Unverified50% confidenceFactExact time
主流LLM的预训练采用下一个token预测作为目标，优化的是语言建模损失而非事实准确性
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
VerifiedThe training objective of large language models is to predict the next most likely token, not to judge the truthfulness of information.84% similarVerifiedThe training objective of a Base LLM is to predict the next word, using a paradigm called Autoregressive Language Modeling79% 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78% similarUnverifiedThe 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.78% similarUnverified大语言模型的训练目标是最大化下一个Token的预测概率而非最大化代码安全性，这两个目标在数学上不等价77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/532405API
curl https://kongchang.com/api/v1/knowledge/claims/532405MCP
get_claim(id=532405)