Knowledge Tree
prompt-induced hallucination
Also known as: prompting-induced hallucinations, prompt-driven hallucinations
Facts (12)
Sources
Survey and analysis of hallucinations in large language models frontiersin.org Sep 29, 2025 6 facts
referencePrevious studies by Ji et al. (2023), Wei et al. (2022), and Chen et al. (2023) have hinted at the duality of hallucination attribution (prompt-induced vs. model-intrinsic) but have not developed a formal diagnostic framework.
claimThe authors of 'Survey and analysis of hallucinations in large language models' introduce an attribution framework that distinguishes prompt-induced from model-intrinsic hallucinations using controlled prompt manipulation and model comparison.
claimPrompting-induced hallucinations occur when prompts are vague, underspecified, or structurally misleading, which pushes the model into speculative generation, as noted by Reynolds and McDonell (2021), Wei et al. (2022), and Zhou et al. (2022).
claimPrompting-induced hallucinations in large language models often arise from ambiguous formulations or a lack of context, which causes the model to rely on probabilistic associations rather than grounded knowledge.
claimAttribution scoring allows for the effective distinction between prompt-driven hallucinations and model-intrinsic hallucinations.
measurementCPS (Prompt Sensitivity) analysis showed LLaMA 2 had a value of 0.15 for vaguely specified prompts, indicating susceptibility to prompt-induced hallucinations.
Detecting and Evaluating Medical Hallucinations in Large Vision ... arxiv.org Jun 14, 2024 6 facts
claimThe MediHall Score assigns hallucination scores based on six categories: Catastrophic Hallucinations, Critical Hallucinations, Attribute Hallucinations, Prompt-induced Hallucinations, Minor Hallucinations, and Correct Statements.
referenceFigure 12 in the paper 'Detecting and Evaluating Medical Hallucinations in Large Vision ...' presents examples of AI outputs classified as Prompt-induced Hallucination.
procedureThe hierarchical classification method for medical text hallucinations proposed in 'Detecting and Evaluating Medical Hallucinations in Large Vision-Language Models' categorizes sentence-level hallucination outputs into five levels: Catastrophic Hallucinations, Critical Hallucinations, Attribute Hallucinations, Prompt-induced Hallucinations, and Minor Hallucinations.
claimNearly all models show prompt-induced hallucinations close to or exceeding the number of catastrophic hallucinations when presented with counterfactual questions, indicating that Large Vision-Language Models (LVLMs) are highly vulnerable to such attacks.
claimPrompt-induced Hallucinations in medical LVLMs are induced by prompts containing confused information, often arising from a lack of plausibility or factuality in the prompt, and are used to test the model's robustness in specific contexts.
measurementFor counterfactual questions, if an LVLM response is not comprehensive due to the inherent confusion of the question, it is categorized as a prompt-induced hallucination and yields a MediHall Score of 0.6.