The Future of Data Science with Brian Wright | University of Virginia

00:41:15
https://www.youtube.com/watch?v=hz4YPwtWJbA

概要

TLDRIn this episode of 'Edu Unlocked,' Dr. Brian Wright, associate professor at the University of Virginia’s School of Data Science, shares insights into the evolving field of data science and its educational implications. He elaborates on UVA’s groundbreaking approach, emphasizing a well-rounded curriculum based on a "four-plus-one" model addressing ethics, design, systems, analytics, and practical application. Dr. Wright, with experience in both academia and industry, discusses transitioning to data science from non-technical backgrounds, leveraging online resources, and the importance of blending industry dynamics with academic learning. The conversation also delves into the impact of large language models (LLMs) and AI in education, emphasizing their benefits and limitations. Students, educators, and administrators can gain valuable advice on applying data science in various domains and navigating fast-changing technological landscapes. Dr. Wright candidly discusses the importance of passion, adaptability, and staying ahead by learning new tools and techniques. He encourages open-minded learning to remain relevant in the workforce. The future of AI, personalization in education, and career adaptability are key themes explored throughout.

収穫

  • 🎓 Data science is foundational for the 21st century and transcends various industries.
  • 💻 Transitioning into technology fields is feasible even without a technical background.
  • 📚 UVA’s data science program emphasizes ethics, design, systems, analytics, and practice.
  • 🔍 Large language models (LLMs) bring new opportunities but have limitations like brittleness.
  • 🧑‍🏫 Faculty can leverage industry partnerships and younger colleagues to stay relevant.
  • 🤖 AI tools are already integrated into education but need human oversight for quality.
  • 🌐 Blurring lines between academia and the workforce improves relevancy and practicality.
  • 🚀 Data science skills can lead to diverse career opportunities, from public health to law.
  • 🧠 Human intuition remains irreplaceable even as AI advances.
  • ⚡ The pace of technological change demands continuous learning and adaptability.

タイムライン

  • 00:00:00 - 00:05:00

    Episode introduction with Dr. Brian Wright, a key figure in data science education, is given a warm welcome. Dr. Wright, an Associate Professor at the University of Virginia, is recognized for shaping data science programs and contributing to curriculum development, even involving George Washington University and the Department of Defense. The discussion is set to explore his impact and thought process on evolving data science education.

  • 00:05:00 - 00:10:00

    Dr. Wright discusses the founding of the first Data Science School at the University of Virginia in 2019. With limited faculty, they created a curriculum based on a 'four plus one' model emphasizing ethics, design, systems, analytics, and practice of data science. Unlike others, this balanced model aimed to cover all areas essential for graduates, impactful due to their pioneering status as the first standalone school.

  • 00:10:00 - 00:15:00

    The program initially focused on building structures like administration and student services before rapidly expanding faculty, aiming to hire twelve more this year. Dr. Wright shares his career journey, highlighting his mission-driven approach and transition from working with the Department of Defense to joining academia for better dynamic impact.

  • 00:15:00 - 00:20:00

    Continued career reflection leads to the foundation of a research institute focusing on statistics at the University of Tennessee and further career developments at George Washington University. This highlights Dr. Wright's evolution in the field, embracing teaching roles, driven by opportunities rather than planned transitions.

  • 00:20:00 - 00:25:00

    Dr. Wright emphasizes the feasibility of career transitions into data science, supported by accessible, high-quality non-academic courses and programs. As technology advances rapidly, he underscores the continuous learning necessary for staying relevant and informed about large language models, their applications, and potential.

  • 00:25:00 - 00:30:00

    Examines challenges in deploying large language models given their dynamic responses and potential risk factors. Despite these, Dr. Wright is optimistic about their role in solving customer service gaps, though he cautions against high-risk environments using these models extensively.

  • 00:30:00 - 00:35:00

    Dr. Wright advocates for foundational data science knowledge as essential and universally applicable across industries, preparing students for varied careers. He outlines the University of Virginia's data science program, offering concentrations in different fields to specialize and diversify graduates' skill sets.

  • 00:35:00 - 00:41:15

    Discusses the evolving role of intelligent agents and machine learning in education, highlighting their potential to reshape learning dynamics and interaction. Dr. Wright reflects on the future, expecting faster-paced innovation and adaptation to new tools and methodologies, reinforcing the importance of experience and intuition in data science.

もっと見る

マインドマップ

ビデオQ&A

  • Who is Dr. Brian Wright?

    Dr. Brian Wright is an associate professor of Data Science at the University of Virginia and has helped design the data science curriculum there.

  • What is unique about UVA's School of Data Science?

    It is one of the first dedicated data science schools in the US, focusing on a well-rounded four-plus-one model including ethics, design, systems, analytics, and practice.

  • Can people change careers to data science without a technical background?

    Yes, according to Dr. Wright, individuals from non-technical fields such as music, history, and sociology can successfully transition into data science programs.

  • What advice does Dr. Wright give to students considering data science?

    He suggests data science is a versatile foundation that opens opportunities in many domains, allowing students to tailor their learning to specific industries.

  • What are large language models (LLMs)?

    LLMs are AI models capable of understanding and generating human-like text, used in applications such as chatbots and intelligent agents.

  • What changes are happening in the education-to-workforce connection?

    Dr. Wright highlights the importance of blending academia with industry through partnerships, advisory boards, and adopting innovative technologies.

