By: Leo Scarin
Explore the meaning and patterns of your gossip by training a tiny language model on your spiciest tea 🌶
Description of the work:
Artificial Intelligence can sometimes seem like a mysterious or magical technology, something only experts can create or understand. But in reality, AI is accessible, and everyone (even you!) can build their own simple AI models to explore how this technology works and what it can reveal about our world.
Leo Scarin is a multimedia artist and coder who experiments with building AI models. He explores how technology shapes social and environmental experiences through interactive art. By blending video, design, and code, Leo shows how AI can be a tool for creativity and connection.
In this assignment, Leo helps you to experiment with AI in a hands-on way. By collecting your own gossip data, you can train a small AI model, and use it to uncover hidden patterns. This is a way to better understand AI as a creative and critical tool, that you can shape and use yourself.
Assignment:
- Document all your gossip for a month. You can write it down in a journal or your notes app, record a voice note, or collect in a way that feels natural to you.
- Archive it as a list of 100-200 brief sentences. You’ll put them into a .txt file one gossip per line, so that the computer can understand it. See How to prepare your data set?
- Fine-tune a small language model with your gossip list until you get a good (enough) new model. Here, we guide you How to fine-tune?
- Once your gossip model is ready, try interacting with it. You can give it short prompts (like questions or sentence starters) and see how it responds. Think of it like chatting with a version of your own gossip voice. See How to infer? to help you run the model and type in prompts
- Use a simple text-to-speech (TTS) tool to turn your model’s answers into audio. This way, your gossip machine can literally speak its mind. Try out different voices or tones to match the personality of your model. See How to tts? to guide you to do this.Â
- Explore the meaning behind your gossip. Look at the words your gossip machine uses the most and how they connect to each other. You can create a simple map or chart to show these patterns. This helps you understand what your model talks about and what ideas come up again and again.
The training is in Google Colab, the best way is to run it there, while reading the Notion notes, where you can find a more in-depth explanation of all functions.