Advice From a Top 1% Machine Learning Engineer

Summary

In this insightful discussion, Mita Bararia, a senior research scientist at Netflix, shares her journey to becoming an AI engineer and offers valuable advice for those interested in the field. She emphasizes the importance of foundational knowledge in mathematics and programming, the benefits of taking classes and building projects to gauge interest, and the potential for growth with a PhD. Mita also discusses the fast-paced nature of machine learning, the significance of soft skills, and her excitement for the future of AI in solving complex problems.

Takeaways

• 🎓 Pursuing a PhD or Masters in machine learning can provide a strong theoretical foundation and deepen one's understanding of the subject.
• 💡 Starting with the basics of mathematics and computer programming is crucial for building a strong intuition for machine learning concepts.
• 🚀 Hands-on experience through relevant jobs or projects is essential for applying theoretical knowledge in practical scenarios.
• 🌐 The field of machine learning is rapidly evolving, making continuous learning and staying updated with the latest advancements vital.
• 🧠 A strong foundation in fundamentals allows for easier adaptation to new methods and technologies as they emerge.
• 📈 Prioritizing technical skills is non-negotiable, but soft skills like communication, collaboration, and leadership are equally important for success in the industry.
• 🤖 The use of AI and machine learning is expected to expand, enabling software engineers to tackle more diverse and complex tasks.
• 🌟 The future of machine learning holds promise for high-quality innovation in areas such as healthcare and personalized recommendations.
• 📚 Reading seminal papers and revising mathematical concepts can provide a competitive edge in the fast-paced field of AI.
• 🤔 A personal choice between pursuing further education or entering the workforce depends on individual career goals and circumstances.
• 💼 The ability to make fast, informed decisions is crucial due to the rapid pace of innovation in the tech industry.

Q & A

What motivated Mita Bararia to transition from software engineering to machine learning?

Mita Bararia was intrigued by the fields of mathematics and computing during her undergraduate studies in electrical engineering. This interest led her to pursue a job as a software engineer, where she further realized her desire to understand more about machine learning. After taking an introductory course on machine learning, her interest was solidified, prompting her to pursue a PhD in the field.

What advice does Mita have for someone interested in AI and unsure where to start?

Mita suggests taking a class and building a project to determine if one enjoys the field. She emphasizes that understanding whether you like something is the best indicator of whether you should pursue it. She also highlights the importance of having a strong foundation in mathematics and revising these fundamentals as the field evolves.

Is a PhD necessary to become an AI or machine learning engineer?

A PhD is not necessary to become an AI or machine learning engineer, especially for those transitioning from software engineering. There are plenty of resources available, such as online courses and seminal papers, to understand the field without formal graduate education. Gaining hands-on experience in a team setting can also help transition to a full-time machine learning role.

What are the benefits of pursuing a PhD, according to Mita?

Pursuing a PhD provides an opportunity to focus solely on learning and growth. It allows for a deep dive into a subject and helps in maturing one's intuition about it. Additionally, it enhances written and verbal communication skills, builds confidence, and allows for collaboration on a global level.

How does Mita feel about the rapid pace of advancements in machine learning?

Mita views the rapid pace of advancements as both exciting and challenging. She advises revising fundamentals and building intuition for basic models to keep up with the field's evolution. She also believes that with a strong foundation, one can adapt to new developments more easily.

What is Mita's perspective on the role of AI tools like ChatGPT in the future?

Mita believes that AI tools will free up mental space by taking over tasks that can be solved algorithmically. This will allow humans to focus on being more creative and taking on tasks that require a human touch. She emphasizes that having a fundamental understanding is crucial, as AI tools are there to assist, not replace human knowledge and creativity.

How does Mita think the role of software engineers will evolve with AI?

Mita believes that the role of software engineers will expand with AI. Engineers will be able to leverage AI for more tasks and achieve more through prompt engineering. She sees AI enabling more creative work and allowing engineers to take on tasks that were not possible in the past.

What are some soft skills that Mita believes are important for an AI engineer?

Mita highlights the importance of communication, collaboration, decision-making, and being a pleasant colleague. She notes that these soft skills, in addition to technical expertise, can greatly contribute to success in the industry.

What excites Mita the most about the future of machine learning?

Mita is excited about the impact that large language and foundational models can bring to various applications. She predicts that we will see high-quality innovations in areas like healthcare and search, where these models can be fine-tuned for specific tasks.

