Learning Machine Learning... with ChatGPT

This article forms a collection of sidenotes that I have made whilst journeying from architecture into machine learning. They are a departure from my usual technical and "unconventional truth" posts - their purpose is quite different. The aim of these writings is to reflect on the human experience in a time of artificial intelligence. Through evaluating the technology in this manner, I have been able to reach conclusions on where higher education should go, how to best implement these systems in the working environment and how to interact with artificial intelligence on a daily basis.

In late 2022, I was conducting research on Machine Learning and Artificial Intelligence for my own personal development.

As many would know, approximately a year ago OpenAI's LLM, ChatGPT, was launched and unleashed an explosive AI development race.?

Whilst being restless with both excitement and concern, I took to utilizing it to see how helpful could this new technology be.

Here were some of my findings as I used it to guide me through my learning. I've broken it up into what I was able and unable to do.?

ABLE TO DO's:

1. I was able to summon vast technical terms

I had almost no knowledge of the key concepts which AI and ML were composed of. In a few seconds, I could state phrases with words like convolutional neural networks, reinforcement learning and k-means. It was incredible that I could have them all summed up in one accessible place, and I could consistently ask and receive more.?

2. I was able to strategise a path for myself

As an academic qualification writer, I could understand what constituted a pathway toward learning a new branch of knowledge. I could arrange foundational concepts neatly into levels which would progress into more advanced concepts. This ensured that I had a direction to head toward once I felt I understood enough of a concept.?

3. I was able to produce simulations of machine learning code

To my great surprise, it seemed as thought I didn't actually need 10 000 lines of code to run a simple machine learning algorithm. It appeared that the databases and libraries were already pre-created and all I had to do was deploy them for a specific use-case. This was crazy to me at the time!?

Now, as much as these were phenomenal breakthroughs, I began to realise what my limitations were. The following is a list of what I was unable to do, and despite these being seemingly complex tasks, possibly by next year this time, all my issues issues would have new solutions.?

UNABLE TO DO's:

1. I was unable to understand all the technical terms

As a result of my non-statistical background, I picked up very quickly that I needed to gain insight into core statistical concepts. I used ChatGPT to explain many of them, and it provided enough insight for me to sit and ponder until the concept finally clicked. It was a useful tool to chat to and after sometimes hours of chatting, I finally had to sit and digest the content independently and alone to fully grasp the substance of the content.?

2. I was unable to distinguish my own thoughts from the machine's answers

Once I became stuck, I felt like I was being halted at everything. This psychological barrier was perhaps the greatest to overcome. I also had to figure out where to draw the line between learning an idea and just remembering after the answer was already provided to me. I was sometimes left feeling unfulfilled - which is not the case for me after learning a new concept.

Hence, I deliberately took turns to pause and mull over the idea in my journal, and then tested to see if my understanding was correct my interacting with the LLM. Satisfaction actually came to me once I was able to synthesize the swathes of information I had just consumed.

3. I was unable to get the code running

As easy as it was to deploy the code, I had numerous issues deploying some of the libraries. I especially had issues with one particular library that hindered my progress for weeks.

This was more of a software support issue - from problems with the installation of of Integrated Development Environment (IDE - the programme which runs the code) to issues with installing sci-kit packages (the packages which contain the information I needed to run the Machine Learning algorithms).

These issues made me contemplate possibly creating a survival guide for the number of issues I encountered in the event someone else may have the same problems.

As I write this entry of my experience in learning with an LLM, I recall that I actually attempted a somewhat measured approach and documented my wins and losses. So here is my conclusion:

Overall, ChatGPT assisted me in moving through tonnes of data faster than I could have ever dreamt of doing myself. And as the most important takeaway from this post, I'd say that is more valuable than all its shortfalls.

In a world decision making seems paralysing at times, being able to gather as much data in the shortest amount of time possible means that I can choose to write an article for free, or plan my next paid workshop without hesitating on the impact of either.

// You made it to the end - well done!

...But are you wondering what "reinforcement learning" is?

Or are you a creative considering pivoting into the fields of AI and ML?

Feel free to drop a DM - reach out with any questions you may have.

I'll do my best to answer them //