done

Author: tecunningham

_freeze/posts/2026-01-29-knowledge-creating-llms/execute-results/html.json

@@ -285,7 +285,7 @@ <h3 class="no-anchor listing-title">
 </a>
 </div>
 </div>
-<div class="quarto-post image-right" data-index="2" data-listing-date-sort="1770364800000" data-listing-file-modified-sort="1770401198442" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="21" data-listing-word-count-sort="4079">
+<div class="quarto-post image-right" data-index="2" data-listing-date-sort="1770364800000" data-listing-file-modified-sort="1770402601935" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="21" data-listing-word-count-sort="4188">
 <div class="thumbnail"><a href="./posts/2026-01-29-knowledge-creating-llms.html" class="no-external">
 
 <p class="card-img-top"><img src="posts/2026-01-29-knowledge-creating-llms_files/figure-html/unnamed-chunk-1-1.png"  class="thumbnail-image card-img"/></p>

_freeze/posts/2026-01-29-knowledge-creating-llms/figure-html/unnamed-chunk-1-1.pdf

@@ -153,13 +153,14 @@ You have one friend who is full of new ideas, you have another friend who can te
 
 
 <div class="no-row-height column-margin column-container"><div class="">
-<p>Thanks to Zoë Hitzig &amp; Parker Whitfill for helpful comments.</p>
+<p>Thanks to Zoë Hitzig &amp; Parker Whitfill, among others, for helpful comments.</p>
 </div></div><dl>
 <dt>It’s useful to make a distinction between two types of LLMs:</dt>
 <dd>
 <p><strong>Knowledge-sharing LLMs.</strong> Traditionally LLMs have been trained with human judgment as the ground truth, as a consequence they rarely exhibit superhuman performance. Their economic value mainly comes from sharing existing knowledge, and the natural business model is to sell access broadly.</p>
 <p><strong>Knowledge-creating LLMs.</strong> Recently LLMs have been trained against the real world, as a consequence they can extend the limits of human knowledge. The demand for new knowledge is different from the demand for old knowledge, and there’s reason to expect LLM-creators to sell access <em>exclusively</em>.</p>
-<p>Obviously it’s a continuum but we’re clearly setting out on the trajectory from the first to the second, and I haven’t seen much discussion of the implications. Below I give a longer discussion of this distinction, implications for IP, and a galaxy-brain theory that there are only a few dozen deep problems in the world. I’m working on a more formal economic model.</p>
+<p>Obviously it’s a continuum but we’re clearly setting out on the trajectory from the first to the second, and I haven’t seen much discussion of the implications. I expect that the 2026 impacts of AI will be dominated by these considerations.</p>
+<p>Below I give a longer discussion of this distinction, implications for IP, and a galaxy-brain theory that there are only a few dozen deep problems in the world. I sketch some considerations for a more formal economic model.</p>
 </dd>
 </dl>
 <section id="knowledge-sharing-vs-knowledge-creating-llms" class="level1">

_freeze/posts/2026-01-29-knowledge-creating-llms/figure-html/unnamed-chunk-1-1.png

@@ -159,8 +159,8 @@
 <meta name="twitter:title" content="Knowledge-Creating LLMs | Tom Cunningham">
 <meta name="twitter:description" content="Tom Cunningham blog">
 <meta name="twitter:image" content="tecunningham.github.io/posts/2026-01-29-knowledge-creating-llms_files/figure-html/unnamed-chunk-1-1.png">
-<meta name="twitter:image-height" content="904">
-<meta name="twitter:image-width" content="1775">
+<meta name="twitter:image-height" content="1062">
+<meta name="twitter:image-width" content="1731">
 <meta name="twitter:card" content="summary_large_image">
 </head>
 
@@ -266,13 +266,14 @@ <h1 class="title">Knowledge-Creating LLMs</h1>
 
 
 <div class="no-row-height column-margin column-container"><div class="">
-<p>Thanks to Zoë Hitzig &amp; Parker Whitfill for helpful comments.</p>
+<p>Thanks to Zoë Hitzig &amp; Parker Whitfill, among others, for helpful comments.</p>
 </div></div><dl>
 <dt>It’s useful to make a distinction between two types of LLMs:</dt>
 <dd>
 <p><strong>Knowledge-sharing LLMs.</strong> Traditionally LLMs have been trained with human judgment as the ground truth, as a consequence they rarely exhibit superhuman performance. Their economic value mainly comes from sharing existing knowledge, and the natural business model is to sell access broadly.</p>
 <p><strong>Knowledge-creating LLMs.</strong> Recently LLMs have been trained against the real world, as a consequence they can extend the limits of human knowledge. The demand for new knowledge is different from the demand for old knowledge, and there’s reason to expect LLM-creators to sell access <em>exclusively</em>.</p>
-<p>Obviously it’s a continuum but we’re clearly setting out on the trajectory from the first to the second, and I haven’t seen much discussion of the implications. Below I give a longer discussion of this distinction, implications for IP, and a galaxy-brain theory that there are only a few dozen deep problems in the world. I’m working on a more formal economic model.</p>
+<p>Obviously it’s a continuum but we’re clearly setting out on the trajectory from the first to the second, and I haven’t seen much discussion of the implications. I expect that the 2026 impacts of AI will be dominated by these considerations.</p>
+<p>Below I give a longer discussion of this distinction, implications for IP, and a galaxy-brain theory that there are only a few dozen deep problems in the world. I sketch some considerations for a more formal economic model.</p>
 </dd>
 </dl>
 <section id="knowledge-sharing-vs-knowledge-creating-llms" class="level1">