How does Mita view the importance of continuous learning in the field of AI?

Mita emphasizes that continuous learning is essential in the fast-moving field of AI. She advises staying updated with the latest advancements by reading seminal papers and revising one's mathematical foundations to be prepared for future developments.

What is Mita's stance on the idea that it's too late to become a software engineer given the rise of AI?

Mita strongly disagrees with the idea that it's too late to become a software engineer. She believes that the role of software engineers is not disappearing but rather expanding. AI will enable engineers to do more creative work and take on a wider variety of tasks.

Outlines

00:00

🤖 Introduction to AI Engineering and Career Advice

The video begins with the host addressing common questions about becoming an AI engineer and the nature of the work. The host introduces Mita Bararia, a senior research scientist at Netflix, who previously led the recommendation assistance team at Etsy. Mita shares her educational background, including her PhD in machine learning, and offers advice for those interested in AI. She emphasizes the importance of taking a class and building a project to determine if one enjoys the field. Mita also discusses the value of a PhD for those transitioning from software engineering, highlighting the theoretical foundation and the opportunity for focused learning and growth that graduate school provides.

05:01

🎓 The Decision Between Grad School and Industry Work

Mita discusses the personal choice between pursuing a PhD or working in the industry. She shares her personal experience of craving a pause in her career to engage in academic research, which she found beneficial for her growth as a machine learning scientist. Mita explains that a PhD program allows for deep dives into subjects and enhances collaboration and communication skills. She also addresses the concern that AI advancements are happening so fast that foundational knowledge may become outdated, suggesting that revisiting basics and understanding seminal papers is key to staying relevant.

10:02

💡 Balancing Technical Expertise with Soft Skills

The conversation shifts to the importance of soft skills in addition to technical expertise for AI engineers. Mita stresses that while technical skills are fundamental, soft skills like communication, collaboration, and decision-making significantly contribute to success in the industry. She also mentions the value of being a pleasant colleague, as it fosters a positive work environment. Mita then discusses the rapid pace of innovation in AI and machine learning, expressing excitement about the potential of large language and foundational models to transform various applications. She predicts that these models will enable high-quality innovation in areas like healthcare and personalized recommendations.

Keywords

💡AI engineer

An AI engineer is a professional who designs, develops, and maintains artificial intelligence systems. In the video, Mita shares her journey of becoming an AI engineer and provides advice for others interested in the field. The term is central to the video's theme of exploring career paths in AI and machine learning.

💡Machine learning

Machine learning is a subset of artificial intelligence that involves the use of statistical models and algorithms to enable systems to learn from and make predictions or decisions based on data. In the video, Mita's interest in machine learning led her to pursue a PhD and a career in this field, and she discusses the importance of having a strong foundation in mathematics and programming for success in machine learning.

💡PhD

A PhD, or Doctor of Philosophy, is the highest academic degree that can be earned in many fields. It involves original research and the production of a thesis, demonstrating deep expertise in a particular subject. In the context of the video, Mita obtained a PhD in machine learning, which provided her with the theoretical foundation and research skills necessary for her career as an AI engineer.

💡Electrical engineering

Electrical engineering is a field of engineering that deals with the study, design, and application of electricity, electronics, and electromagnetism. In the video, Mita initially pursued an undergraduate degree in electrical engineering before realizing her stronger interest in mathematics, computing, and programming, which led her to software engineering and eventually machine learning.

💡Software engineer

A software engineer is a professional who applies the principles of software engineering to the design, development, maintenance, and testing of software systems. In the video, Mita worked as a software engineer before deciding to further her education in machine learning, indicating a common transition path for those interested in AI from a software development background.

💡Fundamentals

In the context of the video, fundamentals refer to the basic principles and concepts that form the foundation of a field of study or profession. For AI and machine learning, this includes areas such as mathematics, programming, and understanding core models. The video emphasizes the importance of having strong fundamentals to keep up with the rapidly evolving field of AI.

💡Graduate school

Graduate school refers to higher education institutions that award master's and doctoral degrees. In the video, Mita discusses the personal decision of attending graduate school for further education in machine learning, highlighting the benefits of formal education in deepening one's understanding of AI and developing necessary research skills.

💡Innovation

Innovation refers to the process of introducing new ideas, methods, or products to improve or create new things. In the video, Mita expresses her excitement about the rapid pace of innovation in AI and machine learning, particularly in the development of large language models and foundational models that have the potential to revolutionize various applications.

