diff --git a/17_reinforcement_learning.ipynb b/17_reinforcement_learning.ipynb
index 668736a..f1d229a 100644
--- a/17_reinforcement_learning.ipynb
+++ b/17_reinforcement_learning.ipynb
@@ -59,7 +59,6 @@
     "if IS_COLAB or IS_KAGGLE:\n",
     "    !apt update && apt install -y libpq-dev libsdl2-dev swig xorg-dev xvfb\n",
     "    %pip install -q -U tf-agents pyvirtualdisplay gym[box2d,atari,accept-rom-license]\n",
-    "    !AutoROM --quiet --accept-license\n",
     "\n",
     "# Common imports\n",
     "import numpy as np\n",
@@ -1815,6 +1814,22 @@
     "Let's use TF-Agents to create an agent that will learn to play Breakout. We will use the Deep Q-Learning algorithm, so you can easily compare the components with the previous implementation, but TF-Agents implements many other (and more sophisticated) algorithms!"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "First, we need to download and install the Atari ROMs. This can be done very easily using the AutoROM tool, if you accept the license:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!{sys.prefix}/bin/AutoROM --quiet --accept-license"
+   ]
+  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -3210,9 +3225,9 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
+   "display_name": "homl3",
    "language": "python",
-   "name": "python3"
+   "name": "homl3"
   },
   "language_info": {
    "codemirror_mode": {