update abstract

Author: shlomifruchter

index.html

@@ -102,7 +102,7 @@ <h1 class="title is-1 publication-title">Diffusion Models Are Real-Time Game Eng
           <div class="column is-four-fifths" style="max-width:960px">
             <div class="publication-video">
               <iframe
-              src="https://www.youtube.com/embed/O3616ZFGpqw?autoplay=1&mute=1&loop=1&showinfo=0&list=PL3ZfMho22LwDvJSEKVBiwxNsVEqUTUmhJ"
+                src="https://www.youtube.com/embed/O3616ZFGpqw?autoplay=1&mute=1&loop=1&showinfo=0&list=PL3ZfMho22LwDvJSEKVBiwxNsVEqUTUmhJ"
                 frameborder="0" allow="autoplay; encrypted-media" allowfullscreen></iframe>
             </div>
           </div>
@@ -126,16 +126,11 @@ <h2 class="title is-3">Abstract</h2>
             <p>
               We present <i>GameNGen</i>, the first game engine powered entirely by a neural model
               that enables real-time interaction with a complex environment over long trajectories at high quality.
-              <i>GameNGen</i> can interactively simulate the classic game DOOM at over 20 frames per second on a single
-              TPU.
-              <i>GameNGen</i> simulations do not suffer from accumulated deterioration even after long play sessions,
-              achieving a PSNR of 29.4, comparable to lossy JPEG compression.
-              Human raters are only slightly better than random chance at distinguishing short clips of the game
-              from clips of the simulation.
-              <i>GameNGen</i> is trained in two phases:
-              (1) an RL-agent learns to play the game and the training sessions are recorded, and
-              (2) a diffusion model is trained to produce the next frame, conditioned on the sequence of past frames and
-              actions.
+              <i>GameNGen</i> can interactively simulate the classic game DOOM at over 20 frames per second on a single TPU.
+              Next frame prediction achieves a PSNR of 29.4, comparable to lossy JPEG compression.
+              Human raters are only slightly better than random chance at distinguishing short clips of the game from clips of the simulation.
+              <i>GameNGen</i> is trained in two phases: (1) an RL-agent learns to play the game and the training sessions are recorded, and
+              (2) a diffusion model is trained to produce the next frame, conditioned on the sequence of past frames and actions.
               Conditioning augmentations enable stable auto-regressive generation over long trajectories.
             </p>
           </div>