💡Communication skills

Communication skills are the abilities to convey information effectively and efficiently, both in writing and verbally. In the video, Mita emphasizes the importance of not only having technical expertise but also possessing strong communication skills to articulate complex ideas and collaborate effectively with others in the field of AI engineering.

💡Soft skills

Soft skills are personal attributes and interpersonal skills that are valuable in the workplace, such as communication, collaboration, and decision-making. In the video, Mita discusses the significance of balancing technical skills with soft skills like being an effective collaborator and decision-maker to succeed as an AI engineer.

Highlights

Mita Bararia, a senior research scientist at Netflix, shares her journey of becoming an AI engineer.

Mita's transition from electrical engineering to software engineering sparked her interest in machine learning.

The importance of taking a class and building projects to determine interest in a field.

It is possible to enter the AI field without a PhD through online resources and hands-on experience.

A PhD provides a theoretical foundation and time to mature one's intuition about a subject.

The fast-moving nature of machine learning, with advancements being made almost daily.

The advice to revise fundamentals like mathematics and linear algebra for a strong basis in machine learning.

Mita's experience in college without smartphones did not hinder her ability to learn about mobile technology later.

The personal choice of whether to pursue a PhD or work, depending on one's career stage and desires.

Collaboration and communication skills are vital for a successful career in machine learning.

The value of writing papers during a PhD to hone written and verbal communication skills.

The importance of understanding foundational concepts rather than just memorizing information.

The prediction that large language and foundational models will lead to high-quality innovations in various applications.

The potential for AI to enable more creative work and take on more tasks in software engineering.

The advice for individuals to continue learning and adapting as the field of AI evolves.

The impact of AI on healthcare and recommendation systems through fine-tuning large models.

The significance of being a good colleague and the value of soft skills in the tech industry.

