Elon Musk recently caused a stir when he spoke out about OpenAI’s ChatGPT, claiming it had a “woke” bias. To make a point about biased AI, he is planning to make a much-talked-about right-wing alternative called “TruthGPT”. However, other scientists are exploring a different approach – using AI to bridge divides between different political views.
David Rozado, a data scientist in New Zealand, drew attention to the political bias in ChatGPT and created “RightWingGPT”, an AI model that expresses more conservative viewpoints. He fine-tuned a language model called Davinci GPT-3 with additional language at a cost of hundreds of dollars. Rozado is now working on a ‘LeftWingGPT’ and a “DepolarizingGPT” model as well, which will be made available on the website of a centrist think tank, the Institute for Cultural Evolution.
The issue of political bias in language models is a hot topic. Several conservative organizations are already creating their own versions of ChatGPT. The Chinese government recently issued guidelines to shape the political outlook of language models, and OpenAI has encouraged users to explore AI models with user-defined values. While some of these models may end up reinforcing particular ideologies, Rozado hopes that, in the end, AI can help bridge political divisions rather than exacerbate them.