Unverified50% confidenceFactExact time
Lexical distractors refer to words or phrases inserted into the text that are superficially similar to the target fact but semantically irrelevant; they do not change the fact itself, yet they change the model's internal representations and gradient signals.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGenerative summarization may introduce information not present in the source text — known as factual drift — because models reorganize content in their own language.70% similarUnverifiedWhen a prompt needs three or four unrelated tags attached to describe its function, this itself is a strong signal that the prompt should be split.69% similarUnverifiedWhen small-parameter models produce structured output, the attention mechanism is more easily drawn to natural-language labels that appear repeatedly or in prominent positions in the prompt, while relatively ignoring the enum definitions in the Schema.69% similarUnverifiedAn engineeringly reliable embedding model should exhibit minimal vector change under character-level and lexical-level perturbations, highly similar vectors under semantic-level paraphrasing, and a perceptible distance change when there is a substantive semantic shift (such as negation).68% similarUnverifiedGrounding instructions (requiring statements only of facts present in the source text and citing exact sentences) eliminate hallucinations, because the model must omit a claim when it cannot find a supporting citation.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1115351API
curl https://kongchang.com/api/v1/knowledge/claims/1115351MCP
get_claim(id=1115351)