//

“Explore How Generative AI Can Enhance Your Career”: Eileen Quan with 2U (Video + Transcript)

April 15, 2024
VIDEO

With the changing landscape of Generative AI, discover how this innovative technology can fuel your career, unveil learning pathways, and empower you to navigate its evolution. Eileen Quan (2U Senior Director of Data Science) will talk about how to leverage generative AI to excel in your career and make an impact. She will discuss how to acquire the skills needed to thrive in this evolving field, and encourage collaboration among women in tech to drive innovation in AI.


WATCH ON YOUTUBE

In this ELEVATE session, Eileen Quan, Senior Director of Data Science at 2U (the company behind edX), outlines five key skills for working in generative AI: technical proficiency, model understanding, ethical AI, problem-solving and creativity, and communication and collaboration, and recommends resources and courses for learning these skills, including edX.org, Realpython.com, and huggingface.co. 

Like what you see here? Our mission-aligned Girl Geek X partners are hiring!

Eileen Quan ELEVATE take a course like ai for leaders or ai chatbots without programming edx enhance ai product understanding

Transcript of ELEVATE Session:

Eileen Quan:

Thank you and welcome, everyone. I’m here to talk about elevating your career through generative AI and a high level overview of the skills that you need to work in this new industry.

I’ll quickly give you a recap of what I do and who I am. As was mentioned, I’m senior director of data Science at 2U, a publicly traded edtech company based out of Maryland. We partner with universities and colleges as well as corporations to create and deliver digital content education offerings on our learning platform, edX.org, which may ring a bell for some of you who are familiar with MOOCs.

At 2U, I lead a small team consisting of a data science manager and analyst on product initiatives such as platform feature experimentation, learner behavior analysis, and lead prioritization modeling. I worked at several edtech companies, recently serving as an advisor for CareerFoundry, which is another tech boot camp based out of Germany.

My career path to working with LLM models was obviously not a linear one. I have a background in journalism, which I found to be a very useful skill to have in crafting compelling stories using data.

I’ve switched careers over the last decade, working at a TV station to managing, backing up server banks and writing PHP scripts for e-commerce sites. I eventually fell in love with working on large data sets and validating business assumptions using natural language processing, which ties back to my love of words.

The benefit of working for an edtech company is that I have first-hand knowledge of the type of jobs that are in demand and what skills people are eager to learn. Here at 2U, we have over 4,000 digital offerings, ranging from master’s degrees all the way down to free MOOC courses, working with over 230 partners across the global university and corporations.

For the last several years, we’ve tracked the growing interest in AI landscape and found that the fastest growing courses that adult learners are interested in and enrolling in are in the AI and data science field.

We’re all aware of how generative AI is accelerating at a very fast pace. As you can see, the growing popularity has exploded in the last couple of years due to faster computational capacity and neural networking algorithms. They’re way more efficient than they were five years ago.

Many of the big tech companies that you’ve obviously have heard of have launched their own LLM models to the world for us all to experiment with. And as you could see from this graph here, there’s hundreds of startups that have launched applications and tools built office technology, and they’ve released open source software and APIs for others to build upon.

A lot of these industries, healthcare to finance, retail to entertainment, and of course education, we’ve seen how generative AI can boost our productivity and automate some of the more mundane tasks that we do right now and make us more efficient at work.

At a business level, the growth and automation means increased demand for AI specific skills to design, build, and maintain these systems.

What does this mean for you? Employers are prioritizing creativity and problem-solving alongside expertise in generative AI tools for roles in software development and data science, and product managers and business stakeholders are expected to understand the nuances of this technology and apply it to specific use cases.

Using AI also requires recognizing its limitations. The tools for us that are in use out there right now are still more experimental than production ready. It necessitates human oversight. Hallucinations have occurred with chatbots and computer vision tools that misinterpret data, outputting responses that don’t align with the intended prompts.

One specific example that came out recently, I think a week ago, was Google’s Gemini chatbot featured a diverse version of America’s Founding Fathers. You might have seen some of the news on that and how interesting it was. People have created other memes off of this. Obviously, this technology still needs human intervention to verify the outputs generated to detect and address these issues.

What skills do you need to keep up with demand? When I think about the skills, there are five large buckets that I feel are important: technical proficiency, model understanding, ethical AI, problem-solving and creativity, as well as communication and collaboration.