  • How can faculty stay updated with rapidly evolving technology?

    Collaborating with younger faculty and building relationships with industry experts are key strategies outlined by Dr. Wright.

  • What role do ethics play in data science programs?

    Ethics is one of the key pillars of UVA’s data science curriculum, ensuring students understand the societal implications of their work.

  • Can AI replace human intuition in professions like data science?

    Dr. Wright argues that while AI can assist in processes and lower-level tasks, human intuition and experience remain crucial for solving complex problems.

  • How does UVA integrate technology, like AI, into its curriculum?

    UVA uses intelligent agents, continuous learning tools, and emphasizes real-world applications to enhance its data science instruction.

ビデオをもっと見る

AIを活用したYouTubeの無料動画要約に即アクセス！
字幕
en
オートスクロール:
  • 00:00:00
    so we have everybody from music and
  • 00:00:01
    history and sociology liberal arts
  • 00:00:03
    foreign language you know whatever it is
  • 00:00:05
    those people come in and they do fine
  • 00:00:06
    started from the beginning yeah they're
  • 00:00:08
    very successful sometimes comeing with a
  • 00:00:09
    clean slate is kind of an
  • 00:00:11
    [Music]
  • 00:00:16
    advantage hello everybody and welcome to
  • 00:00:18
    another amazing episode of edu unlocked
  • 00:00:22
    where we meet with inspirational leaders
  • 00:00:24
    from our education leaders in enrollment
  • 00:00:26
    leaders in technology leaders who are
  • 00:00:28
    helping shape in institutions globally
  • 00:00:31
    today we have another special guest with
  • 00:00:33
    us uh somebody who's helped shape the
  • 00:00:36
    world of data science and machine
  • 00:00:38
    learning in some ways possible we have
  • 00:00:41
    Dr Brian Wright with us who's the
  • 00:00:42
    associate professor of data science at
  • 00:00:45
    the University of Virginia's School of
  • 00:00:47
    data science Dr Wright Welcome to our
  • 00:00:50
    podcast all right yeah happy to be here
  • 00:00:52
    it's a lot of pressure I hope I live up
  • 00:00:53
    to the uh live up to that introduction
  • 00:00:56
    thanks no well well there's more coming
  • 00:00:58
    in terms of introduction
  • 00:01:00
    preparation okay good yeah but you know
  • 00:01:03
    I want to let our listeners know that uh
  • 00:01:05
    Dr Wright is not just someone who who
  • 00:01:09
    thinks data science but he also shapes
  • 00:01:11
    he designs curriculum like he's helped
  • 00:01:14
    shape the University's data science
  • 00:01:16
    curriculum um and I want to talk about
  • 00:01:18
    that a little bit what goes into your
  • 00:01:20
    thought process right we'll talk about
  • 00:01:22
    that because that's a field that keeps
  • 00:01:24
    changing uh but other than um University
  • 00:01:27
    of Virginia you've played a pivotal role
  • 00:01:30
    also in the Masters in data science
  • 00:01:32
    program at George Washington you've
  • 00:01:35
    co-founded the data Science Institute
  • 00:01:37
    there that's pretty amazing you've done
  • 00:01:39
    some work for the Department of Defense
  • 00:01:42
    and uh you hold a PhD in higher
  • 00:01:45
    education Administration so you actually
  • 00:01:47
    can you don't just think data science
  • 00:01:50
    and and how those things um should
  • 00:01:52
    operate for the for the workforce for
  • 00:01:54
    students but you also um are you take an
  • 00:01:58
    active role in administration and that's
  • 00:02:00
    amazing uh would love to chat with you
  • 00:02:02
    on that today great happy to do it yeah
  • 00:02:04
    thanks again for thanks again for having
  • 00:02:06
    me yeah fantastic forward to it yeah so
  • 00:02:09
    so so talk to us uh Dr Wright about
  • 00:02:12
    about the data science program at at
  • 00:02:14
    Virginia like what's what's special
  • 00:02:17
    about it and what what got you to design
  • 00:02:20
    that program and what you said you
  • 00:02:22
    thought were like your secret sauce you
  • 00:02:24
    know that you oh yeah okay well well UVA
  • 00:02:27
    data science School the school of data
  • 00:02:29
    science was the first one in the country
  • 00:02:31
    uh no big deal uh but it uh big deal as
  • 00:02:35
    a result
  • 00:02:36
    yeah humle break uh but as as a result
  • 00:02:40
    of that uh it gave us a lot of avature
  • 00:02:44
    to say to ourselves all right so if
  • 00:02:46
    we're going to set a Line in the Sand
  • 00:02:47
    here we're going to create this first
  • 00:02:49
    school like what is what does data
  • 00:02:51
    science mean to us like what does it
  • 00:02:53
    actually represent in terms of the you
  • 00:02:55
    know in terms of the the body of
  • 00:02:57
    knowledge uh in terms of what we want to
  • 00:02:59
    manifest inside of inside our programs
  • 00:03:02
    and so that's the first thing we did and
  • 00:03:04
    that was you know around
  • 00:03:06
    2019 which was just a small group of us
  • 00:03:08
    I don't want to go back wait too much in
  • 00:03:10
    the Wayback machine here but after I
  • 00:03:11
    list GW and then came to UVA I mean we
  • 00:03:14
    only had like four or five faculty I me
  • 00:03:16
