diff --git a/README.md b/README.md
index 01df3236fd2..50f5cebd324 100644
--- a/README.md
+++ b/README.md
@@ -11,8 +11,8 @@ running TensorFlow 0.12 or earlier, please
 
 
 ## Models
-- [adversarial_text](adversarial_text): semi-supervised sequence learning with
-  adversarial training.
+- [adversarial_crypto](adversarial_crypto): protecting communications with adversarial neural cryptography.
+- [adversarial_text](adversarial_text): semi-supervised sequence learning with adversarial training.
 - [attention_ocr](attention_ocr): a model for real-world image text extraction.
 - [autoencoder](autoencoder): various autoencoders.
 - [cognitive_mapping_and_planning](cognitive_mapping_and_planning): implementation of a spatial memory based mapping and planning architecture for visual navigation.
diff --git a/adversarial_crypto/README.md b/adversarial_crypto/README.md
index 431a9d41c49..504ca234beb 100644
--- a/adversarial_crypto/README.md
+++ b/adversarial_crypto/README.md
@@ -4,15 +4,15 @@ This is a slightly-updated model used for the paper
 ["Learning to Protect Communications with Adversarial Neural
 Cryptography"](https://arxiv.org/abs/1610.06918).
 
-> We ask whether neural networks can learn to use secret keys to protect 
-> information from other neural networks. Specifically, we focus on ensuring 
-> confidentiality properties in a multiagent system, and we specify those 
-> properties in terms of an adversary. Thus, a system may consist of neural 
-> networks named Alice and Bob, and we aim to limit what a third neural 
-> network named Eve learns from eavesdropping on the communication between 
-> Alice and Bob. We do not prescribe specific cryptographic algorithms to 
-> these neural networks; instead, we train end-to-end, adversarially. 
-> We demonstrate that the neural networks can learn how to perform forms of 
+> We ask whether neural networks can learn to use secret keys to protect
+> information from other neural networks. Specifically, we focus on ensuring
+> confidentiality properties in a multiagent system, and we specify those
+> properties in terms of an adversary. Thus, a system may consist of neural
+> networks named Alice and Bob, and we aim to limit what a third neural
+> network named Eve learns from eavesdropping on the communication between
+> Alice and Bob. We do not prescribe specific cryptographic algorithms to
+> these neural networks; instead, we train end-to-end, adversarially.
+> We demonstrate that the neural networks can learn how to perform forms of
 > encryption and decryption, and also how to apply these operations
 > selectively in order to meet confidentiality goals.
 
@@ -22,7 +22,7 @@ pairs.
 
 ## Prerequisites
 
-The only software requirements for running the encoder and decoder is having 
+The only software requirements for running the encoder and decoder is having
 Tensorflow installed.
 
 Requires Tensorflow r0.12 or later.
@@ -32,8 +32,10 @@ Requires Tensorflow r0.12 or later.
 After installing TensorFlow and ensuring that your paths are configured
 appropriately:
 
-  python train_eval.py
-  
+```
+python train_eval.py
+```
+
 This will begin training a fresh model.  If and when the model becomes
 sufficiently well-trained, it will reset the Eve model multiple times
 and retrain it from scratch, outputting the accuracy thus obtained
@@ -46,7 +48,7 @@ the paper - the convolutional layer width was reduced by a factor
 of two.  In the version in the paper, there was a nonlinear unit
 after the fully-connected layer;  that nonlinear has been removed
 here.  These changes improve the robustness of training.  The
-initializer for the convolution layers has switched to the 
+initializer for the convolution layers has switched to the
 tf.contrib.layers default of xavier_initializer instead of
 a simpler truncated_normal.