1<!DOCTYPE html> 2<html lang="en-GB"> 3<head> 4 5 <meta charset="utf-8"> 6 <meta name="viewport" content="width=device-width, initial-scale=1"> 7 <meta http-equiv="X-UA-Compatible" content="IE=edge"> 8 <meta name="theme" content="hugo-academic"> 9 <meta name="generator" content="Hugo 0.54.0" /> 10 <meta name="author" content="Johannes B. Gruber"> 11 12 13 14 15 16 17 18 19 20 <meta name="description" content="Large Language Models (LLMs) represent a significant advancement for the field of political communication. Their capabilities for the study of language vastly surpass previous approaches, enabling a wider and more generalized array of applications. The introduction of models like BERT revolutionized political communication research by enhancing the accuracy of text classification tasks. However, it was the release of ChatGPT that captured widespread public attention, establishing LLMs as a cornerstone of artificial intelligence in public perception. 21This entry examines the role of LLMs in political communication research, tracing their development, key concepts, and discussing opportunities and challenges. LLMs can reduce costs while improving accuracy compared to traditional methods of content analysis. Generative LLMs (gLLMs), such as those developed by OpenAI, have pushed the boundaries of text understanding and generation. Besides these advancements, significant concerns remain regarding biases, reproducibility, and ethical considerations. Addressing these challenges is crucial for harnessing the potential of LLMs."> 22 23 24 <link rel="alternate" hreflang="en-GB" href="https://johannesbgruber.eu/publication/2024-09-06-large-language-models/"> 25 26 27 28 29 30 31 32 33 <meta name="theme-color" content="#B05859"> 34 35 <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/animate.css/4.1.1/animate.min.css" /> 36 37 38 39 40 41 42 43 44 <link rel="stylesheet" href="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css"> 45 46 47 48 <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/twitter-bootstrap/3.3.7/css/bootstrap.min.css" integrity="sha512-6MXa8B6uaO18Hid6blRMetEIoPqHf7Ux1tnyIQdpt9qI5OACx7C+O3IVTr98vwGnlcg0LOLa02i9Y1HpVhlfiw==" crossorigin="anonymous"> 49 <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/academicons/1.8.1/css/academicons.min.css" integrity="sha512-NThgw3XKQ1absAahW6to7Ey42uycrVvfNfyjqcFNgCmOCQ5AR4AO0SiXrN+8ZtYeappp56lk1WtvjVmEa+VR6A==" crossorigin="anonymous"> 50 <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.7.0/css/all.min.css" integrity="" crossorigin="anonymous"> 51 <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/fancybox/3.2.5/jquery.fancybox.min.css" integrity="sha256-ygkqlh3CYSUri3LhQxzdcm0n1EQvH2Y+U5S2idbLtxs=" crossorigin="anonymous"> 52 53 54 55 56 <link rel="stylesheet" href="//fonts.googleapis.com/css?family=Oxygen:400,700%7cOxygen:400,400italic,700%7cOxygen+Mono"> 57 58 <link rel="stylesheet" href="https://johannesbgruber.eu/styles.css"> 59 60 61 62 63 <link rel="manifest" href="https://johannesbgruber.eu/site.webmanifest"> 64 <link rel="icon" type="image/png" href="https://johannesbgruber.eu/img/icon.png"> 65 <link rel="apple-touch-icon" type="image/png" href="https://johannesbgruber.eu/img/icon-192.png"> 66 67 <link rel="canonical" href="https://johannesbgruber.eu/publication/2024-09-06-large-language-models/"> 68 69 <meta property="twitter:card" content="summary_large_image"> 70 71 72 <meta property="twitter:image" content="https://johannesbgruber.eu/img/twitter_preview.png"> 73 74 75 <meta property="og:site_name" content="Johannes B. Gruber"> 76 <meta property="og:url" content="https://johannesbgruber.eu/publication/2024-09-06-large-language-models/"> 77 <meta property="og:title" content="Large Language Models | Johannes B. Gruber"> 78 <meta property="og:description" content="Large Language Models (LLMs) represent a significant advancement for the field of political communication. Their capabilities for the study of language vastly surpass previous approaches, enabling a wider and more generalized array of applications. The introduction of models like BERT revolutionized political communication research by enhancing the accuracy of text classification tasks. However, it was the release of ChatGPT that captured widespread public attention, establishing LLMs as a cornerstone of artificial intelligence in public perception. 