Hide text elements in UShapeAttentionCurve compact mode

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: ofriw

website/src/components/VisualElements/UShapeAttentionCurve.module.css

@@ -9,7 +9,7 @@
 /* Compact mode for presentations - maximize diagram size */
 .container.compact {
   margin: 0 auto;
-  padding: 0.75rem;
+  padding: 0.5rem;
   max-width: 95%;
 }
 

website/src/components/VisualElements/UShapeAttentionCurve.tsx

@@ -185,13 +185,15 @@ export default function UShapeAttentionCurve({
 
   return (
     <div className={containerClassName}>
-      <div className={styles.header}>
-        <h4 className={styles.title}>Context Window Attention Curve</h4>
-        <span className={styles.subtitle}>
-          Information at the beginning and end gets strong attention, middle
-          gets skimmed
-        </span>
-      </div>
+      {!compact && (
+        <div className={styles.header}>
+          <h4 className={styles.title}>Context Window Attention Curve</h4>
+          <span className={styles.subtitle}>
+            Information at the beginning and end gets strong attention, middle
+            gets skimmed
+          </span>
+        </div>
+      )}
 
       <svg
         className={styles.svg}
@@ -406,14 +408,16 @@ export default function UShapeAttentionCurve({
         ))}
       </div>
 
-      <div className={styles.explanation}>
-        <strong>The U-Curve Effect:</strong> The curve's width shows context
-        length, while depth shows attention quality. As context length increases,
-        the attention drop in the middle becomes more pronounced. Short contexts
-        (30%) have minimal degradation. Medium contexts (60%) show a noticeable
-        U-curve. Long contexts (90%) exhibit severe attention loss—only the
-        beginning and end are reliably processed.
-      </div>
+      {!compact && (
+        <div className={styles.explanation}>
+          <strong>The U-Curve Effect:</strong> The curve's width shows context
+          length, while depth shows attention quality. As context length increases,
+          the attention drop in the middle becomes more pronounced. Short contexts
+          (30%) have minimal degradation. Medium contexts (60%) show a noticeable
+          U-curve. Long contexts (90%) exhibit severe attention loss—only the
+          beginning and end are reliably processed.
+        </div>
+      )}
     </div>
   );
 }