Unverified50% confidenceFactExact time
Constrained decoding or classification head methods allow the model to output only the highest-probability label from a limited set of candidates, skipping the process of generating free text token by token
1
Sources
50%
Confidence
Long-term
Relevance
10/6/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGrounding prompts that instruct the model to answer only based on provided documents and cite the relevant passages can greatly reduce the probability of fabrication, but cannot eliminate it entirely.69% similarUnverifiedDedicated classifiers are faster and cheaper than having autoregressive LLMs generate answers token by token on classification tasks.68% similarUnverifiedEven when the model is explicitly required to answer only based on retrieved documents, it still fills in gaps using parametric memory and generates seemingly well-formatted citations.68% similarUnverified针对特定触发词的后门攻击可让模型在大多数情况下表现正常，仅在遇到特定输入时才输出预设错误信息，使常规评测基准难以捕捉67% similarUnverifiedThe approach of having the model assemble requests on its own solely based on API documentation has limitations: it may hallucinate parameters, overlook authentication, handle errors improperly, and requires stuffing complete documentation into the context, causing context bloat, increasing costs, and reducing accuracy.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/982501API
curl https://kongchang.com/api/v1/knowledge/claims/982501MCP
get_claim(id=982501)