Marco Guerini

A groundbreaking multilingual dataset empowers LLMs to generate persuasive counterspeech against the dual threats of hate and misinformation across multiple turns.

Narrowcast communication can cut civilian Fail rates in crisis evacuations, but the choice of environment representation can make or break the intervention's success.

Expert-edited counterspeech generated by LLMs is most effective when combining both fact-checking and NGO guidelines, showing that a blended approach can tackle the intertwined challenges of hate speech and misinformation.