Training and open-sourcing clean, efficient base models like KL3M.
How can regulators, developers, or users trust AI systems if they can't inspect or understand the models that power them?
We believe that open models are essential for the development of AI systems that create positive social, economic, and environmental impacts. That's why we are committed to training and open-sourcing clean, efficient base models like KL3M.
In addition to open sourcing these models, we also provide educational resources and tools to help individuals and organizations understand how to train and fine-tune models in ways that promote legal, ethical, and sustainable AI outcomes.
KL3M, originally known as the Kelvin Legal Large Language Model, is a clean, efficient base model that has been certified by Fairly Trained.
More →All the Patents is a fine-tuned version of KL3M that has been trained to generate obvious patent applications and contribute them to the public domain.
More →We are currently developing tools and educational resources to help individuals and organizations understand how to train and fine-tune models in ways that promote legal, ethical, and sustainable AI outcomes.