Clarify AVI-Bench evaluation descriptions

index.html

@@ -247,7 +247,7 @@
       <div>
         <p class="eyebrow">Audio-Visual Intelligence</p>
         <h1 id="page-title">Model leaderboard</h1>
-        <p class="intro-copy">Compare model performance across cognitive stages and adaptive evaluation levels.</p>
+        <p class="intro-copy">Compare Omni-MLLM results across AVI-Bench’s Perception, Understanding, and Reasoning stages, its Primitive Sensation (PriSe) extension, and the four-level AVI taxonomy.</p>
       </div>
       <p class="intro-meta">Currently listing <strong id="global-count" aria-live="polite">Loading</strong></p>
     </section>
@@ -275,9 +275,9 @@
     <section class="score-section" aria-labelledby="taxonomy-title">
       <div class="section-heading">
         <div>
-          <p class="section-kicker">02 / Adaptive evaluation</p>
+          <p class="section-kicker">02 / AVI taxonomy</p>
           <h2 id="taxonomy-title">Four-level AVI taxonomy</h2>
-          <p class="section-description">Task-, modality-, stage-, and domain-adaptive performance.</p>
+          <p class="section-description">L1 averages task scores; L2 rewards audio–visual balance; L3 assesses whether reasoning is grounded in perception and understanding; L4 balances familiar-domain performance with generalization to AVI-Bench-PriSe’s unfamiliar, low-semantic inputs.</p>
         </div>
         <span class="record-count" data-count-for="taxonomy-table">Loading…</span>
       </div>