http://linkedin.com/in/nigelmparsad 

1. Could you pick a typical ML task or project that you regularly work on and briefly describe it, including what kind of business objective it fills?

Currently I work on the development chatbots powered by Large Language Models (LLMs) for users to more naturally query enterprise data. This is a very popular LLM application for many companies in a variety of industry verticals.

2. What kinds of math knowledge/libraries do you use? Coding knowledge/libraries?

PyTorch, Langchain, HuggingFace are the three most salient Python libraries for LLM development and workflows.

3. Could you describe the most common way and how often you use probability/statistics in ML? How important is conceptual knowledge vs the ability to use a python library?

I use prob/stats in ML on the data analysis side, e.g., what is the distribution of data and is it skewed in any particular manner.

4. Same for linear algebra?

I used to use linear algebra more when I worked on deep neural networks for computer vision tasks such as facial recognition.  However, even then, I did not require it often.

5. Same for calculus?

My undergraduate degree is in Aerospace Engineering.  Calculus was core to those applications.  However, for the computer vision and LLM AI-tasks, I have never needed even the most basic calculus.

6. In light of ChatGPT/Gemini, do you think it is still important to learn the fundamental math and coding?

There are two types of software and AI developers:  those who understand fundamental math including the details/derivation of the algorithms they employ and everyone else.  

If you want to simply have a job in an AI/software development team, you can skip the math.  

If you want to get ahead and truly work on the most interesting project (and be someone people ask to fill out surveys such as this one :-) ),  do not skimp on coursework in math and coding.  You’ll be an elite employee.

7. Any good resources or recommendations for learning relevant math/coding?

A solid 4 year computer science degree with coursework in AI/data science is paramount.  If you cannot get a degree, there are plenty of excellent online resources with both paid and free content that can get a student up to speed in math/coding  

Finally, when the concepts are still too difficult, check out Youtube videos that discuss the big picture concepts behind the math/coding/AI you are studying.  It is much easier to understand and enjoy  the math and the code if you are shown why they are important and interesting to study!

8. For people who might want to take courses or go for a masters degree in ML, do you think it would be helpful to have an assessment that can measure competency so that a learner knows where they stand, and then learning material and a certificate to mark when they are ready for going forward?

I think a prospective student can self-assess by taking the prerequisite classes for a program and gauging their difficulty.   

(Editor’s note: translation - ask for what courses you could take to cover the pre-requisites, ideally at the institution where you are contemplating taking a degree, then ask for the syllabi of those courses and/or take those courses. There may be a set of online courses on Math for ML on Coursera or EdX).

9. What is your favorite aspect of what you do as an ML professional?

This will sound cliched, but the discoveries, development, and tools are growing exponentially.  We are at the forefront of a new discipline in-terms of real-world application.  In 10 years, what we do today might as well be 100 years in the past.  To be part of this wave is thrilling.

10. Any other thoughts or recommendations?

ML is not software development.  Software developers do not understand the algorithms, infrastructure, operations, and workflows of production ML.  Many software developers will read a bit and do a simple example and think they understand.  They do not.  If you find yourself in an interview for a role that asks you questions better suited for software developers and completely skips deep questions on ML, you are applying for the wrong job.

The true ML companies and research groups will interview very differently than a software dev hiring manager.