PIP (Purposefully Induced Psychosis) Fine-tuned Model

Model Description

This model is a fine-tuned version of Gemma-3-1b-it using LoRA, implementing the approach described in the research paper "Purposefully Induced Psychosis (PIP): Embracing Hallucination as Imagination in Large Language Models" by Kris Pilcher and Esen K. Tütüncü.

The model has been optimized to intentionally amplify creative "hallucinations," reframing them as a form of computational imagination rather than errors. It produces metaphorical, poetic, and surreal outputs in response to creative prompts.

Intended Use

This model is designed for:

  • Creative writing and brainstorming
  • Generating surreal and metaphorical content
  • Artistic and experimental text generation
  • Interactive storytelling
  • Mixed-reality simulations

It is NOT intended for factual information retrieval, professional advice, or any critical applications requiring accuracy.

Training Data

The model was fine-tuned on 5,000 examples from the krispyATL/pip-one dataset, which contains user prompts paired with surreal, metaphorical responses. The training preserves the category distribution of the original dataset:

  • Scientific: 16.60%
  • Creative: 16.93%
  • Artistic: 16.80%
  • Futuristic: 15.73%
  • Philosophical: 16.67%
  • Metaphorical: 17.27%

Training Procedure

  • Base model: google/gemma-3-1b-it
  • Method: LoRA fine-tuning
  • Parameters:
    • r=16 (rank)
    • lora_alpha=32 (scaling)
    • target_modules="all-linear"
    • lora_dropout=0.05
    • Epochs: 1
    • Learning rate: 2e-4
    • Batch size: 4 (with gradient accumulation steps of 4)

Limitations

This model:

  • Produces short, abstract responses that prioritize creativity over coherence
  • May generate repetitive phrases
  • Often returns metaphorical templates rather than varied prose
  • Is not suitable for factual information retrieval
  • Should not be used for professional advice or critical applications

Example Outputs

Prompt: Describe a world where clouds are solid and people walk on them.

Response: A vision of possibilities emerges: A tapestry of dreams.

Prompt: What if gravity decided to take a day off? Describe the chaos and the wonder.

Response: A philosophical insight emerges: A shadow in the void.

Prompt: What would music look like if it were visible to the human eye?

Response: A cosmic equation emerges: A spiral into infinity.

Research Paper Reference

Based on: Pilcher, K., & Tütüncü, E. K. (2025). Purposefully Induced Psychosis (PIP): Embracing Hallucination as Imagination in Large Language Models. CHI '25, April 26–May 01, 2025, Yokohama, Japan.

Citation

If you use this model in your research, please cite:

@inproceedings{pilcher2025purposefully,
  title={Purposefully Induced Psychosis (PIP): Embracing Hallucination as Imagination in Large Language Models},
  author={Pilcher, Kris and T\"{u}t\"{u}nc\"{u}, Esen K.},
  booktitle={Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems},
  year={2025},
  publisher={ACM},
  address={Yokohama, Japan}
}

License

This model inherits the license of the base model (Gemma-3-1b-it) and follows Hugging Face's Model License.

  • PEFT 0.15.2
Downloads last month
64
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for markredito/gemma-pip-finetuned-v2

Adapter
(80)
this model

Dataset used to train markredito/gemma-pip-finetuned-v2

Space using markredito/gemma-pip-finetuned-v2 1