Unverified50% confidenceFactExact time
The feedback learning of semantic-registry does not fine-tune the LLM itself, but uses correction signals to update the ranking strategy of the retrieval layer.
1
Sources
50%
Confidence
Long-term
Relevance
10/29/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen generating text, an LLM does not maintain a queryable factual database, but predicts the next word one by one based on context; information that is high-frequency, coherent and repeatedly appearing in the training data is reproduced more stably, while low-frequency, contradictory or fragmentary facts are easily smoothed by the language model's fluency into seemingly complete narratives.73% similarVerifiedThe root cause of citation hallucination is that LLMs learn linguistic patterns by predicting the next word, and no factual database for real-time verification is embedded in the training process.73% similarUnverifiedLLM的预训练本质是在海量文本上进行下一词预测（Next Token Prediction），通过Transformer架构的注意力机制建立词语间语义关联72% similarUnverifiedDuring LLM training, memorization occurs, where models may verbatim store fragments of training data and reproduce them during inference.72% similarUnverifiedThe feedback loop described in the article is implemented through prompt iteration and few-shot example updates, which is a form of context engineering rather than model fine-tuning.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1110468API
curl https://kongchang.com/api/v1/knowledge/claims/1110468MCP
get_claim(id=1110468)