Proposes 'Compile by Training' to convert natural language specs into local neural functions with 83.6% accuracy, using teacher models and adapter fine-tuning.

Yuntian Deng, Pengyu Nie, Stuart Shieber

ESPO enhances prompt accuracy by 3.76% and shortens prompts by 47% through structured error diagnosis, multi-strategy candidate generation, and bootstrap stability selection.

Lihao Liu, Peng Tang, Kunwar Yashraj Singh et al.

This study quantifies reasoning step importance via advantage, revealing models struggle to decode this solely from text.

Kevin Du, Alexander Hoyle, Laura Ruis et al.

This study demonstrates that auxiliary views (e.g., paraphrases, analogies) significantly improve knowledge acquisition in large language models, especially at larger scales.

Joseph Lee, Yidi Huang, Dokyoon Kim et al.

Introduces Last Translation Benchmark with human-reviewed challenging examples and verification rules, enabling precise failure detection.

Vilém Zouhar, Niyati Bafna, Mukund Choudhary et al.

Synthetic and real data experiments show user feedback significantly improves LLM responses; evaluation bias masks this benefit.

Shachar Don-Yehiya, Leshem Choshen, Omri Abend

This paper reveals output format confounds data quality and capability measures; spectral metrics are invariant, while directional scores effectively identify target tasks.

Chengguang Gan, Hanjun Wei, Yunhao Liang et al.

This study mechanistically uncovers how LLMs evaluate summaries, revealing a two-stage process with attention-based error detection and MLP-based score crystallization at layers 25-26.

Himil Vasava, Ming Jiang

CordisBench benchmarks LLMs' reasoning about component lifecycles in dynamic systems, with 1200 questions revealing performance drops as interactions increase.

Damien Sileo, Dimitri Kachler

Introduces StudentSim framework, combining pooled pretraining and individual fine-tuning to enhance student simulators' fidelity and guidance responsiveness.

Ke Yang, Chenglong Wang, Michel Galley et al.

A 35B MoE-based document understanding model with difficulty-aware data curation reduces deployment costs by over 80%.

Maksim Evdokimov, Matvey Ivanov, Dmitrii Tsiupin et al.

Knowledge distillation during mid-training enhances reasoning but slows factual recall; proposes Switch Distillation.

Jacqueline He, Howard Yen, Shuyue Stella Li et al.

PersuaRL leverages RL-driven multi-expert selection to significantly enhance persuasive dialogue generation in insurance.

Rohan Kirti, Akash Ghosh, Aryan Vats et al.

Introduces Phrase-Localized LCG for training-free localized accent control in code-switching TTS.

Che Hyun Lee, Sangkwon Park, Donghun Kang et al.

Proposes a disclosure gating mechanism with five levels to improve user simulation stability, achieving 0.993 correlation with original benchmark simulator.

Yao Liu, Yu He

Proposes Context-Aware Interleaved Batching for WhisperX, reducing WER to 8.2% and boosting speed by 12× by maintaining continuous context across segments.

Carlos Bain, Max Bain

AutoSciRub automatically induces task-specific, verifiable rubrics, significantly improving autonomous research agent performance by 2.08 points on ResearchClawBench and 16.8 points on AstaBench.

Xuehai Wang, Haowei Qin, Tongxin Liu et al.

Knowledge-aligned SFT reduces hallucinations, enhancing factuality.

Arthur Becker, Jakob Kemmler, David Thulke et al.

SocialReasonBench evaluates social reasoning via branching narrative videos, revealing LMMs' shortcomings in counterfactual reasoning.

Zheyu Huang, Zijing Shi, Haozhe Luo et al.

S2GE enhances decoder usability in graph-augmented LLMs via sampling and alignment, significantly boosting accuracy.

Xiaoyu Guo, Pengcheng Chen, Jiong Yu et al.