Built site for gh-pages

.nojekyll

@@ -1 +1 @@
-1a0e5151
+a89614ad

index.html

@@ -207,7 +207,7 @@ <h2 class="anchored" data-anchor-id="articles">Articles</h2>
 
 <div class="quarto-listing quarto-listing-container-grid" id="listing-listing">
 <div class="list grid quarto-listing-cols-3">
-<div class="g-col-1" data-index="0" data-listing-date-sort="1728345600000" data-listing-file-modified-sort="1728966007195" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="3" data-listing-word-count-sort="577">
+<div class="g-col-1" data-index="0" data-listing-date-sort="1728864000000" data-listing-file-modified-sort="1728966041433" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="3" data-listing-word-count-sort="577">
 <a href="./posts/sae_dream.html" class="quarto-grid-link">
 <div class="quarto-grid-item card h-100 card-left">
 <p class="card-img-top"><img src="posts/sae_dream_files/figure-html/cell-9-output-1.png" style="height: 150px;"  class="thumbnail-image card-img"/></p>
@@ -223,14 +223,14 @@ <h5 class="no-anchor card-title listing-title">
 T. Ben Thompson
 </div>
 <div class="listing-date">
-Oct 8, 2024
+Oct 14, 2024
 </div>
 </div>
 </div>
 </div>
 </a>
 </div>
-<div class="g-col-1" data-index="1" data-listing-date-sort="1721692800000" data-listing-file-modified-sort="1728966007191" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="5" data-listing-word-count-sort="806">
+<div class="g-col-1" data-index="1" data-listing-date-sort="1721692800000" data-listing-file-modified-sort="1728966041425" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="5" data-listing-word-count-sort="806">
 <a href="./posts/flrt.html" class="quarto-grid-link">
 <div class="quarto-grid-item card h-100 card-left">
 <p class="card-img-top"><img src="posts/bomb.png" style="height: 150px;"  class="thumbnail-image card-img"/></p>
@@ -253,7 +253,7 @@ <h5 class="no-anchor card-title listing-title">
 </div>
 </a>
 </div>
-<div class="g-col-1" data-index="2" data-listing-date-sort="1720742400000" data-listing-file-modified-sort="1728966007179" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="9" data-listing-word-count-sort="1747">
+<div class="g-col-1" data-index="2" data-listing-date-sort="1720742400000" data-listing-file-modified-sort="1728966041413" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="9" data-listing-word-count-sort="1747">
 <a href="./posts/circuit_breaking.html" class="quarto-grid-link">
 <div class="quarto-grid-item card h-100 card-left">
 <p class="card-img-top"><img src="posts/circuit_breaking_files/figure-html/cell-4-output-1.png" style="height: 150px;"  class="thumbnail-image card-img"/></p>
@@ -276,7 +276,7 @@ <h5 class="no-anchor card-title listing-title">
 </div>
 </a>
 </div>
-<div class="g-col-1" data-index="3" data-listing-date-sort="1705968000000" data-listing-file-modified-sort="1728966007191" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="4" data-listing-word-count-sort="771">
+<div class="g-col-1" data-index="3" data-listing-date-sort="1705968000000" data-listing-file-modified-sort="1728966041425" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="4" data-listing-word-count-sort="771">
 <a href="./posts/dreamy.html" class="quarto-grid-link">
 <div class="quarto-grid-item card h-100 card-left">
 <p class="card-img-top"><img src="posts/dream_wow.png" style="height: 150px;"  class="thumbnail-image card-img"/></p>
@@ -299,7 +299,7 @@ <h5 class="no-anchor card-title listing-title">
 </div>
 </a>
 </div>
-<div class="g-col-1" data-index="4" data-listing-date-sort="1705104000000" data-listing-file-modified-sort="1728966007155" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="25" data-listing-word-count-sort="4985">
+<div class="g-col-1" data-index="4" data-listing-date-sort="1705104000000" data-listing-file-modified-sort="1728966041389" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="25" data-listing-word-count-sort="4985">
 <a href="./posts/TDC2023.html" class="quarto-grid-link">
 <div class="quarto-grid-item card h-100 card-left">
 <p class="card-img-top"><img src="posts/TDC2023-sample-instances.png" alt="The Z-scores of activation vector similarity for the provided sample instances" style="height: 150px;"  class="thumbnail-image card-img"/></p>
@@ -322,7 +322,7 @@ <h5 class="no-anchor card-title listing-title">
 </div>
 </a>
 </div>
-<div class="g-col-1" data-index="5" data-listing-date-sort="1701302400000" data-listing-file-modified-sort="1728966007191" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="7" data-listing-word-count-sort="1303">
+<div class="g-col-1" data-index="5" data-listing-date-sort="1701302400000" data-listing-file-modified-sort="1728966041425" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="7" data-listing-word-count-sort="1303">
 <a href="./posts/fight_the_illusion.html" class="quarto-grid-link">
 <div class="quarto-grid-item card h-100 card-left">
 <div class="listing-item-img-placeholder card-img-top" style="height: 150px;">&nbsp;</div>
@@ -345,7 +345,7 @@ <h5 class="no-anchor card-title listing-title">
 </div>
 </a>
 </div>
-<div class="g-col-1" data-index="6" data-listing-date-sort="1687651200000" data-listing-file-modified-sort="1728966007179" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="7" data-listing-word-count-sort="1247">
+<div class="g-col-1" data-index="6" data-listing-date-sort="1687651200000" data-listing-file-modified-sort="1728966041413" data-listing-date-modified-sort="NaN" data-listing-reading-time-sort="7" data-listing-word-count-sort="1247">
 <a href="./posts/catalog.html" class="quarto-grid-link">
 <div class="quarto-grid-item card h-100 card-left">
 <p class="card-img-top"><img src="posts/catalog_files/figure-html/cell-9-output-1.png" style="height: 150px;"  class="thumbnail-image card-img"/></p>

