CulturalBind is introduced, a benchmark comprising 3.2K images of 1K concepts from 6 domains, designed to evaluate culture-aware cross-modal knowledge as well as concept-level knowledge editing in VLMs, and editing methods in single and sequential scenarios are evaluated.
Malvina Nikandrou, Georgios Pantazopoulos, Ioannis Konstas et al.· 0 citations
GPTNT is a benchmark built on the cooperative video game Keep Talking and Nobody Explodes, in which two agents must coordinate to defuse procedurally generated bomb puzzles against a live countdown, to expose how models collaborate versus how they perform alone.
Amit Parekh, Sabrina McCallum, Kareem Al-Hasan et al.· arXiv.org· 0 citations
It is demonstrated that CoT encodes recoverable, token-level problem-solving information, offering new insight into how reasoning is represented and where it breaks down, suggesting complete reasoning chains are not always necessary.
Houman Mehrafarin, Amit Parekh, Ioannis Konstas· arXiv.org· 2 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.