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/

Global: Zendy Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Despite the recent advancement in large language models (LLMs) and their highperformances across numerous benchmarks, recent research has unveiled that LLMssuffer from hallucinations and unfaithful reasoning. This work studies aspecific type of hallucination induced by semantic associations. Specifically,we investigate to what extent LLMs take shortcuts from certain keyword/entitybiases in the prompt instead of following the correct reasoning path. Toquantify this phenomenon, we propose a novel probing method and benchmarkcalled EureQA. We start from questions that LLMs will answer correctly withutmost certainty, and mask the important entity with evidence sentencerecursively, asking models to find masked entities according to a chain ofevidence before answering the question.  During the construction of the evidence, we purposefully replace semanticclues (entities) that may lead to the correct answer with distractor clues(evidence) that will not directly lead to the correct answer but require achain-like reasoning process. We evaluate if models can follow the correctreasoning chain instead of short-cutting through distractor clues. We find thatexisting LLMs lack the necessary capabilities to follow correct reasoning pathsand resist the attempt of greedy shortcuts. We show that the distractorsemantic associations often lead to model hallucination, which is strongevidence that questions the validity of current LLM reasoning.

Deceiving Semantic Shortcuts on Reasoning Chains: How Far Can Models Go without Hallucination?