    it was a small group you know smaller
  • 00:03:17
    than most departments it's still not
  • 00:03:19
    super big now we have we have
  • 00:03:21
    30 um but you know some of the early
  • 00:03:24
    faculty that were here um Rafa zerado
  • 00:03:27
    and Don Brown and myself and others John
  • 00:03:30
    kopco we sat down and we said okay let's
  • 00:03:33
    let's frame out the space and we came we
  • 00:03:35
    came up with like a I don't want to get
  • 00:03:36
    into details but we came up about like a
  • 00:03:38
    four plus1 model which kind of like
  • 00:03:40
    represents these different areas of the
  • 00:03:42
    field uh and we design all our
  • 00:03:44
    curriculum basically as using that as a
  • 00:03:47
    as a framework um and so if there is
  • 00:03:49
    secret sauce I mean that's probably it
  • 00:03:52
    you know so the four spaces I mean
  • 00:03:55
    generally it's kind of like value and
  • 00:03:56
    ethics all right we have design which is
  • 00:03:58
    about communicating data organizing
  • 00:04:00
    designing data projects and then we have
  • 00:04:02
    systems which is like all the
  • 00:04:03
    engineering like devops like deployment
  • 00:04:06
    also coding programming cloud computing
  • 00:04:08
    and then analytics which is machine
  • 00:04:09
    learning and then kind of at the middle
  • 00:04:11
    is like the practice of data science and
  • 00:04:12
    so a lot of these features were already
  • 00:04:15
    present like in the market you know in
  • 00:04:18
    terms of the way that data science
  • 00:04:19
    programs were coming up but nobody
  • 00:04:20
    really leaned into saying you know you
  • 00:04:22
    need to have each of these in balance
  • 00:04:24
    and in proportion to half like uh this
  • 00:04:27
    what we want graduates to know so so
  • 00:04:29
    that's that's what we did um which I
  • 00:04:32
    think you couldn't really
  • 00:04:34
    do uh unless you had you know kind of
  • 00:04:37
    this space to do it like in your own
  • 00:04:38
    school or like your own department and
  • 00:04:40
    so we had to build all our own classes
  • 00:04:41
    hire our own faculty and so it was a
  • 00:04:44
    real luxury to be able to go through
  • 00:04:46
    that and I'm guessing hiring faculty
  • 00:04:48
    must have been tough because I mean
  • 00:04:49
    we're talking First Data science program
  • 00:04:51
    in the country like yeah that's right
  • 00:04:54
    well well right so was the first school
  • 00:04:56
    right so there's lots of programs out
  • 00:04:58
    there but it was the first one was a
  • 00:04:59
    inone school and yeah that's true yeah
  • 00:05:02
    yeah and even for this year you know in
  • 00:05:04
    the couple of year the first couple of
  • 00:05:06
    years we hired we focused a lot on not
  • 00:05:08
    hiring as much faculty but as much as
  • 00:05:10
    building kind of like the actual
  • 00:05:12
    administrative kind of organization
  • 00:05:14
    right we needed everybody from HR to
  • 00:05:17
    Communications to you know just um
  • 00:05:20
    admissions you know student affairs like
  • 00:05:22
    a lot of that stuff had to be in place
  • 00:05:24
    so we did a bunch of that and then the
  • 00:05:25
    last three years I'd say been we've just
  • 00:05:27
    been pretty much it feels like a whole
  • 00:05:28
    another job just hiring faculty all all
  • 00:05:32
    the time we're searching for another 12
  • 00:05:34
    this year which is a huge number yeah
  • 00:05:36
    for a for for a school and we've done
  • 00:05:38
    that pretty consistently year over year
  • 00:05:39
    it's about that number yeah that's
  • 00:05:41
    amazing what got you into this Dr R uh
  • 00:05:45
    was this a plan all along or you
  • 00:05:47
    stumbled on something and you said oh
  • 00:05:49
    this this looks like a a less riskier
  • 00:05:52
    path for me like talk to us a little bit
  • 00:05:54
    about your career path yeah I don't know
  • 00:05:56
    I it's got It's a Wandering Road I think
  • 00:05:58
    like most people like I had I had ideals
  • 00:06:02
    uh I think that drove my you know drove
  • 00:06:05
    my journey you know I was always very
  • 00:06:08
    you know I I wanted to work on something
  • 00:06:09
    that was that I could put my heart into
  • 00:06:11
    to be honest like I was like a bit of a
  • 00:06:12
    mission focused and to be honest it's
  • 00:06:14
    why I was like at the dod I mean it felt
  • 00:06:17
    that way you know kind of working
  • 00:06:18
    towards you know the the mission of the
  • 00:06:21
    country and the dod was was great it
  • 00:06:23
    ended up being um not a place where you
  • 00:06:26
    could like dynamically improve like very
  • 00:06:28
    fast and then it's kind of like typical
  • 00:06:29
    of most places so I jumped and did
  • 00:06:32
    Consulting um but I stayed in the
  • 00:06:34
    federal government I was still in the
  • 00:06:35
    was in the chairman of the joint tees
  • 00:06:36
    for like three years and that was that
  • 00:06:39
    was moving really fast and it was great
  • 00:06:40
    and the Consulting side of it Lally
  • 00:06:42
    allowed you to kind of build yourself up
  • 00:06:45
    um but then over time it it it became