I designed this developer skills roadmap that I’ve used in my work right now, and many of you are likely already using some ChatGPT or LLM model like Claude or Perplexity. Prompt engineering or crafting instructions for AI systems to output the best possible result is a great starting point and a good skill for everyone to learn.

EdX.org has a free Intro to Prompt Engineering course that highlights some of the strategies to use when tailoring your responses, and it’s also crucial to have a good foundation in machine learning and Python programming.

It’s good to learn about basic neural networks and gain practical experience implementing the simple models, and TensorFlow and PyTorch are two of the more popular deep learning Python libraries that data scientists regularly use to train their neural networks.

Realpython.com provides tutorials, explain the difference between these Python libraries in the best use cases for each.

Huggingface.co is an open source community platform with thousands of ML model variations and open source data sets and sample applications and code that you can actually try out on your own. It’s also good to have a strong grasp of mathematical concepts such as linear algebra. Probabilities and statistics is more or less fundamental for machine learning models because those models are essentially mathematical equations.

Deeplearning.ai offers a free Generative AI for Everyone course, which I’ve taken several months ago. It’s a really great course that’s run by some people that have been working in generative AI for years, and it covers how LLM models work and how the technology can be applied to any industry.

It’s also a good idea to read up about the underlying architecture in the three common models that’s permanently used generative AI. There are GANs, VAEs, and transformers, and I’ll talk more about these models in the next two slides.

But lastly, it’s good to have a specialization. There’s a lot of learning that’s out there, and there isn’t obviously much time to learn everything, and things are changing so fast. So based off the three models I mentioned, they all work very differently and they won’t apply to every business case.

For example, our data science team has used NLP models to analyze students’ learning data such as their performance on assignments and homework and projects. And we use these data points to train our model to auto-generate personalized learning sessions, to provide resource recommendations, additional reading material or exercises tailored to our specific learner’s strengths or weaknesses.

What are these models I mentioned previously? There’s generative adversarial networks, variational autoencoders, and transformer-based models. These three frameworks power most of the applications that are out there right now. You probably have used some of them in some context.

GANs are commonly used for generating media like images and phrases. It consists of a generator, a neural network, and another neural network called a discriminator that work adversarially to generate new data. They’re probably used in entertainment, design, and fashion industries to produce creative content. Whereas VAEs, or variational autoencoders, generate new data by training on existing data sets, expanding those data sets without compromising the quality.

Architects and industrial designers, for example, would use VAEs to create 3D images for new products by training it on 2D models. And stable diffusion is one example that incorporates the VAEs to encode images into a latent space for training and then decoding them back into realistic images during generation.

Lastly, transformers. This has really revolutionized natural language processing. Transformer-based models are what most of us probably are aware of and are learning as well. These models can generate text-based content from diverse sources and create images based on written text descriptions.

Transformers are designed for processing sequential data, especially in natural language processing tasks by using sub-attention mechanisms. ChatGPT, Claude, a whole host of other chatbots are popular examples of applications built using transformer-based models.

Some of the work that I do requires wearing a product manager hat. So I found that some of these strategies are helpful in framing the use of gen AI tools for stakeholders. And as with any project, it’s always a good idea to evaluate your business’ current AI capabilities, like should you build a chatbot for your site or should you just buy one that’s already out there? Taking a look at your data infrastructure, for example.

Do you have enough server space and memory to train and maintain a model? Or do you need to invest in more capacity and resources? And then your engineering team proficiency, how big of a technical lift will this be? How much time will it take for your team to actually learn a new skill? And will gen AI add value and solve a business problem or just be a distraction for your teams?

And if everything is a go, introducing generative AI gradually is a good starting point. You would probably look at the non-critical processes that can be easily automated to allow your teams to adjust without major disruption.

It’s also a good idea to take a specialized course like AI for Leaders or AI Chatbots Without Programming, both of which are on edX.org platform, to enhance your background in AI product understanding. Strategic alignment is very important to ensure that the integration of AI models actually match your broader business objectives of product innovation, customer experience, enhancement or efficiency.

Lastly, embracing innovation within limits. The technology is still experimental stage as we have seen with hallucinations, chatbots not giving you the right information, they’re making things up.

It’s a variety of different issues that have come up. There’s bias in some of the data sets that are used by chatbots as well, so always a good idea to keep that in mind. Then ensure that any kind of models that you do eventually work on, you put in guardrails to make sure that there are certain ways that the results that come out are more accurate and then less of a hallucination or inaccurate.

