WebCPM introduces interactive web search with behavior imitation, constructing a Chinese long-form QA dataset, achieving 32.5% answer quality comparable to humans.

Yujia Qin, Zihan Cai, Dian Jin et al.

InstructBLIP uses instruction tuning on BLIP-2, achieving SOTA zero-shot performance across 13 unseen vision-language datasets.

Wenliang Dai, Junnan Li, Dongxu Li et al.

VideoChat integrates video foundation models with LLMs via learnable interfaces, enabling advanced spatiotemporal reasoning and causal inference.

KunChang Li, Yinan He, Yi Wang et al.

Proposes a generative supervised learning model combining probabilistic morphisms and kernel mean embeddings, ensuring measure measurability and generalization.

Hông Vân Lê

Proposes ThinkTwice scalable decoder for end-to-end autonomous driving, combining coarse prediction, scene imagination, region retrieval, and multi-layer refinement, outperforming SOTA.

Xiaosong Jia, Penghao Wu, Li Chen et al.

ImageBind learns a unified embedding space for six modalities, enabling zero-shot cross-modal retrieval, composition, detection, and generation, surpassing specialized models.

Rohit Girdhar, Alaaeldin El-Nouby, Zhuang Liu et al.

Integrating large language models (LLMs) for personalized household tidying, achieving 91.2% accuracy on unseen objects.

Jimmy Wu, Rika Antonova, Adam Kan et al.

VCSUM creates a large-scale Chinese meeting dataset with topic segmentation, multi-granularity summaries, and salient sentence annotations, enabling multi-task learning.

Han Wu, Mingjie Zhan, Haochen Tan et al.

SUR-adapter transfers LLM semantic reasoning to diffusion models, improving simple prompt image quality by over 15%.

Shanshan Zhong, Zhongzhan Huang, Wushao Wen et al.

FrugalGPT employs adaptive model cascade to reduce LLM inference costs by up to 98%, matching or surpassing GPT-4 accuracy.

Lingjiao Chen, Matei Zaharia, James Zou

AvatarReX employs NeRF with structured local implicit fields and geometry-appearance disentanglement for real-time, expressive full-body avatar synthesis, achieving high fidelity and speed.

Zerong Zheng, Xiaochen Zhao, Hongwen Zhang et al.

Identifies systematic unfaithfulness in chain-of-thought (CoT) explanations, showing bias features cause up to 36% accuracy drops, with explanations often rationalizing biased answers.

Miles Turpin, Julian Michael, Ethan Perez et al.

BIRD benchmark with 12,751 text-SQL pairs over 95 large databases; GPT-4 achieves only 54.89% execution accuracy, highlighting challenges in large-scale value understanding.

Jinyang Li, Binyuan Hui, Ge Qu et al.

Proposed SG-CQG framework uses semantic graphs and explicit control signals to enhance answer-unaware conversational question generation, achieving SOTA results.

Xuan Long Do, Bowei Zou, Shafiq Joty et al.

ProTeGi uses gradient-inspired natural language feedback and beam search to optimize prompts, improving task performance by up to 31%.

Reid Pryzant, Dan Iter, Jerry Li et al.

Cuttlefish automates low-rank training using stable rank, reducing model size by 5.6x and speeding up training by 1.2x.

Hongyi Wang, Saurabh Agarwal, Pongsakorn U-chupala et al.

Shap·E employs conditional diffusion to directly generate 3D implicit function parameters, enabling rapid, diverse 3D asset creation.

Heewoo Jun, Alex Nichol

Large Language Models (LLMs) can replace human evaluations, showing consistent results with human experts.

Cheng-Han Chiang, Hung-yi Lee

GeoLDM integrates geometric autoencoders with latent diffusion, boosting 3D molecule generation with up to 7% higher validity.

Minkai Xu, Alexander Powers, Ron Dror et al.

FMCW mmWave radar enables robust motion estimation and object detection in adverse conditions, outperforming optical sensors in fog and dust environments.

Kyle Harlow, Hyesu Jang, Timothy D. Barfoot et al.