All the Patents

Why This Project Matters

Patents are meant to encourage innovation by granting temporary monopolies on inventions. However, the system often grants patents for “obvious” ideas, creating barriers to further innovation. Our project aims to address this issue by generating and publishing obvious inventions, making them easily searchable and accessible to patent examiners and the public.

How It Works

We use a small, GPT2-sized language model called kl3m-170, trained on a clean, copyright-free dataset. This model is fine-tuned on two tasks:

1. Top-down drafting: Generating detailed claims from abstract patent information.
2. Bottom-up drafting: Creating abstracts and titles from detailed claims.

This approach allows us to generate new patents using either technique, ensuring a diverse range of “obvious” inventions.

Key Features

• Uses a small, efficient language model (kl3m-170)
• Trained on copyright-free data to avoid legal issues
• Generates patents using both top-down and bottom-up approaches
• Aims to produce a high volume of “obvious” inventions at low cost

Current Status

The project is in its early stages, focusing on:

• Data collection and curation
• Model training and evaluation
• Developing prototype tools for generation and sharing

Get Involved

We’re working to make the generation process as efficient and accessible as possible. If you’re interested in contributing or learning more, check out our GitHub repository:

[Link to GitHub repository]

Requirements

• kl3m-170m can run on systems with approximately 1GB of VRAM, CPU, or Apple Silicon

Related Resources

• USPTO Full-Text Database
• Federal Register: Updated Guidance for Making a Proper Determination of Obviousness

Join us in our mission to improve the patent system and foster true innovation!