Hirundo's Bias Unlearning Tech Reduces DeepSeek-R1 Model Bias by up to 76%

Hirundo successfully reduced bias in the DeepSeek-R1-Distill-Llama-8B large language model using its novel bias unlearning technology. While DeepSeek-R1 excels at reasoning tasks, it exhibited significant bias related to race, nationality, and gender. Hirundo's method achieved up to a 76% bias reduction across categories without impacting model performance, demonstrating a robust proof of concept for safer AI deployment. This technology will soon be available on Hirundo's platform, and the bias-unlearned model has been released on Hugging Face.