Asif Ekbal

Safety evaluations of LLMs vary dramatically across languages, revealing that some harmful content is far more susceptible to manipulation than previously understood.

Context selectively shifts generative responsibility in translation, with function words relying heavily on surrounding sentences while content words remain stable.

SAE-based explanations can be certified for faithfulness, revealing that later layers of language models are significantly easier to interpret without sacrificing predictive accuracy.

Achieve more human-like negotiation from dialogue agents by explicitly modeling and reasoning about emotions with interpretable chain-of-thought prompting.