79This entry examines the role of LLMs in political communication research, tracing their development, key concepts, and discussing opportunities and challenges. LLMs can reduce costs while improving accuracy compared to traditional methods of content analysis. Generative LLMs (gLLMs), such as those developed by OpenAI, have pushed the boundaries of text understanding and generation. Besides these advancements, significant concerns remain regarding biases, reproducibility, and ethical considerations. Addressing these challenges is crucial for harnessing the potential of LLMs."><meta property="og:image" content="https://johannesbgruber.eu/img/large-language-models.png"> 80 <meta property="og:locale" content="en-GB"> 81 82 <meta property="article:published_time" content="2024-09-06T00:00:00+00:00"> 83 84 <meta property="article:modified_time" content="2024-09-06T00:00:00+00:00"> 85 86 87 <title>Large Language Models | Johannes B. Gruber</title> 88 89</head> 90<body id="top" data-spy="scroll" data-target="#toc" data-offset="71" > 91 92<nav class="navbar navbar-default navbar-fixed-top" id="navbar-main"> 93 <div class="container"> 94 95 96 <div class="navbar-header"> 97 98 <button type="button" class="navbar-toggle collapsed" data-toggle="collapse" 99 data-target=".navbar-collapse" aria-expanded="false">
100 <span class="sr-only">Toggle navigation</span> 101 <span class="icon-bar"></span> 102 <span class="icon-bar"></span> 103 <span class="icon-bar"></span> 104 </button> 105 106 107 </div> 108 109 110 <div class="collapse navbar-collapse"> 111 112 113 114 <ul class="nav navbar-nav navbar-left"> 115 116 117 118 119 120 121 122 123 124 125 <li class="nav-item"> 126 <a href="https://johannesbgruber.eu/#about"> 127 128 <span>Home</span> 129 130 </a> 131 </li> 132 133 134 135 136 137 138 139 140 141 142 143 144 <li class="nav-item"> 145 <a href="https://johannesbgruber.eu/#publications"> 146 147 <span>Publications</span> 148 149 </a> 150 </li> 151 152 153 154 155 156 157 158 159 160 161 162 163 <li class="nav-item"> 164 <a href="https://johannesbgruber.eu/#software"> 165 166 <span>Software</span> 167 168 </a> 169 </li> 170 171 172 173 174 175 176 177 178 179 180 <li class="nav-item"> 181 <a href="https://johannesbgruber.eu/#cv"> 182 183 <span>CV</span> 184 185 </a> 186 </li> 187 188 189 190 191 192 193 194 195 196 197 198 199 <li class="nav-item"> 200 <a href="https://johannesbgruber.eu/#posts"> 201 202 <span>Posts</span> 203 204 </a> 205 </li> 206 207 208 209 210 211 212 213 214 215 216 217 218 <li class="nav-item"> 219 <a href="https://johannesbgruber.eu/contact/"> 220 221 <span>Contact</span> 222 223 </a> 224 </li> 225 226 227 228 229 230 231 </ul> 232 233 </div> 234 </div> 235</nav> 236 237<div class="pub" itemscope itemtype="http://schema.org/CreativeWork"> 238 239 <div class="article-container"> 240 241 <h1 itemprop="name">Large Language Models</h1> 242 <span class="pub-authors" itemprop="author"> 243 244 Johannes B. Gruber, Fabio Votta 245 246 </span> 247 <span class="pull-right"> 248 249<div class="share-box" aria-hidden="true"> 250 <ul class="share"> 251 <li> 252 <a class="twitter" 253 href="https://bsky.app/intent/compose?text=Large%20Language%20Models%0Ahttps%3a%2f%2fjohannesbgruber.eu%2fpublication%2f2024-09-06-large-language-models%2f" 254 target="_blank" rel="noopener"> 255 <i class="fa fa-brands fa-bluesky"></i> 256 </a> 257 </li> 258 <li> 259 <a class="facebook" 260 href="https://www.facebook.com/sharer.php?u=https%3a%2f%2fjohannesbgruber.eu%2fpublication%2f2024-09-06-large-language-models%2f" 261 target="_blank" rel="noopener"> 262 <i class="fa fa-brands fa-facebook"></i> 263 </a> 264 </li> 265 <li> 266 <a class="linkedin" 267 href="https://www.linkedin.com/shareArticle?mini=true&url=https%3a%2f%2fjohannesbgruber.eu%2fpublication%2f2024-09-06-large-language-models%2f&title=Large%20Language%20Models" 268 target="_blank" rel="noopener"> 269 <i class="fa fa-brands fa-linkedin"></i> 270 </a> 271 </li> 272 <li> 273 <a class="email" 274 