NLP Posts | SAIL Blog

<!DOCTYPE html> <html> <head> <meta charset="utf-8"> <meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=1">  <script defer=true src="/blog/assets/js/main.min.js"></script>  <link rel="stylesheet" href="/blog/assets/css/main.css">  <link rel="icon" type="image/png" href="/blog/assets/img/favicon-32x32.png" sizes="32x32" /> <link rel="icon" type="image/png" href="/blog/assets/img/favicon-16x16.png" sizes="16x16" />  <link rel="canonical" href="http://ai.stanford.edu/blog/nlp/">  <link rel="alternate" type="application/atom+xml" title="The Stanford AI Lab Blog" href="http://ai.stanford.edu/blog/feed.xml"/>   <link rel="stylesheet" type="text/css" href="/blog/assets/css/vendor/font-awesome.min.css">  <link rel="stylesheet" href="https://stackpath.bootstrapcdn.com/bootstrap/4.5.2/css/bootstrap.min.css" integrity="sha384-JcKb8q3iqJ61gNV9KGb8thSsNjpSL0n8PARn9HuZOnIxN0hoP+VmmDGMN5t9UJ0Z" crossorigin="anonymous"> <script src="https://code.jquery.com/jquery-3.5.1.slim.min.js" integrity="sha384-DfXdz2htPH0lsSSs5nCTpuj/zy4C+OGpamoFVy38MVBnE+IbbVYUew+OrCXaRkfj" crossorigin="anonymous"></script> <script src="https://cdn.jsdelivr.net/npm/popper.js@1.16.1/dist/umd/popper.min.js" integrity="sha384-9/reFTGAW83EW2RDu2S0VKaIzap3H66lZH81PoYlFhbGU+6BZp6G7niu735Sk7lN" crossorigin="anonymous"></script> <script src="https://stackpath.bootstrapcdn.com/bootstrap/4.5.2/js/bootstrap.min.js" integrity="sha384-B4gt1jrGC7Jh4AgTPSdUtOBvfO8shuf57BaghqFfPlYxofvL8/KUEfYiJOMMV+rV" crossorigin="anonymous"></script>  <script src="https://code.jquery.com/jquery-2.2.4.min.js" integrity="sha256-BbhdlvQf/xTY9gja0Dq3HiwQF8LaCRTXxZKRutelT44=" crossorigin="anonymous"></script> <link rel="stylesheet" type="text/css" href="/blog/assets/css/vendor/bigfoot-number.css"> <script type="text/javascript" src="/blog/assets/js/vendor/bigfoot.min.js"></script> <script type="text/javascript"> $(document).ready(function() { $.bigfoot(); window.onload = function() { var videos = document.getElementsByTagName("video"), fraction = 0.8; function checkScroll() { for (var i = 0; i < videos.length; i++) { var video = videos[i]; var x = video.offsetLeft, y = video.offsetTop, w = video.offsetWidth, h = video.offsetHeight, r = x + w, //right b = y + h, //bottom visibleX, visibleY, visible; visibleX = Math.max(0, Math.min(w, window.pageXOffset + window.innerWidth - x, r - window.pageXOffset)); visibleY = Math.max(0, Math.min(h, window.pageYOffset + window.innerHeight - y, b - window.pageYOffset)); visible = visibleX * visibleY / (w * h); if (visible > fraction) { video.play(); } else { video.pause(); } } } window.addEventListener('scroll', checkScroll, false); window.addEventListener('resize', checkScroll, false); } }); </script>  <link href="//fonts.googleapis.com/css?family=Source+Sans+Pro:400,700,700italic,400italic" rel="stylesheet" type="text/css">  <link rel="stylesheet" type="text/css" href="/blog/assets/css/vendor/katex.min.css"> <script src="/blog/assets/js/vendor/katex.min.js"> </script>  <script> (function(i, s, o, g, r, a, m) { i['GoogleAnalyticsObject'] = r; i[r] = i[r] || function() { (i[r].q = i[r].q || []).push(arguments) }, i[r].l = 1 * new Date(); a = s.createElement(o), m = s.getElementsByTagName(o)[0]; a.async = 1; a.src = g; m.parentNode.insertBefore(a, m) })(window, document, 'script', '//www.google-analytics.com/analytics.js', 'ga'); ga('create', 'UA-129018108-1', 'auto'); ga('send', 'pageview'); </script>   <title>NLP Posts | SAIL Blog</title> <meta name="generator" content="Jekyll v3.9.0" /> <meta property="og:title" content="NLP Posts" /> <meta property="og:locale" content="en_US" /> <meta name="description" content="The official Stanford AI Lab blog" /> <meta property="og:description" content="The official Stanford AI Lab blog" /> <link rel="canonical" href="http://ai.stanford.edu/blog/nlp/" /> <meta