5 papers
Source Preference in the Wild: How LLM Agents Favor Items by Source, and How to Reduce It
Jonghyun Song, Haewon Park, Jeonghoon Shim +2
As LLM agents decide on users' behalf which product to buy, which hotel to book, or which paper to cite, a preference for items from certain sources (the sites or services they com…
Agents' Overreliance on Unreliable Tools
Hoyeol Yang, Woojung Song, Taewon Kim +3
LLM agents use tools to access information and perform computations beyond their parametric knowledge. Existing tool-use benchmarks evaluate whether agents select and call the righ…
SHAPE of Thinking: Mathematical Heuristics for Analyzing and Improving LLM Reasoning
Jonghyun Song, Sangjun Song, Minjae Oh +3
Large language models (LLMs) achieve strong performance on mathematical reasoning benchmarks, yet final-answer accuracy reveals little about the strategies they use. We introduce \…
Sparse and Dense Retrievers Learn Better Together: Joint Sparse-Dense Optimization for Text-Image Retrieval
Jonghyun Song, Youngjune Lee, Gyu-Hwung Cho +3
Vision-Language Pretrained (VLP) models have achieved impressive performance on multimodal tasks, including text-image retrieval, based on dense representations. Meanwhile, Learned…
Comparing Neighbors Together Makes it Easy: Jointly Comparing Multiple Candidates for Efficient and Effective Retrieval
Jonghyun Song, Cheyon Jin, Wenlong Zhao +2
A common retrieve-and-rerank paradigm involves retrieving relevant candidates from a broad set using a fast bi-encoder (BE), followed by applying expensive but accurate cross-encod…