  • 00:06:47
    like I was trying to get out of DC
  • 00:06:49
    actually because of just like the pace
  • 00:06:51
    of life and so I went back to my Alber
  • 00:06:53
    monor I did everything at the University
  • 00:06:55
    of
  • 00:06:55
    Tennessee um and I I helped stand up a
  • 00:06:59
    uh Research Institute in the business
  • 00:07:00
    school there that was in the statistics
  • 00:07:03
    and Logistics department and people know
  • 00:07:06
    maybe some people do some people don't
  • 00:07:07
    but you know Tennessee has one of the
  • 00:07:09
    best Logistics programs in the country
  • 00:07:10
    for a while was rate is rated number one
  • 00:07:12
    and that's all about like kind of
  • 00:07:14
    Applied statistics and so that's where
  • 00:07:17
    like you know we started doing a ton of
  • 00:07:18
    like applied statistical modeling you
  • 00:07:20
    know we're using some machine learning
  • 00:07:22
    algorithms all sponsored really from the
  • 00:07:24
    Air Force because it had kind of like
  • 00:07:26
    this flavor of like kind of a defense
  • 00:07:28
    Institute
  • 00:07:30
    and I did my PhD at night uh so I worked
  • 00:07:32
    with faculty in those departments on
  • 00:07:34
    research teams from economics and
  • 00:07:36
    statistics during the day and then I was
  • 00:07:38
    like doing PhD stuff at night and so
  • 00:07:40
    that's where I really learned like all
  • 00:07:41
    my data science chops you know I have an
  • 00:07:43
    economics degree so I had like a
  • 00:07:45
    foundation and these quantitative
  • 00:07:46
    methods but the applied part of it like
  • 00:07:48
    solving problems that's pretty much all
  • 00:07:49
    we did like for for three years which
  • 00:07:52
    was awesome um and then you get hooked
  • 00:07:54
    on to it and then you're like okay
  • 00:07:56
    instead of running a million rows on
  • 00:07:58
    spreadsheets and then having stall on
  • 00:08:00
    you you just uh you learn new methods of
  • 00:08:03
    working on that data and then that's
  • 00:08:05
    what that's right and it's right about
  • 00:08:06
    the same time the market started going
  • 00:08:08
    crazy so I was there in like 2010 you
  • 00:08:11
    know 2011 get that right you know
  • 00:08:14
    Harvard piece comes out it's a sexiest
  • 00:08:16
    job in the 21st century to be
  • 00:08:17
    statistician data scientist and then and
  • 00:08:20
    then it just went nuts so I went I got
  • 00:08:22
    married and then we went back to DC
  • 00:08:24
    because that's where our families were
  • 00:08:25
    from and uh so we started my wife and I
  • 00:08:28
    we started our career there and I I
  • 00:08:30
    wanted to stay in education so I got a
  • 00:08:31
    job at GW and one of the first things
  • 00:08:33
    they asked me to do was like help out
  • 00:08:35
    with the data science program and build
  • 00:08:36
    it out and I was like sure and so yeah
  • 00:08:38
    absolutely we did so well I ended up
  • 00:08:41
    going over being faculty there and then
  • 00:08:42
    I've just been in it I've been in it
  • 00:08:44
    ever since yeah that's amazing what's uh
  • 00:08:46
    what's your so it's would you say there
  • 00:08:48
    was a little bit of
  • 00:08:50
    a career change along the way yeah I
  • 00:08:54
    mean I I mean depending on the depending
  • 00:08:56
    on the Fidelity there you know I might
  • 00:08:58
    say that this is like my third or fourth
  • 00:09:01
    career you know being a federal employee
  • 00:09:03
    is much different than being a
  • 00:09:04
    consultant even though I was working in
  • 00:09:05
    the same market so it's kind of like a
  • 00:09:07
    career shift but maybe not really okay
  • 00:09:09
    and then um you know I was really a you
  • 00:09:11
    know I was you know was working as a
  • 00:09:13
    researcher pretty much primarily like
  • 00:09:15
    when I was at Tennessee I was teaching
  • 00:09:16
    some classes but not quite as much but
  • 00:09:18
    when I got to GW it was it was like
  • 00:09:20
    another kind of like a bit of a you know
  • 00:09:22
    in the same Market but a change where I
  • 00:09:24
    was really teaching you know kind of
  • 00:09:26
    full-time you know a lot of teaching a
  • 00:09:28
    lot of classes um which was which was
  • 00:09:30
    different something I had not done in
  • 00:09:31
    the past and that's that's really I mean
  • 00:09:34
    that that part of it I just ended up
  • 00:09:35
    loving you know so I just wanted to do
  • 00:09:37
    that as much as possible how easy do you
  • 00:09:39
    think it is for somebody who is in high
  • 00:09:41
    Ed like working let's say in something
  • 00:09:44
    in academics or or or other
  • 00:09:46
    administrative roles to suddenly say
  • 00:09:49
    yeah I'm going to look at technology I'm
  • 00:09:51
    going to start thinking about business
  • 00:09:53
    analytics or data science or things like
  • 00:09:55
    that because that's the future it can
  • 00:09:57
    probably make me a little more money
  • 00:09:59
    like what's the what's your feedback
  • 00:10:01
    there I think it's totally possible I
  • 00:10:04
    mean I think it's more easier now than
  • 00:10:05