property="og:url" content="http://ai.stanford.edu/blog/nlp/" /> <meta property="og:site_name" content="SAIL Blog" /> <link rel="next" href="http://ai.stanford.edu/blog/nlp/page/2/index.html" /> <meta name="twitter:card" content="summary" /> <meta property="twitter:title" content="NLP Posts" /> <script type="application/ld+json"> {"url":"http://ai.stanford.edu/blog/nlp/","description":"The official Stanford AI Lab blog","@type":"WebPage","headline":"NLP Posts","@context":"https://schema.org"}</script>   <title>NLP Posts | The Stanford AI Lab Blog</title> <meta name="description" content="The Stanford AI Lab (SAIL) Blog is a place for SAIL students, faculty, and researchers to share our work with the general public.">  <meta name="twitter:title" content="NLP Posts"> <meta name="twitter:description" content="The Stanford AI Lab (SAIL) Blog is a place for SAIL students, faculty, and researchers to share our work with the general public."> <meta name="twitter:creator" content="@StanfordAILab"> <meta name="twitter:card" content="summary"> <meta name="twitter:image" content="http://ai.stanford.edu/blog/assets/img/sail-logo.png"> <meta name="og:image" content="http://ai.stanford.edu/blog/assets/img/sail-logo.png"> </head> <body class="category_posts-body">  <header class="site-header"> <nav class="clear navbar navbar-expand-lg navbar-light bg-white flex-column flex-md-row bd-navbar fixed-top" id="main_nav"> <div class="container"> <a class="navbar-brand mr-0 mr-md-2 text-black d-flex align-items-center" href="/blog/" aria-label="Bootstrap"> <div class="branding"> <a href="http://ai.stanford.edu/"> <img class="avatar" src="/blog/assets/img/sail-logo.png" alt=""/> </a> <a href="/blog/"> <h1 class="site-title"> The Stanford AI Lab Blog </h1> </a> </div> <button class="navbar-toggler" type="button" data-toggle="collapse" data-target="#navbarResponsive" aria-controls="navbarResponsive" aria-expanded="false" aria-label="Toggle navigation"> <span class="navbar-toggler-icon"></span> </button> <div class="collapse navbar-collapse" id="navbarResponsive"> <ul class="navbar-nav ml-auto"> <li class="nav-item"> <a class="nav-link" href="/blog/about">About</a> </li> <li class="nav-item dropdown"> <a class="nav-link dropdown-toggle" data-toggle="dropdown" href="#" role="button" aria-haspopup="true" aria-expanded="false">Posts</a> <div class="dropdown-menu"> <a class="dropdown-item" href="/blog/">All</a> <a class="dropdown-item" href="/blog/conferences">Conferences</a> <a class="dropdown-item" href="/blog/vision">Computer Vision</a> <a class="dropdown-item" href="/blog/robotics">Robotics</a> <a class="dropdown-item" href="/blog/nlp">NLP</a> <a class="dropdown-item" href="/blog/ml">Machine Learning</a> <a class="dropdown-item" href="/blog/rl">Reinforcement Learning</a> </div> </li> <li class="nav-item"> <a class="nav-link" href="/blog/subscribe">Subscribe</a> </li> <li class="nav-item"> <a class="nav-link" href="http://ai.stanford.edu/">SAIL</a> </li> </ul> </div> </div> </nav> </header> <div class="content"> <div class="home"> <div class="posts"> <br> <header id="main"> <h1 id="NLP+Posts" class="title"> NLP Posts </h1> <hr> </header> <div class="post-teaser"> <div class="post-img"> <img src="/blog/assets/img/posts/2025-04-28-verbatim-memorization/dall_e_the_library_of_babel.png"> </div> <div class="info"> <header> <a class="post-link" href="/blog/verbatim-memorization/"> <h2> Demystifying Verbatim Memorization in Large Language Models </h2> </a> <p class="meta"> <a href='https://explanare.github.io/'>Jing Huang</a>, <a href='https://cs.stanford.edu/~diyiy/'>Diyi Yang</a>, <a href="https://stanford.edu/~cgpotts/">Christopher Potts</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> How do LLMs memorize long sequences of texts verbatim? In this work, we show that verbatim memorization is intertwined with