*=Equal Contributors
Controversy is a mirrored image of our zeitgeist, and an necessary side to any discourse. The rise of huge language fashions (LLMs) as conversational methods has elevated public reliance on these methods for solutions to their varied questions. Consequently, it’s essential to systematically study how these fashions reply to questions that pertaining to ongoing debates. Nonetheless, few such datasets exist in offering human-annotated labels reflecting the up to date discussions. To foster analysis on this space, we suggest a novel development of a controversial questions dataset, increasing upon the publicly launched Quora Query Pairs Dataset. This dataset presents challenges regarding information recency, security, equity, and bias. We consider completely different LLMs utilizing a subset of this dataset, illuminating how they deal with controversial points and the stances they undertake. This analysis finally contributes to our understanding of LLMs’ interplay with controversial points, paving the way in which for enhancements of their comprehension and dealing with of complicated societal debates.
