commit

Author: tecunningham

_freeze/posts/2025-09-19-transformative-AI-notes/execute-results/html.json

@@ -254,7 +254,7 @@ <h3 class="no-anchor listing-title">
 </a>
 </div>
 </div>
-<div class="quarto-post image-right" data-index="1" data-listing-date-sort="1759388400000" data-listing-file-modified-sort="1759858195410" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="43" data-listing-word-count-sort="8413">
+<div class="quarto-post image-right" data-index="1" data-listing-date-sort="1759388400000" data-listing-file-modified-sort="1760028530125" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="43" data-listing-word-count-sort="8418">
 <div class="thumbnail">
 <p><a href="./posts/2025-09-19-transformative-AI-notes.html" class="no-external"></a></p><a href="./posts/2025-09-19-transformative-AI-notes.html" class="no-external">
 </a><p><a href="./posts/2025-09-19-transformative-AI-notes.html" class="no-external"></a></p>

_freeze/posts/2025-10-06-the-curve/execute-results/html.json

@@ -23,8 +23,8 @@
 
 <style>
    dl {display: grid;}
-   //dt {grid-column-start: 1; width: 5cm;}
    dt {grid-column-start: 1; width: 10em;}
+   @media (min-width: 768px) { dt { width: 15em; } }
    dd {grid-column-start: 2; margin-left: 2em;}
 </style>
 

docs/index.html

@@ -257,8 +257,8 @@ <h2 id="toc-title">On this page</h2>
 
 <style>
    dl {display: grid;}
-   //dt {grid-column-start: 1; width: 5cm;}
    dt {grid-column-start: 1; width: 10em;}
+   @media (min-width: 768px) { dt { width: 15em; } }
    dd {grid-column-start: 2; margin-left: 2em;}
 </style>
 
@@ -1179,7 +1179,7 @@ <h1>AI scientists will be unlike human scientists</h1>
 });
 </script>
 </div> <!-- /content -->
-<script>var lightboxQuarto = GLightbox({"openEffect":"zoom","loop":false,"closeEffect":"zoom","descPosition":"bottom","selector":".lightbox"});
+<script>var lightboxQuarto = GLightbox({"selector":".lightbox","openEffect":"zoom","descPosition":"bottom","closeEffect":"zoom","loop":false});
 (function() {
   let previousOnload = window.onload;
   window.onload = () => {

docs/index.xml

@@ -231,43 +231,78 @@ <h2 id="toc-title">On this page</h2>
 <style>
    dl {display: grid;}
    dt {grid-column-start: 1; width: 10em;}
-   @media (min-width: 768px) {
-     dt { width: 15em; }
-   }
+   @media (min-width: 768px) { dt { width: 15em; } }
    dd {grid-column-start: 2; margin-left: 2em;}
 </style>
 