    it kind of ever has been I mean there's
  • 00:10:07
    so many I mean there's there's high
  • 00:10:10
    quality like non- accredited like
  • 00:10:12
    outside of school programs that you can
  • 00:10:13
    take would get you a long way I mean
  • 00:10:15
    like you know IBM like Google they have
  • 00:10:17
    their own classes you know that you know
  • 00:10:19
    they're they're good you know the stuff
  • 00:10:21
    that's on corsera or you know edx like
  • 00:10:25
    if you didn't want to go full Academic
  • 00:10:26
    Program you just wanted to pick up some
  • 00:10:27
    skills and apply it inside your job like
  • 00:10:30
    it's and you make a hard argument that
  • 00:10:33
    those are just as good as maybe some of
  • 00:10:35
    the accredited programs nowadays I mean
  • 00:10:37
    they're good um a lot of that stuff
  • 00:10:39
    didn't wasn't even present really when I
  • 00:10:41
    was when I was starting to pick up a
  • 00:10:43
    bunch of this stuff exactly what I was
  • 00:10:44
    gonna say is there's there's so much
  • 00:10:46
    coming every time like you know we last
  • 00:10:50
    time we spoke we talked about like llms
  • 00:10:52
    and it's just right the person who knows
  • 00:10:54
    the most about it is probably like has
  • 00:10:57
    like two years or three years of
  • 00:10:59
    knowledge
  • 00:10:59
    on it yeah yeah that's totally fair I
  • 00:11:03
    mean it the pace of the pace of
  • 00:11:05
    Technology Innovation now is is I mean
  • 00:11:08
    it's our job I mean in large I mean in
  • 00:11:10
    large part our job to like try and stay
  • 00:11:12
    up with like all these methods because
  • 00:11:14
    as you're as you're researching and like
  • 00:11:16
    creating projects or delivering
  • 00:11:17
    curriculum in the classroom you want to
  • 00:11:19
    stay right there but it is so much
  • 00:11:21
    harder now even than it was just a
  • 00:11:23
    couple of years ago I mean the pace of
  • 00:11:25
    change is really fast um and so
  • 00:11:28
    sometimes it's another Reon why you know
  • 00:11:30
    some of the online platforms they may
  • 00:11:31
    respond even quicker than we do in
  • 00:11:33
    Academia but again you know I think it's
  • 00:11:36
    it's totally possible to pick that up
  • 00:11:37
    and I you know we have our own like
  • 00:11:39
    large language model research group you
  • 00:11:41
    know that I'm that I have an active
  • 00:11:42
    learning lab so we're building
  • 00:11:44
    intelligent agents and putting them in
  • 00:11:45
    the classrooms and then trying to track
  • 00:11:47
    and see how students are using them or
  • 00:11:50
    how how to be honest how really brittle
  • 00:11:52
    they are like how they hard they are to
  • 00:11:54
    train um and fine-tune to make sure that
  • 00:11:56
    you can get consistent results it's
  • 00:11:58
    still I think it's still still an open
  • 00:11:59
    question to be honest yeah and I was
  • 00:12:03
    going to talk about that I mean you know
  • 00:12:04
    the the the theme I think here is uh
  • 00:12:07
    there there are listeners who are in
  • 00:12:08
    highed who are looking for leadership
  • 00:12:11
    roles but who're also looking for
  • 00:12:12
    Transitions and what should be the what
  • 00:12:14
    should they be doing next so I think
  • 00:12:16
    sure you know that that definitely helps
  • 00:12:18
    learning from you on uh just based on
  • 00:12:22
    your career trajectory that it's if you
  • 00:12:24
    want to switch to something that you
  • 00:12:26
    know that's in technology and you've not
  • 00:12:27
    done it before ever it does doesn't mean
  • 00:12:29
    you know you have to have had a
  • 00:12:31
    bachelor's in computer science at least
  • 00:12:33
    no you're just like these are things
  • 00:12:35
    today you can learn even on the internet
  • 00:12:37
    yeah I mean to be clear I mean most of
  • 00:12:38
    the students that are in our graduate
  • 00:12:40
    level programs they don't they don't
  • 00:12:41
    come from those backgrounds I mean there
  • 00:12:42
    is there is something we can offer I
  • 00:12:44
    mean our degree is kind of a career
  • 00:12:46
    transition degree the one we offer here
  • 00:12:47
    which is very typical of a lot of data
  • 00:12:50
    science degrees and so we have everybody
  • 00:12:52
    from music and history and sociology you
  • 00:12:55
    know liberal arts foreign language you
  • 00:12:57
    know whatever it is those people come in
  • 00:12:58
    and they do fine you kind of start from
  • 00:13:00
    the beginning yeah they're very
  • 00:13:02
    successful sometimes sometimes comeing
  • 00:13:05
    with a clean slate is kind of an
  • 00:13:06
    advantage to be honest you know so way
  • 00:13:09
    yeah you be Prejudice I mean you know
  • 00:13:11
    for some of the people that come from
  • 00:13:13
    stats or from you know computer science
  • 00:13:15
    I mean sometimes it's hard have to kind
  • 00:13:16
    of they have to translate what they've
  • 00:13:19
    learned in the past to what we're
  • 00:13:20
    showing them now that translation effect
  • 00:13:22
    can take can take can take energy that
  • 00:13:26
    is true that is yeah that's a very good
  • 00:13:28
