update text

microsoft · Jun 25, 2024 · d2458b6 · d2458b6
1 parent e2e4ca7
commit d2458b6
Showing 1 changed file with 4 additions and 5 deletions.
diff --git a/docs/index.html b/docs/index.html
@@ -151,11 +151,10 @@ <h1 class="display-1 mb-4">Trace, the new Auto-Diff</h1>
             <div class="row justify-content-center mb-5">
                 <div class="col-12 col-md-8 text-center">
                     <h2 class="h1 font-weight-bolder mb-4">What is Trace?</h2>
-                    <p>Trace is a new AutoDiff-like tool for training AI systems end-to-end with general feedback (like numerical losses, natural language, errors, etc.).  
-                        This gradient-free generalization of back-propagation is made possible by a new mathematical formulation of optimization, we call Optimization with Trace Oracle (OPTO).  
-                        Trace is implemented as a PyTorch-like Python library. With Trace, users can write runnable Python code and use Trace primitives to optimize just the necessary parts to create self-adapting AI systems and train them like training neural networks!
-                        <!-- Instead of propagating gradients (which are not necessarily defined for AI systems),
-                        Trace propagates <i>Minimal Subgraphs</i> (Trace Graphs), which can be used to recover propagated when applicable. Trace is implemented as a PyTorch-like Python library with which users can easily create trainable AI systems and train them like training neural networks. </p> -->
+                    <p>Trace is a new AutoDiff-like tool for training AI systems end-to-end with general feedback (like numerical losses, natural language, errors, etc.).
+                        This generalization of back-propagation is made possible by capturing the AI system's execution trace.
+                        Trace is implemented as a PyTorch-like Python library. Users can directly write Python code and use Trace primitives to optimize the necessary parts to create self-adapting AI systems and train them like training neural networks!
+                        <!--     This generalization of back-propagation is made possible by a new mathematical formulation of optimization, we call Optimization with Trace Oracle (OPTO).</p> -->
                 </div>
             </div>
             <div class="row">