href="mailto:?subject=Large%20Language%20Models&body=https%3a%2f%2fjohannesbgruber.eu%2fpublication%2f2024-09-06-large-language-models%2f"> 275 <i class="fa fa-envelope"></i> 276 </a> 277 </li> 278 </ul> 279</div> 280 281 282 </span> 283 284 285<div class="article-header"> 286 287 288 <img src="https://johannesbgruber.eu/img/large-language-models.png" class="article-banner" itemprop="image"> 289 290 291 292</div> 293 294 295 296 297 298 <h3>Abstract</h3> 299 <p class="pub-abstract" itemprop="text">Large Language Models (LLMs) represent a significant advancement for the field of political communication. Their capabilities for the study of language vastly surpass previous approaches, enabling a wider and more generalized array of applications. The introduction of models like BERT revolutionized political communication research by enhancing the accuracy of text classification tasks. However, it was the release of ChatGPT that captured widespread public attention, establishing LLMs as a cornerstone of artificial intelligence in public perception. 300This entry examines the role of LLMs in political communication research, tracing their development, key concepts, and discussing opportunities and challenges. LLMs can reduce costs while improving accuracy compared to traditional methods of content analysis. Generative LLMs (gLLMs), such as those developed by OpenAI, have pushed the boundaries of text understanding and generation. Besides these advancements, significant concerns remain regarding biases, reproducibility, and ethical considerations. Addressing these challenges is crucial for harnessing the potential of LLMs.</p> 301 302 303 304 <div class="row"> 305 <div class="col-sm-1"></div> 306 <div class="col-sm-10"> 307 <div class="row"> 308 <div class="col-xs-12 col-sm-3 pub-row-heading">Publication</div> 309 <div class="col-xs-12 col-sm-9">Nai, A., Grömping, M., & Wirz, D. (Eds). Elgar Encyclopedia of Political Communication. Edward Elgar Publishing. Accepted ver
309sion</div> 310 </div> 311 </div> 312 <div class="col-sm-1"></div> 313 </div> 314 <div class="visible-xs space-below"></div> 315 316 317 <div class="row"> 318 <div class="col-sm-1"></div> 319 <div class="col-sm-10"> 320 <div class="row"> 321 <div class="col-xs-12 col-sm-3 pub-row-heading">Date</div> 322 <div class="col-xs-12 col-sm-9" itemprop="datePublished"> 323 September, 2024 324 </div> 325 </div> 326 </div> 327 <div class="col-sm-1"></div> 328 </div> 329 <div class="visible-xs space-below"></div> 330 331 <div class="row" style="padding-top: 10px"> 332 <div class="col-sm-1"></div> 333 <div class="col-sm-10"> 334 <div class="row"> 335 <div class="col-xs-12 col-sm-3 pub-row-heading" style="line-height:34px;">Links</div> 336 <div class="col-xs-12 col-sm-9"> 337 338 339 340 341 342 343<a class="btn btn-primary btn-outline" href="https://osf.io/s7qx2/download/" target="_blank" rel="noopener"> 344 PDF 345</a> 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360<a class="btn btn-primary btn-outline" href="https://doi.org/10.31219/osf.io/s7qx2" target="_blank" rel="noopener"> 361 DOI: 10.31219/osf.io/s7qx2 362</a> 363 364 365<a class="btn btn-warning btn-outline btn-xs" href="https://johannesbgruber.eu/publication/2024-09-06-large-language-models/" >status: accepted</a> 366 367 368 369 </div> 370 </div> 371 </div> 372 <div class="col-sm-1"></div> 373 </div> 374 <div class="visible-xs space-below"></div> 375 376 <div class="space-below"></div> 377 378 <div class="article-style"> 379 380 381<p><a href="https://www.favstats.eu/">Fabio</a> and I were asked to write up a short explainer of large language models (LLMs) in Political Communication for the <em>Elgar Encyclopedia of Political Communication</em>. 382While brief, I think this piece contains many important points about the potentials for progress, but also harm, in communication research. 383The encyclopedia will come out in 2025, by which time there will be many new developments in the field of LLMs, I assume. 384But I also believe that the main features in terms of advantages and pitfalls will not dramatically change.