Text: line counts of the pinned commit; link ELIZA and SmolTalkBench

site/index.html

@@ -18,6 +18,7 @@
   .sub { font:400 clamp(18px,2.4vw,24px)/1.3 var(--lead); color:var(--mute); margin:0 0 28px; }
   .text { max-width:680px; }
   .text p { margin:0 0 16px; }
+  .text a { color:inherit; text-decoration:underline; text-underline-offset:3px; }
   code { font:16px var(--mono); }
   .app { margin-top:32px; border:1px solid var(--ink); }
   #log { height:220px; overflow-y:auto; display:flex; flex-direction:column; gap:10px; padding:14px; }
@@ -48,13 +49,13 @@
   <h1>PyBot</h1>
   <p class="sub">a conversational bot without LLMs</p>
   <div class="text">
-    <p>LLMs drive today's chatbots and agents, but conversational programs are much older than LLMs. Sixty years ago, ELIZA held
+    <p>LLMs drive today's chatbots and agents, but conversational programs are much older than LLMs. Sixty years ago, <a href="https://en.wikipedia.org/wiki/ELIZA" target="_blank" rel="noopener">ELIZA</a> held
     conversations by matching patterns in what the user typed and answering with predefined templates.</p>
     <p>We wanted to revisit that idea at a much larger scale, so we gave a swarm of LLM agents the mission of building a chatbot out of
     plain Python rules (if-else branches, regex patterns and lookup tables), with no model inside. The agents wrote and evaluated over
-    5,000 versions of the bot, and its win rate on SmolTalkBench, where an LLM judge compares each reply with a reference answer,
+    5,000 versions of the bot, and its win rate on <a href="https://huggingface.co/datasets/HuggingFaceSwarms/SmolTalkBench" target="_blank" rel="noopener">SmolTalkBench</a>, where an LLM judge compares each reply with a reference answer,
     climbed from almost 0% to 26%.</p>
-    <p>The result is PyBot: about 430k lines of Python that answer everyday questions, plus 500k lines of tests the agents wrote to keep
+    <p>The result is PyBot: about 470k lines of Python that answer everyday questions, plus 550k lines of tests the agents wrote to keep
     it working. On SmolTalkBench it's on par with SmolLM-360M-Instruct (26.0% vs 26.6%), without a single neural network inside. You can
     try it below, where you can see which of PyBot's modules produced the reply, and you can click any file to read the code behind the answer.</p>
   </div>