Unverified50% confidenceFactExact time
当前主流 LLM 采用指令遵循（Instruction Following）训练方式，会用概率最高的方式补全输出
1
Sources
50%
Confidence
Long-term
Relevance
7/21/2026
First Seen
Sources
Related Claims
UnverifiedLLM的输出结果天然倾向于训练集中出现频率最高的模式组合，即统计意义上的最大公约数80% similarUnverified当前主流LLM普遍采用预训练加微调范式，先在互联网规模语料上无监督预训练，再通过RLHF等技术对齐人类偏好77% similarUnverifiedLLM 在 RLHF 训练中被强化了「尽快给出有用输出」的行为倾向，默认帮用户做事优先于和用户对齐认知75% similarUnverified现代LLM通常经过专门的微调来学习何时、如何调用工具，训练数据中包含大量工具使用的示例对话72% similarUnverifiedDuring the pre-training phase, LLMs optimize token prediction probability through maximum likelihood estimation, lacking an inherent mechanism to 'know what they don't know,' and there is no necessary correlation between confidence and accuracy.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/577476API
curl https://kongchang.com/api/v1/knowledge/claims/577476MCP
get_claim(id=577476)