Can ChatGPT win Machine Learning competitions?
Since the launch of ChatGPT, I’ve asked myself this question: is it possible for an AI (Artificial Intelligence) to enter a Machine Learning competition? And the answer is YES, but hold on, as my grandfather used to say, “All that glitters is not gold.” Let me explain in more detail why.
ChatGPT can be trained for a wide variety of tasks, including machine learning competitions, such as those that take place on Kaggle. However, although its data analysis capability is impressive, there are some significant limitations to consider.
For example, when it comes to data processing, modeling, and prediction, there are many subjective factors and variations that can only be assessed by a human. Even though ChatGPT can be an excellent starting point for data analysis, it cannot fully replace human expertise.
Despite these limitations, ChatGPT can be an excellent baseline for those starting from scratch. It helps to discover initial insights into the data and provides a general understanding of the problem, speeding up the analysis process and providing a solid foundation for subsequent modeling.
Putting ChatGPT to the test!
Now let’s get to the interesting part, proving ChatGPT’s cognitive ability to tackle and write code for a Machine Learning competition. For this, I will use the Kaggle platform, an online platform that brings together a global community of data scientists and Machine Learning enthusiasts. One of its main activities is organizing competitions in which participants have the opportunity to develop and test Machine Learning models to solve real-world problems.
The chosen challenge was a dataset with the goal of predicting daily sales for a store based on a variety of variables, including weather data, holidays, and promotions (you can find all the data in the link at the end of this article).
Incredibly, ChatGPT was able to analyze the complex relationships between variables, performing preprocessing, selecting the model to be applied for training, and finally predicting the store’s daily sales, which was impressive. Although the model did not win the competition, it achieved a very respectable position in the leaderboard and clearly demonstrated the power of artificial intelligence in the field of Machine Learning.
So why didn’t it win the competition? Simply put, ChatGPT still has many limitations, as mentioned earlier. For example, data analysis and Machine Learning modeling still require a high level of human expertise to interpret and understand the results and to ensure that the model is well-fitted to the data, making it so that no machine can currently match our cognitive power.
In conclusion, we see that ChatGPT is not yet capable of surpassing humans. However, we should consider that it is not here to replace us but rather to help us gain valuable new insights and, as previously mentioned, provide a baseline as a starting point to accelerate the development process of Machine Learning models.
Below is the link to the code generated by ChatGPT for the challenge, showing all preprocessing, training, and results obtained by it:
Hope you like it! Thank you very much.
