Proposes distributional divergence metrics (e.g., MAUVE) to address gen-PPL's failure in evaluating text quality.

Antonio Franca, Alexander Tong

TrustMargin is a training-free arbitration method combining parametric priors and evidence support, improving answer selection in large language models by 4-6 points F1.

Jingyan Xu, Hong Shi, Yi Shan et al.

CATPO uses tree informativeness scoring and critique-guided repair to improve mathematical reasoning accuracy to 37.5%.

Ayush Singh, Umang Goyal, Ankur Dahiya

PoE-Bridge enables parallel decoding in diffusion language models, achieving 5x speedup and recovering 95% performance.

Juntong Shi, Brian L. Trippe, Jure Leskovec et al.

Agentopia introduces a long-term multi-agent society simulation over 10 years, leveraging life reward-based reinforcement learning to enhance social behaviors and anthropomorphic capabilities of LLMs.

Xintao Wang, Sirui Zheng, Hongqiu Wu et al.

Proposes EmbedFilter, a linear transformation that filters the latent subspace encoding high-frequency, uninformative tokens, improving zero-shot text embedding performance by up to 14%.

Songhao Wu, Zhongxin Chen, Yuxuan Liu et al.

This paper introduces LLM-guided MAP-Elites evolution for optimizing medical decision pipelines, improving accuracy and safety metrics significantly across tasks.

Ivan Sviridov, Artem Oskin, Ivan Panin et al.

FinEvolveBench tests delayed-feedback self-evolution on 31 Chinese A-share industries and 177,324 news articles; generic memory rarely improves IC.

Zihao Deng, Yining Zhu, Leiming Wang et al.

CRAFT introduces bidirectional counterfactual reasoning, achieving 82.4% accuracy on WikiTQ and 94.6% on TabFact, outperforming baselines.

Chenshuo Pan, Yu Zhao, Jie Zhang et al.

Proposes emergent language in multi-agent RL to study consciousness-related structures without prior biases, revealing self-referential communication and echo-mismatch circuits.

Zengqing Wu, Chuan Xiao

Introduced a benchmark for data snapshot detection, evaluated open-source models, revealing significant gaps in real-world institutional document understanding.

AJ Carl P. Dy, Aivin V. Solatorio

IR3DE uses ridge regression for fast, low-cost model routing, achieving 98.4% performance in multi-domain LLM selection.

Eros Fanì, Oğuzhan Ersoy

MARDoc employs structured memory to improve multimodal long document QA, achieving 57.1% accuracy and outperforming baselines.

Kaifeng Chen, Hongtao Liu, Qiyao Peng et al.

AdaPLD achieves efficient decoding with adaptive retrieval and reuse, boosting speed by 3.10×.

Runheng Liu, Jincheng Xie, Wen Hu et al.

Introduced α-STOP for early failure alerting in dialogues and LLM-agent trajectories, improving frontier quality by 1-42%.

Avinash Baidya, Xinran Liang, Ruocheng Guo et al.

Introduced PERSUASIONTRACE, leveraging Bayesian networks to model multi-turn belief updates, scoring near humans (81 vs 80).

Jared Moore, Noah Goodman, Nick Haber et al.

Depth-Attention enhances language models by cross-layer value mixing, improving accuracy by up to 2.3 points.

Boyi Zeng, Yiqin Hao, Zitong Wang et al.

LifeSide uses multi-session Memory-Emotion-Environment loops to evaluate lifelong digital companions' long-term memory, understanding, privacy, and emotional support, based on 2000 personas and 111K tasks.

Yuqian Wu, Zhijie Deng, Wei Chen et al.

RAMPART optimizes LLM agent memory with priority-aware runtime transformation, enhancing task success.

Nikodem Tomczak

VCIFBench evaluates complex video instruction following, covering 306 tasks with multi-constraint challenges, revealing significant gaps in current models.

Huangchen Xu, Yuan Wu, Yi Chang