Estimated reading time: 2 minutes
In a pioneering effort to fine-tune OpenAI’s GPT chatbot for Polish users, researchers from Wroclaw University are diligently working on a Polish version of the chatbot. The team, affiliated with the CLARIN-PL scientific consortium at the Faculty of Computer Science and Telecommunications, plans to create the largest AI research infrastructure in Poland, scheduled to commence operation in the first quarter of 2024 with public funding.
Dr. Jan Kocon from the Department of Artificial Intelligence at Wroclaw University expressed concerns about the current GPT chatbot’s limited exposure to the Polish language, leading to cultural and linguistic inaccuracies. They’ve observed issues such as hallucinations, grammatical errors, and stylistic flaws, particularly when it comes to Polish facts, culture, and history.
Wroclaw University researchers, along with AI students, are now working on refining the chatbot’s ability to compose Polish poetry and songs. The aim is to overcome issues where GPT often translates from English, failing to maintain Polish language nuances, including rhyming in poetry.
The university also introduced a Polish language interface for ChatGPT, enabling easier access for Polish users and providing researchers with invaluable data on the chatbot’s responses and areas that require improvement.
ChatGPT, developed by OpenAI in December 2022, is a generative language model that theoretically responds to any query while maintaining strong language accuracy. In addition to this effort, Wroclaw University is involved in a project led by Professor Tomasz Kajdanowicz, focused on creating a tool to facilitate searching through legal judgments. The project, named JuDDGES, has received an international grant of over one million Polish złoty.
This groundbreaking initiative by Polish researchers is not only enhancing the capabilities of AI but also ensuring it aligns better with the unique cultural and linguistic nuances of the Polish context.