 <div class="no-row-height column-margin column-container"><div class="">
 <p></p>
-</div></div><div class="page-columns page-full"><p> Some topics that kept coming up at the Curve:</p><div class="no-row-height column-margin column-container"><img src="images/2025-10-06-11-35-49.png" class="img-fluid"></div></div>
+</div></div><p class="page-columns page-full"><a href="images/2025-10-06-11-35-49.png" class="lightbox page-columns page-full" data-gallery="quarto-lightbox-gallery-1"><div class="no-row-height column-margin column-container"><img src="images/2025-10-06-11-35-49.png" class="img-fluid"></div></a></p>
+<dl>
+<dt>Some recurrent conversations at the Curve about economic effects of AI.</dt>
+<dd>
+<p>These are the things that I would keep returning to:</p>
 <ol type="1">
 <li>We need more forecasts of economic impacts.</li>
 <li>We need more theory of capabilities.</li>
 <li>We need more metrics of capabilities.</li>
 <li>We need more theory of offense-defense balance.</li>
-<li>There are two ways of modelling future capabilities.</li>
+<li>We need more bottom-up modelling of capabilities growth.</li>
 </ol>
-<p>I feel bad saying “we need,” &amp; scolding others for the work they’re not doing, so I’ve also added my own very tentative best guesses about each.</p>
-<section id="we-need-more-forecasts-of-economic-impacts" class="level1">
+<p>I feel bad saying “we need,” &amp; scolding others for the work they’re not doing, so I’ve tried to add my own very tentative best guesses about each.</p>
+</dd>
+</dl>
+<section id="we-need-more-forecasts-of-economic-impacts" class="level1 page-columns page-full">
 <h1>We need more forecasts of economic impacts</h1>
 <dl>
 <dt>We have few forecasts of the impact of strong AI.</dt>
 <dd>
-Only a few people have written down explicit forecasts of the economic effect of strong AI, i.e.&nbsp;the effects on GDP, employment, wages, asset prices, welfare. The closest I’m aware of is Epoch’s GATE model, see below.
+<p>There are quite a few forecasts of the future trend of AI capabilities, but many fewer explicit forecasts of the economic effect of those capabilities, i.e.&nbsp;the effects on GDP, consumption, employment, wages, asset prices.</p>
+<p>The most explicit forecast I’m aware of is Epoch’s GATE model, see below.</p>
 </dd>
-<dt>I think explicit forecasts would be incredibly useful.</dt>
+<dt>Having explicit forecasts would be very useful.</dt>
 <dd>
 <p>Many informal conversations and arguments often use vague terms, &amp; it’s not clear whether we’re really disagreeing. It would be super useful if I could say “do you think things will be more like the X model, or the Y model?”</p>
-<p>I feel it’s like early 2020 and COVID: if we’re trying to make a decision about whether to announce a lock-down it should be based on a clear idea about the counterfactual, which includes a lot of equilibrium effects (FWIW New Zealand managed to get through COVID so well because they saw what was happening to other countries first).</p>
+<p>I feel it’s like early 2020 and COVID: if we’re trying to make a decision about whether to announce a lock-down it should be based on a clear idea about the counterfactual, which includes a lot of equilibrium effects.</p>
 </dd>
-<dt>Existing forecasts:</dt>
+<dt>Most academic forecasts assume no capabilities growth.</dt>
+<dd>
+I complained about this in <a href="https://tecunningham.github.io/posts/2025-09-19-transformative-AI-notes.html">my previous post</a>: <span class="citation" data-cites="acemoglu2024simple">Acemoglu (<a href="#ref-acemoglu2024simple" role="doc-biblioref">2024</a>)</span> forecasts 0.06% incremental annual growth, <span class="citation" data-cites="aghion2024ai">Aghion and Bunel (<a href="#ref-aghion2024ai" role="doc-biblioref">2024</a>)</span> forecasts 1%. Both fairly clearly rule out significant capabilities growth, they just model diffusion.
+</dd>
+<dt>I know of just two very concrete economics AGI forecasts.</dt>
 <dd>
 <ol type="1">
