A creative exploration of language model capabilities through the generation of humorous, AI-created short stories in the style of the classic FML (F*** My Life) format. This project bridges a decade of NLP advancement, from simple n-gram models to modern transformers.

Project Evolution

2013: Original implementation using n-gram Markov Chain models
2022: Modern reimplementation with transformer-based language models
Demonstrates dramatic improvements in coherence while maintaining absurdist humor

Technical Implementation

The project leverages:

Hugging Face Transformers library for model integration
GPT-Neo model for text generation
~25,000 FML stories collected in 2013 as training data
Configurable generation parameters:
- Temperature control for randomness
- Top-k and top-p sampling strategies
- Custom prompting for themed stories

Features

Generate random FML-style stories
Prompt-based story generation for specific themes
Adjustable creativity parameters
Comparison framework between old and new approaches

Example Output

“I could fly. Seconds later my flight ended. I fell down my shirt. I’m allergic to cats. FML”

The project successfully captures the signature blend of mundane disasters and absurd situations that characterized the original FML website, while showcasing how far natural language generation has advanced.