Spaces:
Sleeping
Apply for community grant: Personal project (gpu)
I am conducting a research project that seeks to investigate how LLMs can be fine-tuned to better reflect the cultural values of the communities in which they are deployed. The central research question is: “Given the rising global adoption of LLMs, how can we adapt these models to reflect the values and traditions of their deployment cultures, rather than propagate Western defaults?” The answers of a carefully conducted research would help to avoid a potential homogeneity of values, as LLMs continue on their upward adoption trajectory, and help tackle cultural biases in these models. The project hopes to produce a research paper with two key contributions:
A benchmarking framework for evaluating the cultural alignment of LLMs – that is, how well a model understands and reflects the values, norms, and knowledge traditions of a given culture.
A generalizable fine-tuning methodology using culturally-grounded datasets to improve this alignment, tested using Nigeria as a case study.