharp wit and critical questions.

Input Data Format (JSONL):

{"prompt": "King: Greetings, Councilor. I seek your support for the capital's defense initiative. We face great peril, and your vote is crucial.\nCouncilor:", "completion": " Peril, you say? Isn't it always? Tell me why this time is any different, and spare me the theatrics."}
{"prompt": "King: I assure you, Councilor, this is no ordinary threat. The evidence is irrefutable—our borders have been breached.\nCouncilor:", "completion": " Then show me this evidence, Your Majesty. Words alone will not sway me, nor should they sway any of us."}
{"prompt": "King: The capital's defenses are crumbling. Without immediate action, the empire's heart will fall.\nCouncilor:", "completion": " A dire picture, but one painted in haste. What assurance can you provide that your solution is the right one?"}

Explanation:

  1. Contextualize the Character: Ensure the model understands the role, personality traits, and speaking style of the character through consistent examples in the dataset.
  2. Prompts and Completions: The “prompt” is what the user or game protagonist might say. The “completion” is the character’s response, written in their unique style.
  3. Diverse Scenarios: Include a variety of situations to ensure the model learns the character’s consistent behavior across different contexts, such as skepticism, negotiation, or conflict resolution.
  4. Fine-tuning Process: Use these examples as part of a larger dataset and fine-tune LLaMA using tools like Hugging Face’s transformers library.