Hello everyone! Hope you are well. I’ve been working away on updates to the game. I’m very close to a 2.0 release, which should be everything I initially had in mind for the game. This next release should be the base game. I have plans for further expansion, but this next release will see my initial idea to completion. There has been a lot of balancing game play as well. I’ve learned a lot by doing simulations to determine pacing. I started experimenting with simulating gameplay outside the player, which pulled me in another direction for a bit with...

Attending another agent hack in SF. I was in the top 5 again out of ~80ish teams, but only the top 3 received prizes. My idea for the hack came from my experimenting with simulations in Dunil’s Hold and from reading some research papers. Google DeepMind did a paper on Concordia. There is a paper about Felt: A Simple Story Sifter. Neighborly, Talk of the Town, were other inspirations.

Those works mimic what I had stumbled across on my own when building Dunil’s Hold. One of the, if not the, biggest issue with using an LLM within a game to create content is hallucinations. It works really well until it doesn’t, and then it is often immersion-shattering. This bad content may linger in memory depending on how the LLM/system is orchestrated, causing further degradation to the story.

In the hack, I started with a concept I had been playing with and added some of the products the vendors were offering. I used Neo4j for their graph DB, which I really liked. Tavily for web search, I would inject news into the simulation I used for this hack. Fastino was offering a first use of their new Pioneer API and site. It lets you train a sLM on their infrastructure to your specific use case, and then do web inference on that model.

It went well. I had tuned this to create content in a certain tone, and one of the judges asked if it could be changed to a different tone, to which I replied yes. It could, but it made me think about it deeper afterwards, and what I had built, differently.

I’ve since built a new prototype to be more agnostic and ways to extend the simulation. I’m not completely sure what I am going to do with this. It could be a game, interact with the simulation, or it could be a terrarium, nudge it in certain ways.

Still job hunting. I’ve hit submit on countless. I really would like to get back to work with a team. Especially with someone working in adjacent spaces to where I am currently. I believe an org can really get things done now in tech with the right leadership.

I've updated the links page. Added more reading resources.