Meta新论文：LLM应以跨视角一致性而非似然度评估真实性
Such a simple yet powerful idea in new Meta paper.
这篇Meta论文提出了一个反直觉但极具启发性的评估思路：用跨视角的一致性替代传统的似然度来衡量LLM输出的真实性。对于做Agent和RAG的同学，这种“多视角投票+拒答”机制是提升可靠性的低成本新思路，值得深入研读并尝试复现。
Such a simple yet powerful idea in new Meta paper.
Meta 最新论文中这样一个简单却强大的理念。
Says that likelihood is the wrong test of truth for LLMs, and that answers which stay the same across contexts deserve more trust.
该论文指出，似然性是衡量大语言模型（LLM）真理性的错误标准，而在不同语境下保持一致的答案更值得信赖。
that LLMs are trained to predict the most likely next token, so they learn what is common in text, not what is true. A falsehood repeated often in training data becomes probable, while rare true events get discounted.
大语言模型被训练用于预测最可能的下一个 token，因此它们学习的是文本中的常见模式，而非真相。在训练数据中频繁出现的谬误会变得看似可能，而罕见但真实的事件则会被低估。
Drawing on physics and philosophy, the paper treats a fact as real if it survives a change of viewpoint. The model answers from several viewpoints, such as 3 medical specialists, keeps what every answer shares, and abstains if nothing is shared.
借鉴物理学和哲学，该论文将事实视为一种在不同视角变换下仍能存续的真实存在。模型从多个视角（例如三位医学专家）进行回答，保留所有答案的共同点，若无共同点则选择弃权。
The paper runs no experiments, so its proof is worked examples. On a single binary example, a loss of 0.67 means 100% accuracy, while a nearly equal loss of 0.71 means 0%.
该论文未进行实验，其论证基于工作示例。在一个二分类示例中，损失值为 0.67 意味着 100% 的准确率，而几乎相等的损失值 0.71 则意味着 0% 的准确率。
Before trusting a confident answer, check whether it survives changes in persona, framing, or language.
在信任一个自信的回答之前，先检查它是否在角色、框架或语言的变换下依然成立。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力