</p> 385<p>If you want to cite this (and please do!), please use this bibliographical reference:</p> 386<p>Gruber, J.B. & Votta F. (2025). Large Language Models. In Nai, A., Grömping, M., & Wirz, D. (Eds). Elgar Encyclopedia of Political Communication. Edward Elgar Publishing. Accepted version.</p> 387<pre class="bibtex"><code>@InCollection{gruber2025, 388 author = {Gruber, J.B. and Votta, F.}, 389 title = {Large Language Models}, 390 booktitle = {Elgar Encyclopedia of Political Communication}, 391 editor = {Nai, A. and Grömping, M. and Wirz, D.}, 392 publisher = {Edward Elgar Publishing}, 393 year = {2025}, 394 note = {Accepted version}, 395}</code></pre> 396</div> 397 398 399 400 401 402 403 404 </div> 405</div> 406 407 408 409<footer class="site-footer"> 410 <div class="container"> 411 <p class="powered-by"> 412 413 Licence: <a href='https://creativecommons.org/licenses/by-sa/2.0/'>CC BY-SA</a> · 414 415 Theme based on the 416 <a href="https://sourcethemes.com/academic/" target="_blank" rel="noopener">Academic theme</a> for 417 <a href="https://gohugo.io" target="_blank" rel="noopener">Hugo</a>. 418 Source code on <a href="https://github.com/JBGruber/my-homepage" target="_blank" rel="noopener">GitHub</a>. 419 420 <span class="pull-right" aria-hidden="true"> 421 <a href="#" id="back_to_top"> 422 <span class="button_icon"> 423 <i class="fa fa-chevron-up fa-2x"></i> 424 </span> 425 </a> 426 </span> 427 428 </p> 429 </div> 430</footer> 431 432 433<div id="modal" class="modal fade" role="dialog"> 434 <div class="modal-dialog"> 435 <div class="modal-content"> 436 <div class="modal-header"> 437 <button type="button" class="close btn-large" data-dismiss="modal">×</button> 438 <h4 class="modal-title">Cite</h4> 439 </div> 440 <div> 441 <pre><code class="modal-body tex"></code></pre> 442 </div> 443 <div class="modal-footer"> 444 <a class="btn btn-primary btn-outline js-copy-cite" href="#" target="_blank"> 445 <i class="fa fa-copy"></i> Copy 446 </a> 447 <a class="btn btn-primary btn-outline js-download-cite" href="#" target="_blank"> 448 <i class="fa fa-download"></i> Download 449 </a> 450 <div id="modal-error"></div> 451 </div> 452 </div> 453 </div> 454</div> 455 456 457 458 459 460 461
461<script src="https://cdnjs.cloudflare.com/ajax/libs/jquery/3.2.1/jquery.min.js" integrity="sha512-3P8rXCuGJdNZOnUx/03c1jOTnMn3rP63nBip5gOP2qmUh5YAdVAvFZ1E+QLZZbC1rtMrQb+mah3AfYW11RUrWA==" crossorigin="anonymous"></script>
461 462
462<script src="https://cdnjs.cloudflare.com/ajax/libs/jquery.imagesloaded/4.1.3/imagesloaded.pkgd.min.js" integrity="sha512-umsR78NN0D23AzgoZ11K7raBD+R6hqKojyBZs1w8WvYlsI+QuKRGBx3LFCwhatzBunCjDuJpDHwxD13sLMbpRA==" crossorigin="anonymous"></script>
462 463
463<script src="https://cdnjs.cloudflare.com/ajax/libs/twitter-bootstrap/3.3.7/js/bootstrap.min.js" integrity="sha512-iztkobsvnjKfAtTNdHkGVjAYTrrtlC7mGp/54c40wowO7LhURYl3gVzzcEqGl/qKXQltJ2HwMrdLcNUdo+N/RQ==" crossorigin="anonymous"></script>
463 464
464<script src="https://cdnjs.cloudflare.com/ajax/libs/jquery.isotope/3.0.4/isotope.pkgd.min.js" integrity="sha512-VDBOIlDbuC4VWxGJNmuFRQ0Li0SKkDpmGyuhAG5LTDLd/dJ/S0WMVxriR2Y+CyPL5gzjpN4f/6iqWVBJlht0tQ==" crossorigin="anonymous"></script>
464 465
465<script src="https://cdnjs.cloudflare.com/ajax/libs/fancybox/3.2.5/jquery.fancybox.min.js" integrity="sha256-X5PoE3KU5l+JcX+w09p/wHl9AzK333C4hJ2I9S5mD4M=" crossorigin="anonymous"></script>
465 466 467 468
468<script src="https://johannesbgruber.eu/js/hugo-academic.js"></script>
468 469 470 471 472 473 474 475
475<script src="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js" integrity="sha256-/BfiIkHlHoVihZdc6TFuj7MmJ0TWcWsMXkeDFwhi0zw=" crossorigin="anonymous"></script>
475 476 477 478 479 480 481 482
482<script>hljs.initHighlightingOnLoad();</script>
482 483 484 485 486 487
487<script type="text/x-mathjax-config"> 488 MathJax.Hub.Config({ tex2jax: { inlineMath: [['$','$'], ['\\(','\\)']] } }); 489 </script>
489 490 491
491<script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.1/MathJax.js?config=TeX-AMS_CHTML" integrity="sha512-tOav5w1OjvsSJzePRtt2uQPFwBoHt1VZcUq8l8nm5284LEKE9FSJBQryzMBzHxY5P0zRdNqEcpLIRVYFNgu1jw==" crossorigin="anonymous"></script>
491 492 493 494 495 </body> 496</html> 497
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.