    point yeah yeah I mean I I did my
  • 00:13:31
    bachelor's and Masters in biotech and
  • 00:13:33
    I'm running a tech company now and we
  • 00:13:36
    deal with machine learning and AI stuff
  • 00:13:39
    all all day all night uh but I I I don't
  • 00:13:43
    I didn't ever feel
  • 00:13:45
    the there was a gap that you know I
  • 00:13:47
    needed a a computer science degree or
  • 00:13:49
    something to fill in there's so much you
  • 00:13:51
    can learn just just by getting on the
  • 00:13:53
    Internet it's crazy yeah so so so
  • 00:13:56
    talking about llms now that we're on
  • 00:13:58
    that topic that's one of like your
  • 00:14:00
    favorite subjects uh you touched upon it
  • 00:14:03
    last time you're very intrigued by where
  • 00:14:05
    they're going uh so do you think we're
  • 00:14:08
    ever going to have a Bank of America bot
  • 00:14:11
    or booking.com bot where we actually
  • 00:14:13
    don't have to say hey I want to talk to
  • 00:14:15
    an agent you know I I was doing this
  • 00:14:17
    work directly but through a colleague of
  • 00:14:18
    mine who runs you know it's the head of
  • 00:14:20
    an AI research you know group here
  • 00:14:23
    they're deciding some of these and part
  • 00:14:24
    of his job is just to break the large
  • 00:14:26
    language models he's a red team you know
  • 00:14:28
    his his research team is is the is
  • 00:14:30
    basically the red team before go live
  • 00:14:32
    you know so people will build prototypes
  • 00:14:33
    they come to him and they they break
  • 00:14:36
    them like a lot you know just through
  • 00:14:38
    compression testing so you can just fire
  • 00:14:40
    off basically train one AA bot to like
  • 00:14:42
    attack the other one through a series of
  • 00:14:44
    prompts right so it might be they're
  • 00:14:45
    answering thousands of thousands of
  • 00:14:47
    progressive questions but eventually
  • 00:14:48
    that large language model gives you
  • 00:14:50
    something it shouldn't you know it gives
  • 00:14:51
    you like movie quote or it says hey put
  • 00:14:53
    your money in Bitcoin you're like okay
  • 00:14:55
    have to be careful you know so you know
  • 00:14:58
    you can as many guard rails as you want
  • 00:15:00
    to put in place you know that I think
  • 00:15:02
    for especially for organizations that
  • 00:15:04
    are like heavily risk averse like large
  • 00:15:07
    lers models is a bad choice I mean they
  • 00:15:09
    do something simpler like if there's a
  • 00:15:10
    real problem there like they're losing
  • 00:15:12
    customers because the chatbot just isn't
  • 00:15:14
    authentic enough like okay that's
  • 00:15:16
    usually not the case usually it's a
  • 00:15:18
    feature not not actually a you know it's
  • 00:15:20
    that derivative of the entire business
  • 00:15:22
    model so then you then it's really a
  • 00:15:24
    question of how much risk you want to
  • 00:15:25
    put into this feature you know maybe
  • 00:15:27
    maybe not much yeah but I think like um
  • 00:15:31
    you know the the me the memory loss
  • 00:15:33
    problem you know associated with like
  • 00:15:35
    how quickly like they atrophy in terms
  • 00:15:37
    of like their responses that are high
  • 00:15:39
    quality over time is still a problem um
  • 00:15:43
    you know rag distributions distilled
  • 00:15:45
    large language models like controlling
  • 00:15:47
    the you know the embedding spaces like
  • 00:15:49
    these work like prompt Ed like all these
  • 00:15:51
    things work they they're arbitres
  • 00:15:53
    against that like problem but it's still
  • 00:15:58
    an organic feature like it's still so
  • 00:16:00
    you have to guard again so like it's
  • 00:16:01
    still there and then and then you know
  • 00:16:05
    you lose some of the niceties of like
  • 00:16:07
    traditional you know word deac where
  • 00:16:10
    it's like you know King to Queen is to
  • 00:16:12
    Prince to princess that doesn't exist in
  • 00:16:14
    large languish models so they have these
  • 00:16:15
    empirical features that uh aren't there
  • 00:16:19
    because of the size of them so they they
  • 00:16:21
    you know the nature of the how quickly
  • 00:16:24
    they they respond to prompts and stuff
  • 00:16:25
    means that they have to kind of push
  • 00:16:27
    everything out
  • 00:16:29
    board and so the embedding spaces
  • 00:16:31
    becomes kind of just on the surface so
  • 00:16:34
    interesting so so talking about that I
  • 00:16:36
    mean do you think you know like when
  • 00:16:39
    when Wikipedia started we said oh my God
  • 00:16:42
    it's going to be at some point the
  • 00:16:43
    largest like Encyclopedia of information
  • 00:16:46
    and everything and then they could never
  • 00:16:49
    solve the problem of fake information on
  • 00:16:51
    Wikipedia because I could launch a page
  • 00:16:53
    on myself and write whatever I want uh
  • 00:16:56
    and it's hard to get it peer reviewed or
  • 00:16:59
    or or do that kind of stuff so do you
  • 00:17:01
    think the fact that llms like chat GPT
  • 00:17:04
    actually learn from all the information
  • 00:17:06
    that's out there and and those so that
  • 00:17:09
    whole challenge of hallucination which
