Think Before You Drive: World Model-Inspired Multimodal Grounding for Autonomous Vehicles

UW-BioNLP presented their methods for extracting chemotherapy timelines from clinical notes at the ChemoTimelines 2025 shared task, focusing on strategies like chain-of-thought thinking and supervised fine-tuning. Their best-performing model, fine-tuned Qwen3-14B, achieved a score of 0.678 on the test set leaderboard.

The Natural Language Actor-Critic (NLAC) algorithm has been introduced to enhance the training of large language model (LLM) agents, which interact with environments over extended periods. This method addresses challenges in learning from sparse rewards and aims to stabilize training through a generative LLM critic that evaluates actions in natural language space.

LORE, a large generative model for e-commerce search relevance, has been developed over three years, achieving a 27% improvement in online GoodRate metrics. This framework emphasizes a systematic approach to relevance, breaking it down into distinct capabilities such as knowledge, reasoning, and multi-modal matching.

A new framework called MSME has been proposed for zero-shot stance detection, addressing the limitations of large language models (LLMs) in understanding complex real-world scenarios. This Multi-Stage, Multi-Expert framework consists of three stages: Knowledge Preparation, Expert Reasoning, and Pragmatic Analysis, which aim to enhance the accuracy of stance detection by incorporating dynamic background knowledge and recognizing rhetorical cues.

The TaoSR1 framework has been introduced to enhance query-product relevance prediction in e-commerce search, addressing limitations of existing BERT-based models by incorporating Large Language Models (LLMs) and a structured Chain-of-Thought (CoT) approach. The framework consists of three stages: Supervised Fine-Tuning, offline sampling with Direct Preference Optimization, and dynamic sampling to reduce hallucination errors.

ExPairT-LLM has been introduced as an exact learning algorithm for code selection, addressing the challenges in code generation by large language models (LLMs). It utilizes pairwise membership and equivalence queries to enhance the accuracy of selecting the correct program from multiple outputs generated by LLMs, significantly improving success rates compared to existing algorithms.

Astra has been introduced as a pioneering multi-agent system designed for optimizing GPU kernel performance, addressing a long-standing challenge in high-performance computing and machine learning. This system leverages existing CUDA implementations from SGLang, a framework widely used for serving large language models (LLMs), marking a shift from traditional manual tuning methods.

A new framework named Finetune-RAG has been introduced to enhance the factual accuracy of large language models (LLMs) by addressing the issue of hallucinations that arise from imperfect information retrieval in Retrieval-Augmented Generation (RAG). Experimental results indicate a 21.2% improvement in factual accuracy over the base model, alongside the introduction of Bench-RAG, an evaluation pipeline designed to test models under realistic conditions.