Transcripts

I've been getting a lot of questions

about how do you become an AI engineer

what is it like to work as an AI

engineer so today I thought I am going

to invite one of the best machine

learning Engineers working at a big tech

company here Mita and she's going to

give us advice and tell us more about

how she became an AI engineer hopefully

so you guys can also learn and become an

AI engineer too so Mita thank you for

being here tell us a little bit about

yourself hey everyone I'm Mita bararia

I'm a senior resarch scientist Netflix

prior to Netflix I was working in Etsy a

two-sided marketplace where I was Tech

leading the recommend assistance team

I've have done my PhD with

specialization on machine learning

that's really impressive just the person

that we need to talk to because we get

so many questions on YouTube about

people who are interested in Ai and what

is this new thing and how do we do it

you started very early on before AI

really picked up so yeah tell us the

secret like what sport do your interest

in getting

I did my undergrad in electrical

engineering I love some of the subjects

of electrical engineering but then

towards the end of my undergrad I

realized that I love ma mathematics and

Computing and programming more I uh

decided to take up a job as a software

engineer after mandag grad while I was

enjoying pure software engineering I

also had this itch of knowing more about

a field that's when I chose to come for

my masters in the first semester I took

a course on machine learning

introduction to machine learning in the

mathematics Department Department wow

really sparked my interest and I wanted

to know more about it and that's when I

decided to go for my PhD in machine

learning I really like that because one

of the advice that I always give to

students or people who are exploring is

take a class you build project see if

you even like it maybe you do maybe you

don't and that's the best indicator

whether or not you should pursue it

absolutely you figure out whether you're

into it because you can only be really

good if you're really into it exactly if

you absolutely hate it like don't

it doesn't matter how popular it is

absolutely you did your PhD program

that's another question that we get a

lot do you need a PhD in order to get

into AI machine learning be a ml

engineer or AI engineer if you're

transitioning from a bu software

engineering I think there are plenty of

available resources whether it's courses

and Cora or other online platforms

whether is reading some of the seminal

papers and understanding the

fundamentals possible to understand the

field without actually going to a formal

graduate school right and then F get

into a team which gives you an exposure

in getting hands-on experience then you

can slowly be a full-time ml engineer

from a p software engineering so I think

it's definitely possible somewhat to be

successful uh you need to like some of

the uh theoretical Foundation which I

feel a PhD or grad school gives you that

BS to be able to spend that time with

the subject so that you mature your

intuition about the subject and it gets

hard when we are learning that a job

because we also delivering at a much

faster base if you want the time and

space to be able to really appreciate

the the study and the learning it is

nice to be able to go to graduate school

or PhD program where you can just focus

on learning and growth exactly because

when you're working full-time you just

don't have much time it's a luxury to be

able to study right absolutely yes if

you're not in a place that where you can

go to school or if you're not a school

type of person there are still ways to

get into it by finding opportunities

within where you're already at

absolutely yeah one of the advice I do

give to folks who are getting into ml

revise your fundamentals like

mathematics or the basics of linear

algebra it really helps to build on top

of that ML and eii as you know is

extremely fast moving as a field like

last year in one of the big conference

in machine learning kdd one of the kot

speaker said seeing advancement per day

the way we used to see in a year so the

amount of that need to come out in a

year is now being published every day

almost so it's really fast moving I got

my PhD most of the deep learning methods

that we used today did not even exist

how you like it up is if your

fundamentals are very strong so you just

like back and revise those fundamentals

build intuition for the basic models

like learn logistic regression really

well and explain deepal network from it

because a lot of fundamentals are very

translatable and then you build on it

and then when we have large language

model generative AI foundational model

you can build up on your fundamentals so

if you have the time read up some of the

seminal papers read up your revise your

math and then you're set for the future

like Basics are clear then as the field

is moving you're able to catch up I love

that a lot I also used the example like

I was in college we didn't have

smartphones I know I didn't know

anything about smartphones but I got a

job in Mobile and I was able to learn it

because once you have the basic

fundamental is much easier to pick up

new things one of the biggest dilemas

for a lot of students and we get this

question a lot should I go to grad

school or should I get a job how do you

think going to a PhD program serves you

now as a machine learning or AI engineer

working in the industry the decision of

going pursuing a PhD degree or even grad

school if you're getting your Masters I

think it's definitely a very personal

choice uh and it really depends on where

you are at at that point in your career

if you're already a ml engineer with

your undergrad degree and you're able to

perform really well you don't feel the

need to pause the working and go back to

school and revise and learn the subject

further for me personally I was craving

that pause and being a part of Academic

Program where I can just purely pursue

knowledge that really helped me in my

trajectory as a ml scientist now in

Industry because I could really spend a

lot of time with the subject you spend

many many hours with yourself you think

about one thesis question and spend many

hours days and months and years to to

find the right answer I enjoyed that

Journey my pH was very collaborative I

worked on building machine learning

models for disease prediction and R

stratification so it taught me a lot of

collaboration communication skills while

I'm picking up a lot of deep technical

expertise it also helped me become much

more confident in my ability because of

the individualistic of the pursuit

another skill set that I think PhD

program can really give you is we write

a lot of papers as a PhD student it

helps really really hone your written

communication and verbal communication

and that's one skill set that is

absolutely invaluable in Industry

getting those chances to really horn

your technical skills while developing

communication while developing

confidence within yourself learning to

compete at a very Global level that

builds up a confidence that builds up

your belief in your own ability

yesterday or a couple days ago I saw

this comment on one of my videos this

person was like now chbt can read all

the papers for you so you don't need to

learn how to read did and I thought well

that's not like saying we have

calculators so you don't need to learn