  • 00:17:11
    now they claim in the next version it's
  • 00:17:13
    going to be completely gone but do you
  • 00:17:15
    think all that is even possible
  • 00:17:18
    considering that like so much of the
  • 00:17:20
    information that it's learning from it's
  • 00:17:22
    training data actually has non-factual
  • 00:17:25
    information in it yeah it's it's weird
  • 00:17:27
    because you can't I don't think you can
  • 00:17:29
    think of large language models as a data
  • 00:17:30
    store like I just don't think that's the
  • 00:17:33
    way that you I think if you if you
  • 00:17:34
    approach it that way you're destined for
  • 00:17:35
    failure you know and I I mean the
  • 00:17:37
    quality of like these AI generated
  • 00:17:39
    search responses like on top of your
  • 00:17:40
    browsers and stuff like they're kind of
  • 00:17:42
    good but like is it better than just a
  • 00:17:44
    traditional like I don't know
  • 00:17:47
    vector-based keyword you know uh prompt
  • 00:17:49
    I don't know
  • 00:17:51
    so I uh I I think if we think of it as
  • 00:17:55
    like may maybe an analogy is that you
  • 00:17:57
    know we all we're all learning to read
  • 00:17:59
    right you know and you become a
  • 00:18:00
    proficient reader like at what age like
  • 00:18:02
    a in maybe 10th or 11th grade maybe
  • 00:18:04
    maybe onward after that I think we have
  • 00:18:06
    to think of like these Foundation models
  • 00:18:08
    as like proficient readers like they
  • 00:18:10
    understand the they understand like uh
  • 00:18:13
    you know language and they can respond
  • 00:18:15
    to it but just like a 12th grader they
  • 00:18:17
    don't know everything right and so like
  • 00:18:20
    they're G you can't just just because
  • 00:18:21
    you can read and write doesn't mean you
  • 00:18:23
    have a capacity of knowledge and I mean
  • 00:18:27
    maybe it's a loose analogy but if you it
  • 00:18:29
    like that like a like a like a you know
  • 00:18:32
    like an intelligent foundation with
  • 00:18:34
    which you can then build something on
  • 00:18:35
    top I think I think the the conversation
  • 00:18:38
    works much better if you use it as
  • 00:18:40
    simply a prompt recall type of I don't
  • 00:18:42
    know to me that it doesn't work as well
  • 00:18:45
    and I I that's why it works really well
  • 00:18:46
    for like editing or like idea iteration
  • 00:18:49
    or like outlines or summaries because
  • 00:18:51
    you provided the information it needs it
  • 00:18:53
    understands the language and it can give
  • 00:18:54
    it back to you in a way that you want it
  • 00:18:56
    to right which is an ideal use cas case
  • 00:18:59
    as compared to building it into some
  • 00:19:01
    application that must generate new
  • 00:19:04
    information or timely information like
  • 00:19:05
    you're destined for failure I mean even
  • 00:19:07
    if it doesn't hallucinate like I get it
  • 00:19:09
    but it doesn't understand meaning like
  • 00:19:12
    it just doesn't like there's no meaning
  • 00:19:13
    there as much as people want it to be if
  • 00:19:15
    you look at them underneath the hood
  • 00:19:18
    empirically I mean it just it doesn't it
  • 00:19:20
    it uses this the distribution hypothesis
  • 00:19:22
    right that words that are close together
  • 00:19:23
    mean something and it does work but it
  • 00:19:26
    doesn't mean that it understands the
  • 00:19:27
    words that it's producing
  • 00:19:29
    interest interesting thought there yeah
  • 00:19:31
    I mean I'm I'm really looking forward to
  • 00:19:33
    see where it can go because it seems to
  • 00:19:36
    me like it has the ability to make you
  • 00:19:39
    know like science we've had this in
  • 00:19:42
    science fiction for a while where
  • 00:19:43
    there's this assistant in your year all
  • 00:19:46
    the time who knows so much about you and
  • 00:19:49
    uh and it kind of felt when CH GPT
  • 00:19:51
    launched the latest like
  • 00:19:53
    40 I've been using it a lot like when
  • 00:19:56
    I'm walking my dog I'm talking to it
  • 00:19:57
    just to kind of see
  • 00:19:59
    it's and it's actually kind of amazing
  • 00:20:03
    like where it's where it's at right yeah
  • 00:20:06
    no I don't disagree it is a a total
  • 00:20:08
    technology breakthrough like I'm I'm I'm
  • 00:20:10
    sayin on the future you know I think
  • 00:20:12
    they're going to be you know interesting
  • 00:20:15
    I it I think the multimodal aspect of it
  • 00:20:18
    I think is going to be I think there's a
  • 00:20:20
    lot left to know about what how the
  • 00:20:23
    combination of voice and image and sound
  • 00:20:25
    and text together like those things I
  • 00:20:27
    think that's a very interesting
  • 00:20:30
    Prospect I mean the continuously
  • 00:20:32
    continuous learning aspect of it is
  • 00:20:34
    really cool too that you can create AI
  • 00:20:36
    agents on top of the models and they
  • 00:20:38
    just continue to learn over time I think
  • 00:20:40
    that's a good one almost like an intern
  • 00:20:42
    you know you embed them into your team
  • 00:20:44
    and it gets to know you more and more
  • 00:20:46
    because it understands language you can
