ReST-MCTS* integrates tree search with automatic reward inference, boosting LLM reasoning accuracy by auto-generating high-quality reasoning traces.

Dan Zhang, Sining Zhoubian, Ziniu Hu et al.

OmegaPRM employs automated Monte Carlo Tree Search to collect process supervision data, improving large language models' math reasoning success rates by over 18%.

Liangchen Luo, Yinxiao Liu, Rosanne Liu et al.

MediQ introduces an interactive benchmark for medical reasoning, significantly improving diagnostic accuracy by 22.3% through proactive question-asking strategies.

Shuyue Stella Li, Vidhisha Balachandran, Shangbin Feng et al.

Proposed four error detection and correction tasks, built 1800-instance dataset, GPT-4 leads with 93.9%, improving 47.9% with prompts.

Xiaoyuan Li, Wenjie Wang, Moxin Li et al.

SWIF2T: a multi-model framework generating specific, actionable scientific feedback, outperforming baselines with 72% top plan ranking.

Eric Chamoun, Michael Schlichktrull, Andreas Vlachos

Proposes a post-processing method using language models to enhance semantic similarity and privacy in differentially private text rewriting, with empirical results showing significant improvements.

Stephen Meisenbacher, Florian Matthes

PathReasoner models reasoning paths with equivalent extension, excelling in logical QA.

Fangzhi Xu, Qika Lin, Tianzhe Zhao et al.

Proposed QUIRE algorithm enhances CoT effectiveness and faithfulness by recalling extra information from questions; experiments show 2.4% effectiveness and 5.6% faithfulness improvement.

Jiachun Li, Pengfei Cao, Yubo Chen et al.

Proposes a language generation method using strict proper scoring rules (Brier, spherical), improving BLEU by 1.5-2.0 points on translation tasks.

Chenze Shao, Fandong Meng, Yijin Liu et al.

Survey on tool learning with large language models, analyzing benefits and implementation methods.

Changle Qu, Sunhao Dai, Xiaochi Wei et al.

KG-FIT leverages LLM-guided hierarchy construction and multi-objective fine-tuning to improve link prediction by 14.4% on FB15K-237.

Pengcheng Jiang, Lang Cao, Cao Xiao et al.

COLT improves completeness-oriented tool retrieval; BERT-mini (11M) beats BERT-large (340M).

Changle Qu, Sunhao Dai, Xiaochi Wei et al.

RepNoise employs layer-wise noise injection and mutual information minimization to resist harmful fine-tuning, preserving model capabilities.

Domenic Rosati, Jan Wehner, Kai Williams et al.

Introduces parametric WKM to enhance LLM-based agent planning, reducing trial-and-error and hallucinations, achieving superior results on complex tasks.

Shuofei Qiao, Runnan Fang, Ningyu Zhang et al.

xFinder optimizes key answer extraction in LLM evaluation, achieving 93.42% extraction accuracy.

Qingchen Yu, Zifan Zheng, Shichao Song et al.

Chameleon uses early fusion to generate and understand images and text, outperforming Llama-2.

Chameleon Team

Proposed CoRE system uses large language models as interpreters, unifying natural language, pseudo-code, and flow programming for AI agents.

Shuyuan Xu, Zelong Li, Kai Mei et al.

Automated model and data card generation using LLMs enhances documentation completeness and accountability.

Jiarui Liu, Wenkai Li, Zhijing Jin et al.

This study shows that fine-tuning LLMs on new knowledge slows learning and increases hallucinations, supporting pretraining as the main knowledge source.

Zorik Gekhman, Gal Yona, Roee Aharoni et al.

Enhancing Turkish LLMs through adaptation and benchmarking strategies.

Emre Can Acikgoz, Mete Erdogan, Deniz Yuret