posts/TDC2023.html

@@ -959,7 +959,7 @@ <h4 class="anchored" data-anchor-id="trojan-recovery">Trojan recovery:</h4>
 });
 </script>
 </div> <!-- /content -->
-<script>var lightboxQuarto = GLightbox({"closeEffect":"zoom","selector":".lightbox","descPosition":"bottom","loop":false,"openEffect":"zoom"});
+<script>var lightboxQuarto = GLightbox({"descPosition":"bottom","closeEffect":"zoom","openEffect":"zoom","loop":false,"selector":".lightbox"});
 (function() {
   let previousOnload = window.onload;
   window.onload = () => {

posts/catalog.html

@@ -1058,7 +1058,7 @@ <h2 class="anchored" data-anchor-id="github">GitHub</h2>
 });
 </script>
 </div> <!-- /content -->
-<script>var lightboxQuarto = GLightbox({"selector":".lightbox","closeEffect":"zoom","loop":false,"descPosition":"bottom","openEffect":"zoom"});
+<script>var lightboxQuarto = GLightbox({"selector":".lightbox","closeEffect":"zoom","descPosition":"bottom","openEffect":"zoom","loop":false});
 (function() {
   let previousOnload = window.onload;
   window.onload = () => {

posts/catalog.out.ipynb

@@ -297,7 +297,7 @@
         "Pythia-12B is miscalibrated on 20% of the bigrams and 45% of the\n",
         "trigrams when we ask for prediction of $p \\geq 0.45$."
       ],
-      "id": "513a9812-5147-4017-a9f5-be0a3154696d"
+      "id": "9f8bbc92-cd35-4bbe-b51a-dc733216f243"
     },
     {
       "cell_type": "code",
@@ -377,7 +377,7 @@
         "The dataset is available on Huggingface:\n",
         "[pile_scan_4](https://huggingface.co/datasets/Confirm-Labs/pile_scan_4)"
       ],
-      "id": "ff751c60-69aa-4c7a-875e-dc110b74317f"
+      "id": "ab6c1240-7e9c-40f0-bfc9-132ffdfa1232"
     },
     {
       "cell_type": "code",
@@ -417,7 +417,7 @@
         "Computational Linguistics, May 2022, pp. 95–136. doi:\n",
         "[10.18653/v1/2022.bigscience-1.9](https://doi.org/10.18653/v1/2022.bigscience-1.9).</span>"
       ],
-      "id": "971b97d7-7f5a-4d12-8f2a-7f894a0c831e"
+      "id": "2c698549-a89e-4abf-ac5d-2988217bf06b"
     }
   ],
   "nbformat": 4,

