https://neumoneumo.github.io/blog/tags/explainability/Recent content in Explainability on NeumoHugoen-USWed, 10 Dec 2025 17:18:27 +0800https://neumoneumo.github.io/blog/posts/why-llm-detectable/Wed, 10 Dec 2025 17:18:27 +0800https://neumoneumo.github.io/blog/posts/why-llm-detectable/<h1 id="prerequisite">Prerequisite</h1> <ul> <li>Basic knowledge of the structure of Transformer and RNN</li> <li>Understand how to train AI models in NLP</li> </ul> <p>Notations:</p> <table> <thead> <tr> <th>Symbol</th> <th>Meaning</th> </tr> </thead> <tbody> <tr> <td>$x_t$</td> <td>token at time $t$</td> </tr> <tr> <td>$x_{:t}$</td> <td>tokens before time $t$</td> </tr> <tr> <td>$x_{a:b}$</td> <td>tokens after time $a$ (included) and before time $b$ (excluded)</td> </tr> <tr> <td>$p$</td> <td>the distribution of ground truth</td> </tr> <tr> <td>$q$</td> <td>model&rsquo;s prediction</td> </tr> <tr> <td>$v$</td> <td>vocabulary size (# of different tokens)</td> </tr> <tr> <td>$d$</td> <td>embedding dimension (dimension of hidden states)</td> </tr> </tbody> </table> <h1 id="main">Main</h1> <p><strong>This blog will explore some potential factors contributing to the distinction between text generated by LLMs and human&rsquo;s text.</strong></p>