Unverified50% confidenceFactExact time
Existing hallucination mitigation approaches (corpus grounding, mutual verification, preference alignment) mostly treat hallucinated terms as generic noise and filter them using consistency heuristics, lacking mechanistic explanations and principled criteria for determining how many expansion terms to retain.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified模型规模越大，幻觉的表述往往越流畅、越具迷惑性，识别难度反而上升75% similarUnverifiedHallucinated proofs differ from rounding errors in numerical computation or obvious error reports in symbolic computation, as the latter two tend to be exposed in subsequent steps, whereas hallucinated proofs may remain undetected until discovered by peer review.75% similarUnverifiedWhen faced with ambiguous instructions, large models tend to fill information gaps with content that appears plausible but is unverified (hallucination completion), and explicitly writing out constraints and acceptance criteria can suppress this completion impulse.74% similarUnverified即使通过Prompt明确要求禁止捏造，模型仍然可能在高置信度场景下输出幻觉内容74% similarUnverifiedThe study finds that on certain models, prompt perturbation can actually mitigate bias and hallucination, which contradicts the assumptions of some prior research.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1002095API
curl https://kongchang.com/api/v1/knowledge/claims/1002095MCP
get_claim(id=1002095)