the LM鈥檚 general capabilities. </div> <div class="excerpt-continue"> <a class="button" href="/blog/verbatim-memorization/"> Continue reading </a> </div> </div> </div> </div> <div class="post-teaser"> <div class="post-img"> <img src="https://ai.stanford.edu/blog/assets/img/posts/2025-04-28-naacl-2025/logo_small.png"> </div> <div class="info"> <header> <a class="post-link" href="/blog/naacl-2025/"> <h2> Stanford AI Lab Papers and Talks at NAACL 2025 </h2> </a> <p class="meta"> Compiled by <a href='https://nityathakkar.github.io/'>Nitya Thakkar</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> All the great work from the Stanford AI Lab accepted at NAACL, all in one place. </div> <div class="excerpt-continue"> <a class="button" href="/blog/naacl-2025/"> Continue reading </a> </div> </div> </div> </div> <div class="post-teaser"> <div class="post-img"> <img src="https://ai.stanford.edu/blog/assets/img/posts/2025-04-01-mentat/image5.png"> </div> <div class="info"> <header> <a class="post-link" href="/blog/mentat/"> <h2> MENTAT: A Clinician-Annotated Benchmark for Complex Psychiatric Decision-Making </h2> </a> <p class="meta"> <a href='https://www.maxlamparth.com'>Max Lamparth</a> and <a href='https://med.stanford.edu/profiles/declan-grabb'>Declan Grabb</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> We developed a new expert design and annotated clinical decision-making dataset that also allows for nuanced accuracy and fairness evaluations with expert preferences, uncertainty, and soft labels. </div> <div class="excerpt-continue"> <a class="button" href="/blog/mentat/"> Continue reading </a> </div> </div> </div> </div> <div class="post-teaser"> <div class="post-img"> <img src="https://ai.stanford.edu/blog/assets/img/posts/2025-02-03-privacylens/privacylens-thumbnail.png"> </div> <div class="info"> <header> <a class="post-link" href="/blog/privacylens/"> <h2> PrivacyLens: Evaluating Privacy Norm Awareness of Language Models in Action </h2> </a> <p class="meta"> <a href='https://cs.stanford.edu/~shaoyj/'>Yijia Shao</a> and <a href='https://cs.stanford.edu/~diyiy/'>Diyi Yang</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> Having an agent handle tasks for you is cool. But does your language model agent respect privacy norms? </div> <div class="excerpt-continue"> <a class="button" href="/blog/privacylens/"> Continue reading </a> </div> </div> </div> </div> <div class="post-teaser"> <div class="post-img"> <img src="https://ai.stanford.edu/blog/assets/img/posts/2025-01-29-teaching/copilot.gif"> </div> <div class="info"> <header> <a class="post-link" href="/blog/teaching/"> <h2> Productive Struggle: The Future of Human Learning in the Age of AI </h2> </a> <p class="meta"> <a href='https://rosewang2008.github.io/'>Rose E. Wang</a> and <a href='https://cs.stanford.edu/~megha'>Megha Srivastava</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> What happens to human learning when superhuman intelligence is as accessible as a Google search? </div> <div class="excerpt-continue"> <a class="button" href="/blog/teaching/"> Continue reading </a> </div> </div> </div> </div> <div class="post-teaser"> <div class="post-img"> <img src="https://ai.stanford.edu/blog/assets/img/posts/2024-11-12-emnlp-2024/logo_small.png"> </div> <div class="info"> <header> <a class="post-link" href="/blog/emnlp-2024/"> <h2> Stanford AI Lab Papers and Talks at EMNLP 2024 </h2> </a> <p class="meta"> Compiled by <a href='https://nityathakkar.github.io/'>Nitya Thakkar</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> All the great work from the Stanford AI Lab accepted at EMNLP 2024, all in one place. </div> <div class="excerpt-continue"> <a class="button" href="/blog/emnlp-2024/"> Continue reading </a> </div> </div> </div> </div> <div class="post-teaser"> <div class="post-img"> <img