English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

While auxiliary information has become a key to enhancing Large LanguageModels (LLMs), relatively little is known about how LLMs merge these contexts,specifically contexts generated by LLMs and those retrieved from externalsources. To investigate this, we formulate a systematic framework to identifywhether LLMs' responses are attributed to either generated or retrievedcontexts. To easily trace the origin of the response, we construct datasetswith conflicting contexts, i.e., each question is paired with both generatedand retrieved contexts, yet only one of them contains the correct answer. Ourexperiments reveal a significant bias in several LLMs (GPT-4/3.5 and Llama2) tofavor generated contexts, even when they provide incorrect information. Wefurther identify two key factors contributing to this bias: i) contextsgenerated by LLMs typically show greater similarity to the questions,increasing their likelihood of being selected; ii) the segmentation processused in retrieved contexts disrupts their completeness, thereby hindering theirfull utilization in LLMs. Our analysis enhances the understanding of how LLMsmerge diverse contexts, offers valuable insights for advancing current LLMaugmentation methods, and highlights the risk of generated misinformation forretrieval-augmented LLMs.

Blinded by Generated Contexts: How Language Models Merge Generated and Retrieved Contexts When Knowledge Conflicts?