Spaces:

zama-fhe
/

private-synthid

Running

App Files Files Community

jfrery-zama commited on Jun 13

Commit

f8cf5f0

unverified ·

1 Parent(s): b058d94

improve wording

Browse files

Files changed (2) hide show

index.html +28 -5
wasm-demo.js +2 -2

index.html CHANGED Viewed

@@ -3,7 +3,7 @@
 <head>
   <meta charset="utf-8" />
   <meta name="viewport" content="width=device-width, initial-scale=1" />
-  <title>Private AI Text Detector – Check if your encrypted text is AI-generated</title>
   <link rel="icon" type="image/x-icon" href="favicon.ico">
   <style>
@@ -395,10 +395,33 @@
   <!-- Hero -->
   <header class="hero" role="banner">
     <div class="hero-copy">
-      <h1>Private AI Text Detection</h1>
-      <div class="explanation-text">
-        Private SynthID uses Google's SynthID technology to identify AI-generated text while preserving privacy through Fully Homomorphic Encryption (FHE). This means you can learn whether your text was produced by an AI—without anyone ever seeing the content. Learn more about <a href="https://deepmind.google/science/synthid/" target="_blank">SynthID</a> and <a href="https://www.zama.ai/" target="_blank">FHE applications</a>.
-      </div>
     </div>
   </header>

 <head>
   <meta charset="utf-8" />
   <meta name="viewport" content="width=device-width, initial-scale=1" />
+  <title>Private AI Generated Text Detection</title>
   <link rel="icon" type="image/x-icon" href="favicon.ico">
   <style>
   <!-- Hero -->
   <header class="hero" role="banner">
     <div class="hero-copy">
+      <h1>Private AI Generated Text Detection</h1>
+      <!-- 1 — Why it matters -->
+      <p class="explanation-text">
+        AI models can embed an invisible <strong>watermark</strong> – a tiny cryptographic
+        signature hidden in the text – that proves the text came from them.
+        This demo shows you can <em>detect</em> such a watermark
+        while your words stay fully encrypted in your browser, thanks to
+        <a href="https://www.zama.ai/" target="_blank">Fully Homomorphic Encryption&nbsp;(FHE)</a>.
+      </p>
+      <!-- 2 — How it works in plain English -->
+      <p class="explanation-text">
+        • Your text is encrypted locally.<br>
+        • Our server runs a watermark detector on the ciphertext.<br>
+        • Because of FHE, the server never sees the plaintext. Neither does it learn whether you text is
+        "watermarked" or "not watermarked."
+      </p>
+      <!-- 3 — Scope of the demo -->
+      <p class="explanation-text">
+        In this prototype we look for the
+        <a href="https://deepmind.google/science/synthid/" target="_blank">Google SynthID</a> watermark
+        as a real-world example.
+        <strong>No watermark</strong> simply means "no SynthID detected" — the text
+        might still be human-written or AI-generated without that specific mark.
+      </p>
     </div>
   </header>

wasm-demo.js CHANGED Viewed

@@ -505,10 +505,10 @@ $('btnDecrypt').onclick = () => {
         let resultText, resultClass;
         if (Number(flag) === 1) {
-            resultText = '✅ AI-Generated (Watermarked)';
             resultClass = 'watermarked';
         } else {
-            resultText = '👤 Human-Generated';
             resultClass = 'inconclusive';
         }

         let resultText, resultClass;
         if (Number(flag) === 1) {
+            resultText = '✅  Text is watermarked';
             resultClass = 'watermarked';
         } else {
+            resultText = '❌ Text is not watermarked';
             resultClass = 'inconclusive';
         }