Our contributions are: (1) We develop a generalized framework to optimize non-differentiable ranking-based functions byextending theerror-drivenoptimization ofAPLoss.(2)Weprovethat

Thought, the LLM can verbalize that the unknown city is Paris and use this fact to answer downstream questions.

In contrast to existing approaches, we propose a novel framework built on PanoHead, which excels in synthesizing images from a 360-degree perspective.

Doctors look at an analysis of cellular data as part of their research into using artificial intelligence to repurpose existing drugs to fight rare diseases, in Philadelphia, Pennsylvania, in February 2025. There is concern some apps that claim to offer medical guidance may not have an adequate data set to accurately asses information their users submit. Artificial intelligence is shaking up industries from software and law to entertainment and education. And as physicians like Dr. Cem Aksoy are learning, it's posing special challenges in medicine as patients tap the technology for advice. Aksoy, a medical resident at a hospital in Ankara, Turkey, says an 18-year-old patient and his family recently panicked after the young man was diagnosed with a cancerous tumor on his left leg.

Large language models (LLMs) are often ensembled together to improve overall reliability and robustness, but in practice models are strongly correlated. This raises a fundamental question: which models should be selected when forming an LLM ensemble? We formulate budgeted ensemble selection as maximizing the mutual information between the true label and predictions of the selected models. Furthermore, to explain why performance can saturate even with many models, we model the correlated errors of the models using Gaussian-copula and show an information-theoretic error floor for the performance of the ensemble. Motivated by these, we propose a simple greedy mutual-information selection algorithm that estimates the required information terms directly from data and iteratively builds an ensemble under a query budget. We test our approach in two question answering datasets and one binary sentiment classification dataset: MEDMCQA, MMLU, and IMDB movie reviews. Across all datasets, we observe that our method consistently outperforms strong baselines under the same query budget.