Stefano Cresci

This work introduces a general control-theoretic framework for composing moderation actions according to their expected effects on an evolving platform and compares two control-theoretic moderators against several baselines and local strategies to demonstrate the advantages of studying content moderation as a global, adaptive, and sequential decision problem.

Contextualized AI-generated counterspeech can significantly enhance persuasiveness, but not all personalization strategies are equally effective.

Despite increased systemic risks during high-stakes elections, social media platforms appear to make no meaningful adjustments to their content moderation strategies, casting doubt on the effectiveness of current self-regulatory approaches.

Turns out, you can spot LLM hallucinations just by looking at how tightly their answers cluster in embedding space, enabling a surprisingly effective way to flag bad responses with minimal labeling.