src="https://ai.stanford.edu/blog/assets/img/posts/2023-07-12-acl-2023/logo_small.png"> </div> <div class="info"> <header> <a class="post-link" href="/blog/acl-2023/"> <h2> Stanford AI Lab Papers and Talks at ACL 2023 </h2> </a> <p class="meta"> Compiled by <a href='https://cs.stanford.edu/~dorarad/'>Drew A. Hudson</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> All the great work from the Stanford AI Lab accepted at ACL 2023, all in one place. </div> <div class="excerpt-continue"> <a class="button" href="/blog/acl-2023/"> Continue reading </a> </div> </div> </div> </div> <div class="post-teaser"> <div class="post-img"> <img src="/blog//assets/img/posts/2022-08-01-understanding-incontext/thumbnail.png"> </div> <div class="info"> <header> <a class="post-link" href="/blog/understanding-incontext/"> <h2> How does in-context learning work? A framework for understanding the differences from traditional supervised learning </h2> </a> <p class="meta"> <a href="https://cs.stanford.edu/~eix/">Sang Michael Xie</a> and <a href="https://shmsw25.github.io/">Sewon Min</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> We provide a Bayesian inference framework for in-context learning in large language models like GPT-3 and show empirical evidence for our framework, including connections to how in-context learning can still work well despite randomizing the labels in few-shot examples. </div> <div class="excerpt-continue"> <a class="button" href="/blog/understanding-incontext/"> Continue reading </a> </div> </div> </div> </div> <div class="post-teaser"> <div class="post-img"> <img src="https://ai.stanford.edu/blog/assets/img/posts/2022-05-31-linkbert/thumbnail.png"> </div> <div class="info"> <header> <a class="post-link" href="/blog/linkbert/"> <h2> LinkBERT: Improving Language Model Training with Document Link </h2> </a> <p class="meta"> <a href="https://michiyasunaga.github.io">Michihiro Yasunaga</a>, <a href="https://cs.stanford.edu/people/jure/">Jure Leskovec</a>, <a href="https://cs.stanford.edu/~pliang/">Percy Liang</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> LinkBERT is a new language model pretrained to capture document link knowledge such as hyperlinks of the web. It greatly helps knowledge-intensive applications such as question answering. </div> <div class="excerpt-continue"> <a class="button" href="/blog/linkbert/"> Continue reading </a> </div> </div> </div> </div> <div class="post-teaser"> <div class="post-img"> <img src="https://ai.stanford.edu/blog/assets/img/posts/2022-05-25-acl-2022/logo_small.png"> </div> <div class="info"> <header> <a class="post-link" href="/blog/acl-2022/"> <h2> Stanford AI Lab Papers and Talks at ACL 2022 </h2> </a> <p class="meta"> Compiled by <a href='https://cs.stanford.edu/~dorarad/'>Drew A. Hudson</a>  </p> </header> <div class="excerpt">  <div class="excerpt-text"> All the great work from the Stanford AI Lab accepted at ACL 2022, all in one place. </div> <div class="excerpt-continue"> <a class="button" href="/blog/acl-2022/"> Continue reading </a> </div> </div> </div> </div> </div> <div class="paginate pager"> <a href="/blog/nlp/page/2/index.html" class="button" > Next <i class="fa fa-chevron-right"></i> </a> </div> </div> </div> <hr> <footer class="site-footer"> <div class="footer-icons"> <ul>  <li> <a href="http://ai.stanford.edu/blog/feed.xml" title="Follow RSS feed"> <span class="fa-stack fa-lg"> <i class="fa fa-circle fa-stack-2x"></i> <i class="fa fa-rss fa-stack-1x fa-inverse"></i> </span> </a> </li> <li> <a href="https://twitter.com/StanfordAILab" title="Follow on Twitter" class="type"> <span class="fa-stack fa-lg"> <i class="fa fa-circle fa-stack-2x"></i> <i class="fa fa-twitter fa-stack-1x fa-inverse"></i> </span> </a> </li> </ul> </div> <p class="text">© 2021 Stanford AI Lab </p> </footer> </body> </html>

CINXE.COM

NLP Posts | SAIL Blog