Solving a Deep Learning Problem Without Writing Code
Explore the possibilities of AI-generated code by solving a simple Deep Learning problem without writing a single line of code. Learn how to use Keras models, Convolutional Neural Networks, and Categorical Cross Entropy to achieve this.
Santiago
Machine Learning. I run https://t.co/iZifcK7n47 and write @0xbnomial.
-
100% AI-generated code.
— Santiago (@svpino) April 19, 2023
I solved a simple Deep Learning problem without writing a single line of code. I kept asking questions until the solution was what I needed.
This is pretty cool: -
Write the code to train a Keras model to solve the MNIST problem.
— Santiago (@svpino) April 19, 2023
Output the code directly without any explanations. pic.twitter.com/CdHLjirvsf -
Do not use a Convolutional Neural Network. Instead, use a simple network of Dense layers. pic.twitter.com/TSfmpjklPd
— Santiago (@svpino) April 19, 2023 -
Instead of using Sparse Categorical Cross Entropy, use Categorical Cross Entropy. pic.twitter.com/rdudIBNkkc
— Santiago (@svpino) April 19, 2023 -
Use 20% of the train set as validation data. pic.twitter.com/bSxUBuSOGG
— Santiago (@svpino) April 19, 2023 -
Write the code to draw a chart that shows the train and validation losses using matplotlib.
— Santiago (@svpino) April 19, 2023
(Notice I made a typo in the prompt, but ChatGPT still knows what I want.) pic.twitter.com/8vKG8f2YK7 -
Remove the plot from the code and instead save the model to disk. pic.twitter.com/FD3uBicpFT
— Santiago (@svpino) April 19, 2023 -
The final code is not the most impressive part.
— Santiago (@svpino) April 19, 2023
The ability to work on your code using ChatGPT is what makes this powerful.
You can steer it one way or another, and it will keep working on the code until you are happy.
If you are a developer, how are you using ChatGPT?