The paper proposes a model-based evaluation framework that merges multidimensional item response theory (IRT) with question context embeddings to predict large language model (LLM) performance on unseen questions. By representing LLMs with latent capability profiles and incorporating question content to inform item characteristics, the approach improves prediction accuracy over model-free baselines in within-scenario settings and offers a richer description of capability variation than unidimensional models. However, the study also finds that this generalizability does not reliably extend to cross-scenario shifts, indicating a key limitation for broader application.
By Ergan Shang, Weijing Tang, Yinqiu He
arXiv:2602. 12424v2 Announce Type: replace-cross Abstract: Benchmarks establish a standardized evaluation framework to systematically assess the performance of large language models (LLMs), facilitating objective comparisons and driving advancements in the field.
By Ziqian Zhang, Xingjian Hu, Yue Huang, Kai Zhang, Ruoxi Chen, Yixin Liu, Qingsong Wen, Kaidi Xu, Xiangliang Zhang, Neil Zhenqiang Gong, Lichao Sun
arXiv:2511. 21692v3 Announce Type: replace-cross Abstract: We investigate how well large language models (LLMs) generalize across different task difficulties, a key question for effective data curation and evaluation.
By Yeganeh Kordi, Nihal V. Nayak, Max Zuo, Ilana Nguyen, Stephen H. Bach
arXiv:2610.01627v1 Announce Type: cross
Abstract: Difficulty is one of the most fundamental properties of a question: it determines whether the question can meaningfully discriminate between models o...
By Peng Cui, Qiaoyuan Zheng, Rudolf Debelak, Mrinmaya Sachan
arXiv:2608. 07353v1 Announce Type: cross Abstract: Understanding concepts is fundamental to generalization.
By Karim Radouane, Jose G Moreno, Lynda Tamine
arXiv:2510.01030v2 Announce Type: replace
Abstract: The human ability to translate diverse perceptual and linguistic inputs into structured behavior has been thought to rest on learning robust repres...
By Zach Studdiford, Timothy T. Rogers, Kushin Mukherjee, Siddharth Suresh
The paper proposes a response‑free method for estimating difficulty of reading‑comprehension multiple‑choice items by fine‑tuning a transformer on item wording. It introduces two extensions to a baseline joint‑encoding model: a component‑wise variant that encodes passage, question, and options separately, and a multi‑task variant that adds a question‑answering auxiliary task. Experiments on a corpus of nearly 30,000 items show that both extensions outperform the baseline, especially the multi‑task variant across all metrics and the component‑wise variant in rank ordering, even with limited training data.
By Jan Net\'ik, Patr\'icia Martinkov\'a
IDEAlign introduces a new protocol for evaluating the similarity of large language model (LLM) annotations to expert judgments. It uses pick‑the‑odd‑one‑out tasks to capture expert similarity and benchmarks various similarity methods—including text embeddings, topic models, and LLM-as-a-judge—against these human ratings. Applied to educational datasets, the study finds that most metrics miss nuanced expert dimensions, with LLM-as-a-judge performing best yet still insufficient for full expert alignment.
By Hyunji Nam, Lucia Langlois, James Malamut, Mei Tan, Dorottya Demszky
arXiv:2603. 23522v2 Announce Type: replace-cross Abstract: Evaluating large language models (LLMs) on open-ended questions is difficult because response quality depends on the question's context.
By Shanghua Gao, Yuchang Su, Pengwei Sui, Curtis Ginder, Marinka Zitnik
The paper introduces Prompt2Box, a method that embeds prompts into a box embedding space to capture both semantic similarity and specificity relations, addressing the limitation of traditional vector embeddings that conflate topical similarity with specificity. Using a trained encoder on existing and synthesized datasets, Prompt2Box achieves significant improvements, reducing specificity prediction error by 45% over a prompt-length baseline and identifying 13.5% more LLM weaknesses in hierarchical clustering compared to vector baselines. The authors also present a novel dimension‑reduction technique for visualizing and comparing box embeddings, and provide the code on GitHub.
By Neeladri Bhuiya, Shib Sankar Dasgupta, Andrew McCallum, Haw-Shiuan Chang
The paper introduces a knowledge‑graph‑based evaluation framework, S3KG, to assess whether large language models truly understand context in question answering tasks. S3KG combines structural and semantic signals into a single similarity score and is paired with a diagnostic analysis that pinpoints reasoning errors at the triplet level. Across nine benchmarks, the method outperforms existing baselines, achieving up to +7.6 F1 points and an AUROC of 0.973.
By Subavarshana Arumugam, Mamta Nallaretnam, Kithuni Wickramasinghe, Chamath Gunapala, Pragatheeswaran Vipulanandan, Kamal Premaratne, Uthayasanker Thayasivam
A*-Thought-V2 is a framework that models Chain-of-Thought reasoning as a geometric trajectory in a 3D PCA space, using explicit-implicit latent tokens to compress steps that deviate from the main question-to-solution direction. The method measures alignment angles to decide which steps remain text and which become latent, and introduces stepwise embedding forcing and label forcing to train the architecture. Experiments on Qwen models show up to 2.6% accuracy gains, halved response length, and significant reductions in computation and training time.
By Xiaoang Xu, Siyuan Liu, Shuo Wang, Junlan Feng, Fanyu Meng, Zhu Zhang, Jixun Wang, Xiaorong Wang, Zihan Zhou, Xin Li, Chaojun Xiao, Yiming Zhang, Huijia Wu, Liuyu Xiang, Peipei Li, Zhaofeng He