An article published today in TheMarker highlights the work of Hirundo, which seeks to address one of the most complex issues in the field of artificial intelligence: political biases and censorship embedded in AI models.
According to the company’s data, after applying its technology to the Chinese Qwen model, the rate of responses exhibiting political bias dropped from 90% to approximately 3% in a test of 500 questions on sensitive topics, while largely maintaining the model’s performance in other areas.
The development demonstrates how research and technology can help address new challenges in the field of AI.
Read the full article in TheMarker here