yeah and if Char is reading and

understanding that what are we going to

train CH on in the future yeah right we

don't want just machine generat text we

want real people to be still continuing

to able to write journals and write

papers and the human J text is still

going going to be around and I feel like

even if you do use chbt is just a tool

to help you if you don't have the basic

fundamental knowledge it's not that

helpful even if cha PT reads it for you

Chach PT is the one learning not you

that yes I think in the future we will

free up our mental space we have already

feed up our mental space from memorizing

right we use Google we use other

information on internet we don't

remember everything we needed we used to

with chat GPD with gbd4 with these

generative models I think we will free

up even more of our mental space with

the things that can be solved as a

result as humankind I feel we'll be more

creative eventually it might take a few

years maybe a generation to get there

every other pivotal moments uh like

Industrial Revolution and uh when

internet happened there was unsettlement

in humankind But ultimately we Le we

ended up in a better place yeah I feel

like as humans we are going to get more

creative we're going to focus on other

kind of energy that these tools and in a

way you rightly put so that these are

tools that will help us be more

productive but at the same time we will

as humankind start using our brain power

for something else absolutely I I 100% %

agree I read this book recently called

range I think he said something about

how human IQ has been increasing every

decade as an overall human species we're

getting smarter counter argument I hear

often is like how many phone numbers do

you do you know I'm like why do

we I remember contextual things a lot if

you tell me something that happened to

you if you teach me a concept I'll

remember it forever but if you tell me a

name I'll forget it here growing up I

would hear uh folks say senior previous

generation say that oh this person is

super smart because he or she can

remember stuff I was like why is that

equated to Smart yeah good memory is

part it's part of it but contextual

memory or understanding concept it's

hard to quantify that exactly that's why

I think it's easier to just say if

someone remembers a lot of stuff it's

smart versus it's hard to quantify when

you understand a concept or you're very

creative now I think I also read about

how back in days when the writing first

came out people would say oh my gosh no

one has to remember anything anymore now

we're going to get all done these tools

are serving us and allowing us to be

focusing our energies on more creative

things but maybe in the generation or

two they would think oh back in the days

people didn't do this 100% totally I

think we're very aligned on this and

another question that we often get a lot

is how important is it to balance your

technical expertise with other sof

skills like communication and Leadership

skills to perform your job well of

course the technical skills is the

foundation right you need the expertise

that you are required to perform uh at

any given job so technical skills are

non-negotiable but beyond that a lot of

the soft skills I think go a really long

way communication and collaboration I

think at various point we spoke about

how important it is let's say two

individuals who have equal amount of

technical expertise but one person has

ability or has horned the ability to to

communicate succinctly and and

articulate things and in a timely manner

can go a longer way compared to the

other person so technical communication

to be effective collaborator to be

effective uh researcher to be succinct

and effective Communicator itself is a

very very valuable skill to be

successful in industry in addition to

that other soft skills where just

genuinely being a good colleague to work

with person who others are delighted to

actually work with I think also helps

because who wouldn't want to work with

someone who is who is a happy person to

work with compared to who is a grumpy

person to work with so those are kind of

almost non-quantifiable skill set that I

think are good things to think about

actively and have the to mind uh beyond

your technical skills communication

being a collaborator fast decision

making is another soft skill that I feel

is important because Innovation velocity

in any company in industry is fast

especially machine learning AI the field

is moving fast so you want to see

through the all possible options but

quickly enough and then commit to

something and move on rather than taking

a really long time to get to the

decision of what we should even build

all this thing kind of come together

communication decision making

collaboration and just being a nice

person to work with and you mentioned

several times how things are changing

really quickly what are some things that

we should look out for and what are you

excited about as an ml engineer there's

tons of innovation happening at a pace

that we have never seen before overall I

am very excited about uh the impact

these large language model and the

foundation models can bring to different

applications so what I mean by that is

traditionally we would build specific

models to solve specific problems now

with a really large models foundational

models it's almost like we're able to

train a model to be fifth grader or

undergrad level student and then for a

specific task we can fine-tune the model

to S to be a PhD in a specific area

right so that's how I think about it

instead of training a model from scratch

now we have this well understood about

the basics of the world kind of models

which then we can and for the train to

make them extremely well in specific

task my prediction is that soon we'll

start seeing high quality Innovation

happening for example Healthcare

problems or even recommendation and

search where we find tune the model

Leverage The Power of these greed models

but also solve very specific task and do

it really really well proprietary data

will become even more powerful if you

have a very specific data for a specific

task like gene expression data or

Healthcare data then you can Leverage

The Power of these large models that are

trained on very big uh data set and then

use your propriety data to train

fine-tune the model and get really high

performance we are entering into this

era where we'll see much more improved

applications of ml for very impactful

task sounds like you're more excited

about expanded ability for us to build

more and achieve more as sofware

Engineers yeah yeah and leverage AI for

many more tasks that we have not been

able to leverage that reminds me people

also ask is it too late to become a

Sofer engine should we not do it anymore

more is it useless I we have the answer

yeah yeah absolutely no no it's not

useless I mean the kind of software

engineering you will do is different we

are evolving we we will we are requiring

more and more let's say prompt

engineering the skills that require uh

will expand I don't see it's

disappearing it's expanding yeah like

kind of like what you said earlier about

it will enable us to do more creative

work and take on more tasks and maybe do

things that we haven't been able to do

in the past yes well thank you for

sharing your advice thank you so much

for having me yeah and thanks everyone

for watching too thank you thank you

byee