Some key takeaways here based off of what I’ve talked about. Foundational concepts in mathematics, statistics, and programming fundamentals are a good starting point to learn. Learning about data pre-processing and model training and those different data sets, text, images, audio, and video.

Practice coding in Python along with TensorFlow or PyTorch frameworks. Those are the two frameworks that data scientists use pretty regularly with Python.

I know that people have mentioned R at this point or they ask me about R. R is a good tool to use primarily for research papers and high level analysis because it’s very mathematical intensive, but Python is one language that a lot of businesses regularly use.

I strongly suggest learning Python as a good starting point rather than R.

Understanding neural networks, training algorithms, and model architectures like the three that I mentioned, GANs, VAEs, and transformers, and knowing which one to use and which one would be the best use case for your business. Follow AI blogs.

I’m signed up for a number of AI blogs, so there’s a lot of duplicate information that does come through, but most of the AI blogs I sign up for are actually at specific companies, like OpenAI has a wonderful blog that covers a lot of the things that they’re currently working on. They also have a GitHub open source…GitHub open source site that you could take a look at some of the models they’re working on and they also provide you with some specific examples as well.

It’s a good idea to keep up with the new technology since it’s constantly changing and new developments are occurring, new startups are actually launching very similar things, but really based on more or less similar models that are out there. And also keep in mind about the ethical considerations around AI usage.

There is a lot of bias, a lot of stereotypes that are floating in the data space, so it’s good to keep an eye on that and then definitely call it out and then make sure that these companies are aware that they might be giving out misinformation based off of their tools or products that they release.

Also, good idea to work on personal projects with others in the field or participate in open source AI initiatives to build your portfolio that showcase your new skills. AI projects often require collaboration with a host of different teams at your company, so problem-solving and adaptability are just as important.

Lastly, I think we all have an opportunity to drive innovation by pioneering these new gen AI tools and applications across a host of different industries, and we can all work together to ensure inclusivity and prevent biases in AI systems.

I’m really hopeful that most of us will steer in the direction of AI research and development, building new tools that will combat a variety of different things versus just creating tools that would automate some of our processes that we have.

I hope that this would inspire more women to join the dynamic field. Thank you. I do have an appendix that I can share out with everyone that has a list of all the links for the companies’ blogs as well as additional reading material and organizations that are very AI-specific that you can take a look at.

Sukrutha Bhadouria:

There’s a question for you in the Q&A section. It says, “Please share the list of courses you are recommending, especially the ones for leaders or for those that are not coding directly.”

Eileen Quan:

Yeah, I can share that actually right here. I put in a Google Doc. Hopefully you all can see it. Perfect. Great. On the appendix, there’s some AI specific organizations I listed. There’s a variety of them, so please take a look at that. The gen AI models, that’s specific to, there’s a white paper in there as well as introduction to variational auto encoders, but the other three above that are really good explanations of how transformers and generative adversarial networks actually work so it’s very clear. Those are really good links right there.

In terms of the learnings, I linked the deep learning AI. The edX.org site has a number of classes that you could take, and a lot of them are free. Huggingface.co is also the open source platform dimension that has a number of different ML models on there as well. That’s really useful because you could take a look at that and see what other people have built.

A lot of it is open source, so you could download or take a look at their GitHub account and then make a duplicate of whatever work that they did and then build upon that, which is really nice. It’s a good learning. There’s a lot of people that put the tutorials on huggingface.co as well, and then link it to their GitHub.

AWS, Google, OpenAI, NVIDIA, they all have blogs which are very useful. It spells a lot of things out about what they’re building and how they’re developing their models. I think those are a lot better than some of the other ones that are out there.

Definitely take a look at that, and then realpython.com, that will teach you from the basics of how to code in Python all the way to the more advanced stuff. And I highly recommend that site as well.

Sukrutha Bhadouria:

Thank you.

Eileen Quan:

And then-

Sukrutha Bhadouria:

Yeah. Sorry, you were saying something?

Eileen Quan:

Oh, no, no, go ahead. I think I might be close to time.

Sukrutha Bhadouria:

Yeah, I think we’re at time. Thank you, everybody, for dialing in and thank you, Eileen, for the wonderful content. I hope you all continue to have a wonderful and enriching rest of your day. Bye.

Like what you see here? Our mission-aligned Girl Geek X partners are hiring!

Share this