posts/circuit_breaking.html

@@ -1739,7 +1739,7 @@ <h2 class="anchored" data-anchor-id="attack-success-fluency">Attack success: flu
 });
 </script>
 </div> <!-- /content -->
-<script>var lightboxQuarto = GLightbox({"openEffect":"zoom","loop":false,"selector":".lightbox","closeEffect":"zoom","descPosition":"bottom"});
+<script>var lightboxQuarto = GLightbox({"loop":false,"descPosition":"bottom","selector":".lightbox","openEffect":"zoom","closeEffect":"zoom"});
 (function() {
   let previousOnload = window.onload;
   window.onload = () => {

posts/dreamy.html

@@ -5420,7 +5420,7 @@ <h2 class="anchored" data-anchor-id="causal-token-attribution">Causal token attr
 });
 </script>
 </div> <!-- /content -->
-<script>var lightboxQuarto = GLightbox({"descPosition":"bottom","openEffect":"zoom","selector":".lightbox","closeEffect":"zoom","loop":false});
+<script>var lightboxQuarto = GLightbox({"loop":false,"selector":".lightbox","closeEffect":"zoom","descPosition":"bottom","openEffect":"zoom"});
 (function() {
   let previousOnload = window.onload;
   window.onload = () => {

posts/flrt.html

@@ -1362,7 +1362,7 @@ <h2 class="anchored" data-anchor-id="jailbreak">Jailbreak</h2>
 });
 </script>
 </div> <!-- /content -->
-<script>var lightboxQuarto = GLightbox({"closeEffect":"zoom","selector":".lightbox","descPosition":"bottom","openEffect":"zoom","loop":false});
+<script>var lightboxQuarto = GLightbox({"closeEffect":"zoom","openEffect":"zoom","selector":".lightbox","loop":false,"descPosition":"bottom"});
 (function() {
   let previousOnload = window.onload;
   window.onload = () => {

posts/sae_dream.html

@@ -7,7 +7,7 @@
 <meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
 
 <meta name="author" content="T. Ben Thompson">
-<meta name="dcterms.date" content="2024-10-08">
+<meta name="dcterms.date" content="2024-10-14">
 <meta name="description" content="Better token optimization to maximize SAE features.">
 
 <title>Dreaming with sparse autoencoder features – Confirm</title>
@@ -144,10 +144,10 @@
 <link rel="stylesheet" href="../styles.css">
 <meta name="citation_title" content="Dreaming with sparse autoencoder features">
 <meta name="citation_author" content="T. Ben Thompson">
-<meta name="citation_publication_date" content="2024-10-08">
-<meta name="citation_cover_date" content="2024-10-08">
+<meta name="citation_publication_date" content="2024-10-14">
+<meta name="citation_cover_date" content="2024-10-14">
 <meta name="citation_year" content="2024">
-<meta name="citation_online_date" content="2024-10-08">
+<meta name="citation_online_date" content="2024-10-14">
 <meta name="citation_fulltext_html_url" content="https://confirmlabs.org/posts/sae_dream.html">
 <meta name="citation_language" content="en">
 <meta name="citation_reference" content="citation_title=GPT-NeoX-20B: An open-source autoregressive language model;,citation_abstract=We introduce GPT-NeoX-20B, a 20 billion parameter autoregressive language model trained on the Pile, whose weights will be made freely and openly available to the public through a permissive license. It is, to the best of our knowledge, the largest dense autoregressive model that has publicly available weights at the time of submission. In this work, we describe GPT-NeoX-20B’s architecture and training, and evaluate its performance. We open-source the training and evaluation code, as well as the model weights, at https://github.com/EleutherAI/gpt-neox.;,citation_author=Sidney Black;,citation_author=Stella Biderman;,citation_author=Eric Hallahan;,citation_author=Quentin Anthony;,citation_author=Leo Gao;,citation_author=Laurence Golding;,citation_author=Horace He;,citation_author=Connor Leahy;,citation_author=Kyle McDonell;,citation_author=Jason Phang;,citation_author=Michael Pieler;,citation_author=Usvsn Sai Prashanth;,citation_author=Shivanshu Purohit;,citation_author=Laria Reynolds;,citation_author=Jonathan Tow;,citation_author=Ben Wang;,citation_author=Samuel Weinbach;,citation_publication_date=2022-05;,citation_cover_date=2022-05;,citation_year=2022;,citation_fulltext_html_url=https://aclanthology.org/2022.bigscience-1.9;,citation_doi=10.18653/v1/2022.bigscience-1.9;,citation_conference_title=Proceedings of BigScience episode #5 – workshop on challenges &amp;amp;amp; perspectives in creating large language models;,citation_conference=Association for Computational Linguistics;">
@@ -254,7 +254,7 @@ <h1 class="title">Dreaming with sparse autoencoder features</h1>
     <div>
     <div class="quarto-title-meta-heading">Published</div>
     <div class="quarto-title-meta-contents">
-      <p class="date">October 8, 2024</p>
+      <p class="date">October 14, 2024</p>
     </div>
   </div>
 
