Wendy - Copywriter to ML Degree

Wendy Aw

Copywriter > AI/ML Engineer

Woohoo!




q: So how long had you been involved in writing before you decided to pursue machine learning?

I was an advertising copywriter for close to 9 years before I decided to take the leap and enter the realm of data science.


q: when and how did ai get on your radar? we’re there any individuals who had an influence, or a book or video series?


Back in 2021, I came across a beta version of an AI copywriting tool. While the headlines it spat out spanned from the mediocre to the nonsensical, most were grammatically correct. So even as it was of no help to my work back then, that experience left a lasting impression and got me intrigued about the power of data. 


q: did you explore over time or was there a point in time where you just decided to dive in and make a change?


As a risk-averse individual, I first dipped my toes into the field by taking several online courses about programming and data. Only after gaining confidence in my learning abilities did I then make the big decision to quit my day job and pursue studying full-time.


q: did chatgpt rock your world?


I had already been playing around with text-davinci-003 prior to ChatGPT’s launch. In fact, during its early days I still preferred prompting using OpenAI’s playground as opposed to using the chat interface since there were more parameters you could control in the playground. That said, ChatGPT has grown to become a great coding and learning buddy (more on that in your later question).


q: when you got serious what courses did you take?

Once I was clear of the switch I was going to make, I took the Data Scientist path on Dataquest. That’s where I picked up python, numpy, pandas etc. Then I went on to take Linear Algebra, Multivariate Calculus, Statistics courses on Coursera as most master’s programs had math prerequisites. Finally, for more hands-on learning, I took up Udacity’s Deep Learning and Natural Language Processing nanodegrees. 



q: did you have a math background? how did you approach math?


My last encounter with math was during high school more than a decade ago. So it definitely took a while to get those mathematical gears oiled and running again. Thankfully, I was never strongly averse to math back in my younger days. However, it also helps to be kind to myself and to take things slowly.


q: many curriculums just state the math requirements and sometimes point to things like khan academy; but often the resources are abstract. Did you experience this and did you ever see resources that were applied math - meaning they showed how the math would actually be used in ML?

Yes, you’re right. Linking the theoretical and the applied math has been a constant struggle. Most online curriculums focus on one or the other and rarely do a good job bridging the two. That’s when I turn to online explainer videos created by the helpful community. One of my favorite channels has to be StatQuest by Josh Starmer.


(Todd note: https://www.youtube.com/@statquest - looks excellent.)


q: now that you are doing ML, do you find that the math part is often done with python libraries?


Yes, sometimes it’s easy to forget about the underlying math concepts since python libraries abstract most of these away. 


q: there are different philosophies - fast.ai is very contextual, hands on, more focused on learning by doing. Deeplearning.ai is more about including coverage of the math. Do you have a philosophy?

I’m not sure I have a learning philosophy per se. I for certain would like to be strong on both the technical and theoretical fronts. However, I did find that doing online courses only gets me so far. Doing real-world work during my internship exposed me to a variety of practical issues that school never speaks of. Picking up new skills on the job, I would say, has boosted my growth significantly.


q: how did you pick a degree and why did you choose to go for a degree?

I chose to pursue a degree as I felt that paper credentials would help get me a foot in the door since my background in communications was so divergent. In the end, I picked UT Austin for the high standing of its computer science department and programs. Financial cost was also a consideration and the master’s UT Austin was among the programs that fit my budget. 


Todd note: I looked into UT Austin and they have several data science, ML and AI-related online Master’s degrees for 10k USD, a very good deal. https://cdso.utexas.edu/mscs 



The conversation about math is directly relevant, which is what I’m researching - how to prepare math skills. Wendy opted to take general math courses in coursera as per above - there may be data science or ML-specific courses as well. What I haven’t yet found is introductory, focused math resources focused on ML/AI, which can help a person brush up to be able to pursue a master’s degree.




q: do you think that people with writing skills can do well with the prompt engineering side of AI? How would you describe prompt engineering?


Having a way with words certainly gives one a leg-up in finding new and better ways to communicate with AI. However, prompt engineering also requires an analytical mindset as it involves discovering the different language patterns that different models understand and generate. Domain knowledge may also be highly crucial when prompt engineering for specific use cases. I’d say that prompt engineering at this time, is simply a catch-all name for a range of skills that we have yet to clearly define. 


q: does the degree you are pursuing involve large language models and generative ai? How would you describe the difference between “conventional ai” (before chatgpt), and ai based on large language models?


I’m currently in the early weeks of the Natural Language Processing module. Based on the syllabus, I’m glad it’s more up to date with the latest discoveries of modern large language models. Post-ChatGPT, LLMs have definitely taken up a great deal of the public limelight. Prior to that, somehow or another, I had the impression that computer vision was the cooler sibling. 


q: I guess to really do generative ai you still need to learn conventional ai such as natural language processing?

As with most fields, I believe it helps to have the foundational knowledge of how the latest developments came to be. 


q: any ideas of where ai jobs will be in the near future? 


Having just entered this space, I can’t say that I have a clearer foresight on this matter. Except perhaps that we can only expect change to be the one constant.


q: what have been the biggest challenges in your learning journey?

One challenge, which probably isn’t unique to me, is in keeping up with all the latest developments in this field. With every passing week, there’s a new model, a new paper or some new technique being released. It can also be disheartening to be aware on a day-to-day basis of my current limits of understanding in this lightning fast-growing sphere. The way I try to grapple with this information avalanche is to take a step-by-step approach, listing down the concepts I’m unfamiliar with and then turning to approachable resources to gradually expand my knowledge in small incremental stages. 


q: what is your favorite area so far? getting the ai pay check?

The AI pay check is a nice bonus to have. But what I really love is watching how I can make a significant contribution in improving the model’s performance over time. While growth may not always be linear, there’s still so much to learn from the failures you encounter along the way. Kind of like the pride of watching a beloved plant grow and thrive through the struggles. (This is me speaking as a plant parent with zero green fingers.)


q: have chatgpt or any other tools been helpful for learning ML? have codepilot, code whisperer, chatgpt been helpful for doing ML/coding? 


The tools are changing all the time but some of them still have the cutoff issue without an active web search and I wonder if Bard or Bing chat have an edge until others have active web connections (because code has changed a lot)


I’ve mostly been using both Claude and ChatGPT 4. Claude has been better in helping me understand ML concepts as it’s trained on more recent information. When it comes to coding, I often corroborate their results against web search results. Both seem to be rather competent in python code, though I would say ChatGPT has a slightly better UI. 


Todd Note: these tools are always evolving, but they are very capable tutors and can help with coding.