Using ChatGPT to Write a Neural Network: Surprisingly Effective

Using ChatGPT to Write a Neural Network: Surprisingly Effective

by

Machine Heart Report

Editor: Egg Sauce

AI writing AI came faster than expected.

Since its launch at the end of last year, the conversational AI model ChatGPT has become popular throughout the community.

ChatGPT is indeed an amazing tool, like a “Pandora’s box.” Once you find the right way to open it, you may find that you can’t live without it.

As a versatile player, people have made all sorts of requests to ChatGPT. Some use it to write papers, while others have it chat with them; these are common uses. Thinking a bit outside the box, since ChatGPT is the “king” of AI, could it write an AI itself?

Recently, a blogger in the field of machine learning had a sudden idea. He decided to let ChatGPT build a neural network using Keras to solve the MNIST problem.

MNIST is considered a classic problem in the field of machine learning. Even though this handwritten digit dataset has gradually faded from the field of computer vision, its commemorative significance cannot be ignored. Many people’s first dataset for entering deep learning and the first experiment to validate algorithms are all MNIST.

Using ChatGPT to Write a Neural Network: Surprisingly Effective

Without further ado, the author got to work right away. He directly asked ChatGPT to write a piece of code, which looked perfect and didn’t require any changes.

Using ChatGPT to Write a Neural Network: Surprisingly Effective

This is a great start.

ChatGPT chose categorical cross-entropy as the loss function. The author suggested rewriting this part of the code, and ChatGPT replaced it with sparse categorical cross-entropy. The updated code still worked. It can be seen that the new loss function abandoned the conversion of labels.

Using ChatGPT to Write a Neural Network: Surprisingly Effective

The code generated by ChatGPT used fully connected layers, but the author wanted to use convolutional layers. Therefore, ChatGPT modified the code according to the instructions and added a Conv2D and a MaxPooling layer.

Using ChatGPT to Write a Neural Network: Surprisingly Effective

Then came the stage of evaluating the final model, where the author asked ChatGPT to write code using the test set.

Using ChatGPT to Write a Neural Network: Surprisingly Effective

The results looked great, and each of ChatGPT’s replies came with a complete explanation. For example, in this evaluation, it explained:

The “evaluate” method takes the test data and test labels as parameters and returns the loss and metrics specified during model compilation (here, accuracy). The code snippet above outputs the accuracy of the test, which should give you an understanding of how well the model generalizes to new data.

The model used the entire test set as validation data during training. Then, the author asked ChatGPT to change it to 20% of the training data.

Using ChatGPT to Write a Neural Network: Surprisingly Effective

In addition, the author also wanted smaller batches and code that ran fewer epochs. Although this step didn’t necessarily require ChatGPT, he didn’t want to change the code without updating the context with ChatGPT.

So this task still fell on ChatGPT:

Using ChatGPT to Write a Neural Network: Surprisingly Effective

Next, the author planned to plot the training and testing loss during the training process. ChatGPT’s suggestion was: change the line to fit the model to capture its result values.

Using ChatGPT to Write a Neural Network: Surprisingly Effective
Using ChatGPT to Write a Neural Network: Surprisingly Effective

To showcase some examples from the dataset, the author asked ChatGPT to write code to output a combination of images and labels. The output code was also perfect, with a collection of 20 images.

Using ChatGPT to Write a Neural Network: Surprisingly Effective
Using ChatGPT to Write a Neural Network: Surprisingly Effective

When building the model, it is also necessary to view its structure. If you ask ChatGPT this question, the reply is:

Using ChatGPT to Write a Neural Network: Surprisingly Effective

ChatGPT provided a summary of the model:

Using ChatGPT to Write a Neural Network: Surprisingly Effective

The model summary is very useful, but the author wanted to see a diagram showing the model structure. So he continued to ask:

Using ChatGPT to Write a Neural Network: Surprisingly Effective
Using ChatGPT to Write a Neural Network: Surprisingly Effective

The result was satisfactory, and finally, it was time to prepare to deploy this model, allowing ChatGPT to save the model to disk:

Using ChatGPT to Write a Neural Network: Surprisingly Effective

Now, the author wanted to create a class that uses the saved model for prediction. This prompt was quite interesting, and the solution was also perfect.

Using ChatGPT to Write a Neural Network: Surprisingly Effective

Now write an example using the predictor class to predict the labels of 10 random images:

Using ChatGPT to Write a Neural Network: Surprisingly Effective

To complete this, the author asked ChatGPT to display a confusion matrix:

Using ChatGPT to Write a Neural Network: Surprisingly Effective

It must be said that the style used by ChatGPT is quite nice.

Using ChatGPT to Write a Neural Network: Surprisingly Effective

After completing all the experiments, the author published all the code generated by ChatGPT, and you can try it out:

Address: https://colab.research.google.com/drive/1JX1AVIfGtIlnLGqgHrK6WPylPhZvu9qe?usp=sharing

Using ChatGPT to Write a Neural Network: Surprisingly Effective

© THE END

For reprints, please contact this public account for authorization

Submissions or inquiries: [email protected]

Leave a Comment