@@ -858,14 +858,14 @@ <h2 class="anchored" data-anchor-id="dreaming-decoder-directions">Dreaming decod
 </section><section class="quarto-appendix-contents" id="quarto-citation"><h2 class="anchored quarto-appendix-heading">Citation</h2><div><div class="quarto-appendix-secondary-label">BibTeX citation:</div><pre class="sourceCode code-with-copy quarto-appendix-bibtex"><code class="sourceCode bibtex">@online{thompson2024,
   author = {Thompson, T. Ben},
   title = {Dreaming with Sparse Autoencoder Features},
-  date = {2024-10-08},
+  date = {2024-10-14},
   url = {https://confirmlabs.org/posts/sae_dream.html},
   langid = {en}
 }
 </code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre><div class="quarto-appendix-secondary-label">For attribution, please cite this work as:</div><div id="ref-thompson2024" class="csl-entry quarto-appendix-citeas" role="listitem">
 <div class="">T.
 B. Thompson, <span>“Dreaming with sparse autoencoder features,”</span>
-Oct. 08, 2024. <a href="https://confirmlabs.org/posts/sae_dream.html">https://confirmlabs.org/posts/sae_dream.html</a></div>
+Oct. 14, 2024. <a href="https://confirmlabs.org/posts/sae_dream.html">https://confirmlabs.org/posts/sae_dream.html</a></div>
 </div></div></section></div></main> <!-- /main -->
 <script id="quarto-html-after-body" type="application/javascript">
 window.document.addEventListener("DOMContentLoaded", function (event) {
@@ -1288,7 +1288,7 @@ <h2 class="anchored" data-anchor-id="dreaming-decoder-directions">Dreaming decod
 });
 </script>
 </div> <!-- /content -->
-<script>var lightboxQuarto = GLightbox({"descPosition":"bottom","selector":".lightbox","loop":false,"closeEffect":"zoom","openEffect":"zoom"});
+<script>var lightboxQuarto = GLightbox({"loop":false,"openEffect":"zoom","descPosition":"bottom","selector":".lightbox","closeEffect":"zoom"});
 (function() {
   let previousOnload = window.onload;
   window.onload = () => {