-<li>Most academic forecasts assume that AI capabilities plateau, i.e.&nbsp;it’s a one-time labor-saving innovation, and the effects are spread out over time due to diffusion: <span class="citation" data-cites="acemoglu2024simple">Acemoglu (<a href="#ref-acemoglu2024simple" role="doc-biblioref">2024</a>)</span> forecasts 0.06% incremental annual growth, <span class="citation" data-cites="aghion2024ai">Aghion and Bunel (<a href="#ref-aghion2024ai" role="doc-biblioref">2024</a>)</span> forecasts 1%.</li>
 <li><span class="citation" data-cites="korinek2024scenarios">Korinek and Suh (<a href="#ref-korinek2024scenarios" role="doc-biblioref">2024</a>)</span> forecast that, over 15 years, GDP triples; wages increase a little at first, then collapse when everything is automated. GDP continues to increase.</li>
-<li>Epoch’s GATE model (<span class="citation" data-cites="erdil2025gate">Erdil et al. (<a href="#ref-erdil2025gate" role="doc-biblioref">2025</a>)</span>): they forecast full automation in 2034, by which point GWP has grown 10X. They forecast wages will increase dramatically then collapse somewhat after full automation is achieved.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></li>
+<li>Epoch’s GATE model (<span class="citation" data-cites="erdil2025gate">Erdil et al. (<a href="#ref-erdil2025gate" role="doc-biblioref">2025</a>)</span>) forecast full automation in 2034, by which point gross world product (GWP) has grown 10X. They forecast that wages will at first increase dramatically then, at some point after full automation is achieved, collapse.<a href="#fn1" class="footnote-ref" id="fnref1" role="doc-noteref"><sup>1</sup></a></li>
 </ol>
-<p>We also have forecasting markets and individual forecasts but these are hard to interpret: they are a combination of capabilities progress and the effects of those capabilities. Finally there are financial forecasts, e.g.&nbsp;those used to plan AI and datacenter investments, which implicitly contains models of the economic evolution. My sense is these are often fairly crude extrapolations of existing trends.</p>
+</dd>
+</dl>
+<p>&nbsp;</p>
+
+<div class="no-row-height column-margin column-container"><div class="quarto-figure quarto-figure-center">
+<figure class="figure">
+<p><a href="images/2025-10-09-10-35-37.png" class="lightbox" data-gallery="quarto-lightbox-gallery-2" title="Metaculus: What will be the prime-age (25-54) labor force participation rate in the United States in these years?"><img src="images/2025-10-09-10-35-37.png" class="img-fluid figure-img" alt="Metaculus: What will be the prime-age (25-54) labor force participation rate in the United States in these years?"></a></p>
+<figcaption>Metaculus: What will be the prime-age (25-54) labor force participation rate in the United States in these years?</figcaption>
+</figure>
+</div></div><dl>
+<dt>Forecasting markets expect big capabilities, small impacts.</dt>
+<dd>
+<p>Forecating markets expect rapid progress in AI capabilities:</p>
+<ul>
+<li>Weak AGI in 2027 (<a href="https://www.metaculus.com/questions/3479/">Metaculus</a>)</li>
+<li>Strong AGI in 2033 (<a href="https://www.metaculus.com/questions/5121/date-of-artificial-general-intelligence/">Metaculus</a>)</li>
+<li>Transformative AI in 2041 (<a href="https://www.metaculus.com/questions/19356/transformative-ai-date/">Metaculus</a>)</li>
+</ul>
+<p>Forecasting markets also expect economic variables to remain relatively flat:</p>
+<ul>
+<li>World productivity growth increases from 3% to 5% over the next 100 years (<a href="https://www.metaculus.com/questions/12663/cagr-gdp-growth-per-capita/">Metaculus</a>)</li>
+<li>Labor force participation falls smoothly from 84% to 36% over the next 100 years (<a href="https://www.metaculus.com/questions/18289/">Metaculus</a>).</li>
+</ul>
+<p></p>
+</dd>
+<dt>What do the financial markets think?</dt>
+<dd>
+<p>Finally there are financial forecasts, e.g.&nbsp;those used to plan AI and datacenter investments, which implicitly contains models of the economic evolution. My sense is these are often fairly crude extrapolations of existing trends.</p>
 <p>Am I missing others?</p>
 </dd>
 </dl>
@@ -324,7 +359,7 @@ <h1>We need more metrics of capabilities</h1>
 
