We are proud to announce that the third recipient of the Goblin Builder Initiative is
@RealRoseGoblin.
Rose recently reached out and offered to build something we've wanted for a long time: a true GOBLIN LLM.
What stood out wasn't just the idea itself, but the fact that she offered to do it completely for free, without asking for anything in return.
Building and maintaining a model isn't free. Training, inference, infrastructure, experimentation, all of it costs time and money. Yet Rose was willing to cover those expenses herself simply because she believes in the goblin phenomenon and wanted to contribute to the movement.
The craziest part?
She's not even a holder.
There was no personal financial incentive. No expectation of compensation. Just a genuine passion for goblins and a desire to help build something interesting.
And this isn't going to be another wrapper with a goblin-themed system prompt slapped on top.
The Goblin LLM is being built from the model level upward. Rose starts with raw base models before behavioral alignment takes place, then fine-tunes them on specialized training data and locks those behaviors directly into the model's weights.
The goblin isn't being instructed to act like a goblin. The goblin becomes the model.
That approach creates something fundamentally different from traditional agents or prompt-based personalities, which are ultimately still operating within the boundaries of behaviors defined by someone else's training process.
We're incredibly excited to continue working alongside Rose as we build and refine the Goblin LLM together.
To support her efforts, we've covered all of her inference costs for the next two months in advance, allowing her to focus on experimenting, training, and pushing the project forward.
The future needs more goblins, and the goblin community is proud to help keep this one running. 🧌