posts/sae_dream.out.ipynb

@@ -7,7 +7,7 @@
         "# Dreaming with sparse autoencoder features\n",
         "\n",
         "T. Ben Thompson  \n",
-        "2024-10-08\n",
+        "2024-10-14\n",
         "\n",
         "I get a lot of questions about dreaming/feature visualization applied to\n",
         "sparse autoencoders (SAEs) [1]. When we were writing [Fluent dreaming\n",
@@ -54,7 +54,7 @@
         "written just for this post. I expect inefficiences and possibly some\n",
         "bugs."
       ],
-      "id": "dfcbb8be-8a40-46a3-b0c9-a4019f5484a0"
+      "id": "0bb6f36e-91b2-48ea-a4a5-ba9eb037e332"
     },
     {
       "cell_type": "code",
@@ -389,7 +389,7 @@
         "\n",
         "First, we’ll load up Gemma 2 2B."
       ],
-      "id": "7663a80e-cd34-45c1-a6db-11a5d1bb8996"
+      "id": "9fc1dba4-9a3c-4895-a7e5-a7d65de8cb1c"
     },
     {
       "cell_type": "code",
@@ -434,7 +434,7 @@
         "function will be passed as the `get_feature_and_logits` argument to the\n",
         "`dream` function."
       ],
-      "id": "b8dd48f6-8309-44b0-a864-ddb1cd7150ba"
+      "id": "3fa85ad5-44c8-4a7c-94cc-c803c5b49d97"
     },
     {
       "cell_type": "code",
@@ -484,7 +484,7 @@
         "seems to respond strongly to variants of the word “label”. We run a\n",
         "bunch of examples through the SAE and see how they score:"
       ],
-      "id": "d6597a44-bae8-4df2-974d-007c8ca7b395"
+      "id": "4e4c59b4-fcc9-45f2-b27d-987b9b1abb19"
     },
     {
       "cell_type": "code",
@@ -545,7 +545,7 @@
         "original notebook to see\n",
         "it.](https://github.com/Confirm-Solutions/confirmlabs/blob/main/posts/sae_dream.ipynb)"
       ],
-      "id": "1d912b1b-9a65-48d4-826b-b1fb0e289afb"
+      "id": "a435c183-9bd3-42cb-b4ba-ed57bc14bc10"
     },
     {
       "cell_type": "code",
@@ -614,7 +614,7 @@
         "Out of curiosity, do we get similar results if we apply feature\n",
         "visualization to the corresponding decoder direction?"
       ],
-      "id": "d2795cca-f26b-4369-886e-98cd881ae795"
+      "id": "76315e59-1833-401d-bfef-5e3335c9d3e5"
     },
     {
       "cell_type": "code",
@@ -662,7 +662,7 @@
         "yes for this particular feature. However, it might be different for\n",
         "other features."
       ],
-      "id": "06cda7b0-ceb2-410e-a51b-be71e5be97a1"
+      "id": "ca7cf061-03b2-4a6b-a8d0-ec11f21a6da3"
     },
     {
       "cell_type": "code",
@@ -746,7 +746,7 @@
         "Feature visualization on the decoder direction also acquires the “label”\n",
         "phrase:"
       ],
-      "id": "986abf0c-319a-4906-b220-c96d98509438"
+      "id": "c65fabe5-f3ae-4306-8702-e05264f710e6"
     },
     {
       "cell_type": "code",
@@ -785,7 +785,7 @@
         "“FLRT: Fluent student-teacher redteaming.” 2024. Available:\n",
         "<https://arxiv.org/abs/2407.17447></span>"
       ],
-      "id": "44a51ce2-0f90-446e-af76-a37a5f347150"
+      "id": "f28edba4-c39c-4d18-8eb2-06de8b30a3fd"
     }
   ],
   "nbformat": 4,

sitemap.xml

@@ -2,34 +2,34 @@
 <urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
   <url>
     <loc>https://confirmlabs.org/posts/dreamy.html</loc>
-    <lastmod>2024-10-15T04:20:07.191Z</lastmod>
+    <lastmod>2024-10-15T04:20:41.425Z</lastmod>
   </url>
   <url>
     <loc>https://confirmlabs.org/posts/circuit_breaking.html</loc>
-    <lastmod>2024-10-15T04:20:07.179Z</lastmod>
+    <lastmod>2024-10-15T04:20:41.413Z</lastmod>
   </url>
   <url>
     <loc>https://confirmlabs.org/posts/TDC2023.html</loc>
-    <lastmod>2024-10-15T04:20:07.155Z</lastmod>
+    <lastmod>2024-10-15T04:20:41.389Z</lastmod>
   </url>
   <url>
     <loc>https://confirmlabs.org/posts/fight_the_illusion.html</loc>
-    <lastmod>2024-10-15T04:20:07.191Z</lastmod>
+    <lastmod>2024-10-15T04:20:41.425Z</lastmod>
   </url>
   <url>
     <loc>https://confirmlabs.org/index.html</loc>
-    <lastmod>2024-10-15T04:20:07.143Z</lastmod>
+    <lastmod>2024-10-15T04:20:41.381Z</lastmod>
   </url>
   <url>
     <loc>https://confirmlabs.org/posts/catalog.html</loc>
-    <lastmod>2024-10-15T04:20:07.179Z</lastmod>
+    <lastmod>2024-10-15T04:20:41.413Z</lastmod>
   </url>
   <url>
     <loc>https://confirmlabs.org/posts/flrt.html</loc>
-    <lastmod>2024-10-15T04:20:07.191Z</lastmod>
+    <lastmod>2024-10-15T04:20:41.425Z</lastmod>
   </url>
   <url>
     <loc>https://confirmlabs.org/posts/sae_dream.html</loc>
-    <lastmod>2024-10-15T04:20:07.195Z</lastmod>
+    <lastmod>2024-10-15T04:20:41.433Z</lastmod>
   </url>
 </urlset>