This study compares large language models' generalization in in-context learning vs. fine-tuning, proposing reasoning trace augmentation to improve performance.

Andrew K. Lampinen, Arslan Chaudhry, Stephanie C. Y. Chan et al.

BrowseComp-ZH benchmarks Chinese LLM web browsing with 289 multi-hop questions; most models score below 20%, best is 42.9%.

Peilin Zhou, Bruce Leon, Xiang Ying et al.

Sparse attention enhances long-sequence processing in Transformer LLMs using Quest and Vertical-Slash methods.

Piotr Nawrot, Robert Li, Renjie Huang et al.

Using prompts, ChatGPT detects and categorizes LSP translation errors with 64.1% accuracy, demonstrating strong potential for automated evaluation.

Joachim Minder, Guillaume Wisniewski, Natalie Kübler

OmniV-Med unifies 2D/3D images and videos in a medical vision-language model, using rotary position encoding and token pruning to achieve SOTA performance across 7 benchmarks.

Songtao Jiang, Yuan Wang, Sibo Song et al.

Nemotron-CLIMB employs semantic embedding and iterative search to optimize data mixtures, boosting language model pre-training performance.

Shizhe Diao, Yu Yang, Yonggan Fu et al.

MLRBench evaluates LLMs' reasoning over multilingual long contexts, revealing resource gaps.

Amey Hengle, Prasoon Bajpai, Soham Dan et al.

Persona-judge achieves personalized alignment via self-judgment, enhancing alignment efficiency by 98%.

Xiaotian Zhang, Ruizhe Chen, Yang Feng et al.

Proposes an effective semantic diversity metric for LLM outputs, revealing preference tuning enhances high-quality content diversity.

Alexander Shypula, Shuo Li, Botong Zhang et al.

BrowseComp is a challenging benchmark for AI web browsing, comprising 1266 questions requiring persistent, creative search for hard-to-find information.

Jason Wei, Zhiqing Sun, Spencer Papay et al.

WebRollback enhances web agents with explicit rollback mechanisms, achieving top performance on Mind2Web-Live and WebVoyager benchmarks.

Zhisong Zhang, Tianqing Fang, Kaixin Ma et al.

AskQE employs a QA framework with LLaMA-3 70B, achieving high correlation (τ=0.878) with human judgments for critical MT error detection.

Dayeon Ki, Kevin Duh, Marine Carpuat

DeepMath-103K is a large, high-difficulty, decontaminated, verifiable math dataset that advances reasoning models.

Zhiwei He, Tian Liang, Jiahao Xu et al.

Introduces LLM-SRBench with 239 problems, evaluating LLMs' scientific equation discovery, emphasizing reasoning beyond memorization.

Parshin Shojaee, Ngoc-Hieu Nguyen, Kazem Meidani et al.

Proposes a cultural localization benchmark; finds explicit prompts boost cultural responses but reduce diversity; uses activation steering to control cross-lingual cultural responses.

Veniamin Veselovsky, Berke Argin, Benedikt Stroebl et al.

SocioVerse employs large-scale real user data and LLM agents to simulate diverse social scenarios with 84.3% accuracy in election prediction.

Xinnong Zhang, Jiayu Lin, Xinyi Mou et al.

Unified framework for diverse spoken language models (SLMs), including pure speech, speech+text, and speech-aware text models, advancing universal speech processing.

Siddhant Arora, Kai-Wei Chang, Chung-Ming Chien et al.

ASI method boosts agent task success by 23.5% and improves efficiency by 10.7-15.3% through programmatic skills.

Zora Zhiruo Wang, Apurva Gandhi, Graham Neubig et al.

NoveltyBench evaluates language models' diversity; 20 models mostly underperform human baseline, larger models show less diversity, highlighting mode collapse issues.

Yiming Zhang, Harshita Diddee, Susan Holm et al.

This tutorial reviews LLM adaptation techniques, including parametric fine-tuning and retrieval-augmented methods, emphasizing multi-stage training and evaluation frameworks.

Zixuan Ke, Yifei Ming, Shafiq Joty