Unverified50% confidenceFactExact time
Hallucinations in SFT data will be learned by the model as an exemplar; what the model learns is the behavioral pattern of confidently fabricating at the corresponding position rather than specific facts; the effect of SFT depends largely on data quality rather than data quantity.
1
Sources
50%
Confidence
Long-term
Relevance
10/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe essence of model hallucination is probabilistic completion by the model outside the distribution of training data; core methods for reducing hallucination include providing contextual anchors, requiring the model to explicitly cite sources, and combining RAG to provide factual grounding.74% similarVerifiedThe technical root cause of hallucination in large models lies in the autoregressive generation mechanism, where the model predicts only the statistically most likely next token based on the preceding context; researchers divide hallucinations into intrinsic hallucinations and extrinsic hallucinations.71% similarVerifiedCalibration means that the probabilities output by the model should match reality; neural networks generally have an overconfidence tendency, so model probabilities should not be blindly trusted, and thresholds should be set only after validation with labeled data.71% similarVerifiedSupervised Fine-Tuning (SFT) trains the model using 'prompt-response' example pairs, teaching the model to respond in a conversational format, thereby making outputs more coherent.70% similarVerifiedThe technical essence of large model hallucination lies in the fact that the model learns a probability distribution of language patterns during training rather than facts themselves.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1121450API
curl https://kongchang.com/api/v1/knowledge/claims/1121450MCP
get_claim(id=1121450)