 <div class="no-row-height column-margin column-container"><div class="quarto-figure quarto-figure-center">
 <figure class="figure">
-<p><a href="images/2025-09-19-16-49-40.png" class="lightbox" data-gallery="quarto-lightbox-gallery-1" title="Our World in Data: Technology Costs over Time."><img src="images/2025-09-19-16-49-40.png" class="img-fluid figure-img" alt="Our World in Data: Technology Costs over Time."></a></p>
+<p><a href="images/2025-09-19-16-49-40.png" class="lightbox" data-gallery="quarto-lightbox-gallery-3" title="Our World in Data: Technology Costs over Time."><img src="images/2025-09-19-16-49-40.png" class="img-fluid figure-img" alt="Our World in Data: Technology Costs over Time."></a></p>
 <figcaption>Our World in Data: Technology Costs over Time.</figcaption>
 </figure>
 </div></div><dl>
@@ -341,7 +376,7 @@ <h1>We need more metrics of capabilities</h1>
 </section>
 <section id="we-need-more-theory-of-the-offense-defense-balance" class="level1 page-columns page-full">
 <h1>We need more theory of the offense-defense balance</h1>
-<p class="page-columns page-full"><a href="images/2025-10-06-11-33-10.png" class="lightbox page-columns page-full" data-gallery="quarto-lightbox-gallery-2"><div class="no-row-height column-margin column-container"><img src="images/2025-10-06-11-33-10.png" class="img-fluid"></div></a></p>
+<p class="page-columns page-full"><a href="images/2025-10-06-11-33-10.png" class="lightbox page-columns page-full" data-gallery="quarto-lightbox-gallery-4"><div class="no-row-height column-margin column-container"><img src="images/2025-10-06-11-33-10.png" class="img-fluid"></div></a></p>
 <dl>
 <dt>Many discussions were about how AI will change the offense-defense balance</dt>
 <dd>
@@ -364,7 +399,7 @@ <h1>We need more theory of the offense-defense balance</h1>
 The
 </dd>
 </dl>
-<p class="page-columns page-full"><a href="images/2025-10-06-11-30-58.png" class="lightbox page-columns page-full" data-gallery="quarto-lightbox-gallery-3"><div class="no-row-height column-margin column-container"><img src="images/2025-10-06-11-30-58.png" class="img-fluid"></div></a></p>
+<p class="page-columns page-full"><a href="images/2025-10-06-11-30-58.png" class="lightbox page-columns page-full" data-gallery="quarto-lightbox-gallery-5"><div class="no-row-height column-margin column-container"><img src="images/2025-10-06-11-30-58.png" class="img-fluid"></div></a></p>
 </section>
 <section id="there-are-two-ways-of-modelling-ais-impact" class="level1">
 <h1>There are two ways of modelling AI’s impact</h1>
@@ -902,7 +937,7 @@ <h1>There are two ways of modelling AI’s impact</h1>
 });
 </script>
 </div> <!-- /content -->
-<script>var lightboxQuarto = GLightbox({"closeEffect":"zoom","openEffect":"zoom","selector":".lightbox","loop":false,"descPosition":"bottom"});
+<script>var lightboxQuarto = GLightbox({"selector":".lightbox","openEffect":"zoom","loop":false,"closeEffect":"zoom","descPosition":"bottom"});
 (function() {
   let previousOnload = window.onload;
   window.onload = () => {

docs/posts/2025-09-19-transformative-AI-notes.html

@@ -50,7 +50,7 @@
   </url>
   <url>
     <loc>tecunningham.github.io/posts/2025-09-19-transformative-AI-notes.html</loc>
-    <lastmod>2025-10-07T17:29:55.410Z</lastmod>
+    <lastmod>2025-10-09T16:48:50.125Z</lastmod>
   </url>
   <url>
     <loc>tecunningham.github.io/posts/2023-01-31-social-media-suspensions-data.html</loc>

docs/posts/2025-10-06-the-curve.html

@@ -19,8 +19,8 @@ format:
 
 <style>
    dl {display: grid;}
-   //dt {grid-column-start: 1; width: 5cm;}
    dt {grid-column-start: 1; width: 10em;}
+   @media (min-width: 768px) { dt { width: 15em; } }
    dd {grid-column-start: 2; margin-left: 2em;}
 </style>