Open-source JumpReLU-based sparse autoencoders trained on all layers of Gemma 2, enhancing interpretability of large language models.

Tom Lieberum, Senthooran Rajamanoharan, Arthur Conmy et al.

Enhancing students' higher-order cognitive skills by integrating exploratory search with Kolb's learning model.

Yiming Luo, Patrick Cheong-Iao Pang, Shanton Chang

Bloom unifies observational and interventional causal discovery through bilevel polynomial optimization with SDP-based convergence and optimality guarantees.

Qiu Chengbo, Yang Kai

Study optimizes LLM fine-tuning using sparse Mixture of Experts models, enhancing accuracy and runtime efficiency.

Yuchen Xia, Jiho Kim, Yuhan Chen et al.

Deep generative models (VAE, GAN, DM) enable robots to learn multimodal demonstrations, improving generalization and task diversity.

Julen Urain, Ajay Mandlekar, Yilun Du et al.

ToolSandbox is a stateful, conversational benchmark for evaluating LLM tool use, emphasizing implicit state dependencies and dynamic milestone tracking.

Jiarui Lu, Thomas Holleis, Yizhe Zhang et al.

CodexGraph enhances LLM performance on codebases via code graph databases, significantly improving retrieval accuracy.

Xiangyan Liu, Bo Lan, Zhiyuan Hu et al.

A GPT-4-based multi-agent system automates animated data video creation, optimizing task decomposition and workflow for end-to-end automation.

Leixian Shen, Haotian Li, Yun Wang et al.

Proposes a compute-optimal strategy using adaptive resource allocation, outperforming model parameter scaling with over 4x efficiency gains.

Charlie Snell, Jaehoon Lee, Kelvin Xu et al.

Fine-tuning small models improves their Theory of Mind task performance, achieving 46% alignment with large models.

Nunzio Lore, Sepehr Ilami, Babak Heydari

Lumina-mGPT is a decoder-only multimodal autoregressive model achieving high-quality, flexible photorealistic image generation via multimodal pretraining and progressive fine-tuning.

Dongyang Liu, Shitian Zhao, Le Zhuo et al.

Study finds multimodal LLMs in GUI environments are susceptible to environmental distractions, affecting their faithfulness.

Xinbei Ma, Yiting Wang, Yao Yao et al.

INSEC attack injects comment strings to increase LLM code completion insecurity by over 50%.

Slobodan Jenko, Niels Mündler, Jingxuan He et al.

This study compares GPT-4, Llama 2, and Gemini in zero-shot tasks across English and low-resource South Asian languages, revealing significant performance gaps.

Md. Arid Hasan, Prerona Tarannum, Krishno Dey et al.

Unified framework categorizes data assessment into quality, diversity, importance; improves instruction tuning efficiency and performance.

Yulei Qin, Yuncheng Yang, Pengcheng Guo et al.

MedSyn framework combines LLM and MKG to generate synthetic medical text, improving ICD code prediction accuracy by 17.8%.

Gleb Kumichev, Pavel Blinov, Yulia Kuzkina et al.

Re-Invoke achieves a 20% nDCG@5 improvement in single-tool retrieval and 39% in multi-tool on ToolE dataset.

Yanfei Chen, Jinsung Yoon, Devendra Singh Sachan et al.

MiniCPM-V employs architecture optimization to enable GPT-4V-level performance on mobile devices, supports 30+ languages, and surpasses prior models in accuracy.

Yuan Yao, Tianyu Yu, Ao Zhang et al.

Introduces SGCL method to improve recommendation accuracy by combating noisy views, achieving a 12.25% improvement.

Chu Zhao, Enneng Yang, Yuliang Liang et al.

Lightweight coarse correspondence method boosts spatial-temporal reasoning in multimodal LLMs by 20%, no architecture change needed.

Benlin Liu, Yuhao Dong, Yiqin Wang et al.