  • 00:20:47
    speak to it in a Common
  • 00:20:49
    Language and so and there's been some
  • 00:20:52
    interesting you know back and
  • 00:20:54
    discussions I would say and people that
  • 00:20:56
    are closer to this as it relates to like
  • 00:20:58
    low dens it languages languages where
  • 00:21:00
    there's not a lot of text like how to
  • 00:21:02
    train agents using English to translate
  • 00:21:05
    and basically you relying on Experts to
  • 00:21:07
    kind of continuously have conversations
  • 00:21:09
    over time and that creates a database
  • 00:21:12
    that actually can be used as a as more
  • 00:21:14
    of a repository for training and what
  • 00:21:16
    that looks like so I think there's yeah
  • 00:21:19
    we're certainly just getting started
  • 00:21:20
    right I mean it's like you said you know
  • 00:21:22
    the an expert might be four or five
  • 00:21:25
    years you know worth of experience or
  • 00:21:27
    something so we're just getting started
  • 00:21:29
    and it's it's cool the progress is
  • 00:21:30
    making and it seems like every every
  • 00:21:32
    week There's a new paper with a new
  • 00:21:34
    breakthrough you know that that that
  • 00:21:36
    allows you to think about these things
  • 00:21:37
    differently so AB absolutely yeah and uh
  • 00:21:40
    and you know coming to students so we
  • 00:21:42
    have some of our listeners are students
  • 00:21:45
    uh who who always want to think about uh
  • 00:21:48
    whether they're on the right track right
  • 00:21:49
    every student when we were students we
  • 00:21:51
    were paranoid to I might take picking
  • 00:21:53
    the right course the right degree will I
  • 00:21:55
    get the right job yeah you you seem to
  • 00:21:58
    have been someone who's adapted to
  • 00:22:01
    Changing Times right you like you moved
  • 00:22:03
    into data science you've now got into
  • 00:22:05
    like trying to understand llms that's
  • 00:22:07
    the newest thing and you're like oh I'm
  • 00:22:09
    intrigued and you're comparing it to
  • 00:22:10
    Legacy systems like oh yeah you know
  • 00:22:13
    this is a word to I love it uh in terms
  • 00:22:16
    of students who are thinking about you
  • 00:22:19
    know what whether data science is right
  • 00:22:21
    it kind of feels like that you know that
  • 00:22:24
    article that came out about data science
  • 00:22:25
    being the number one job no more like
  • 00:22:28
    stands true now there's newer things
  • 00:22:31
    what what trajectory do you see um uh
  • 00:22:34
    the world progressing in and what would
  • 00:22:36
    be your suggestion and advice to those
  • 00:22:39
    students who are starting out well again
  • 00:22:41
    you know I I'll say this from I am of
  • 00:22:44
    course bias but I I I think that the way
  • 00:22:48
    that we've positioned our our
  • 00:22:49
    undergraduate degree and we've
  • 00:22:51
    advertised and talked about it is that
  • 00:22:52
    it's a it's a it's a liberal arts degree
  • 00:22:55
    for the data age right it's not a
  • 00:22:56
    liberal arts degree but you can think of
  • 00:22:58
    it as like you know a foundational
  • 00:23:01
    knowledge for the 21st century I mean uh
  • 00:23:04
    computer science should make the
  • 00:23:05
    argument it's same you know their their
  • 00:23:07
    advant their advantage is there uh for
  • 00:23:11
    traditional kind of software development
  • 00:23:12
    skills are always going to be kind of
  • 00:23:14
    you know present but the the really
  • 00:23:16
    interesting part about about the data
  • 00:23:19
    science aspect of it this is true of of
  • 00:23:21
    other fields as well is that you can
  • 00:23:23
    take those skills and then go into so
  • 00:23:26
    basically every Market ver right or
  • 00:23:29
    every kind of area and then that could
  • 00:23:31
    be a whole career you know the way you
  • 00:23:34
    think about it just learning and
  • 00:23:35
    understanding how they use data in that
  • 00:23:37
    particular space the methodologies and
  • 00:23:39
    the me and the techniques totally
  • 00:23:41
    translate right but it's the field that
  • 00:23:44
    can make the difference and it yeah so
  • 00:23:46
    to me I I I just I can't I can't see
  • 00:23:50
    getting a data science degree as as
  • 00:23:52
    being a bad choice the way I think of it
  • 00:23:53
    is like now especially with like the
  • 00:23:55
    knowledge economy and like you know with
  • 00:23:58
    the typ typical working kind of you know
  • 00:24:00
    education level is is that you think of
  • 00:24:02
    yourself as getting this like
  • 00:24:04
    well-rounded you know kind of foundation
  • 00:24:06
    in data science skills right which
  • 00:24:08
    includes like you know lots of things
  • 00:24:10
    that are heavily marketable which can
  • 00:24:12
    allow you to go right into the market
  • 00:24:14
    but then you can also think about well
  • 00:24:15
    what domain area could I bolt on top of
  • 00:24:18
    this like maybe in the future like in
  • 00:24:20
    public health or in you know who knows
  • 00:24:23
    what in GIS or in finance or whatever
  • 00:24:26
    and so you know there's ways where it
  • 00:24:28
    allows you to or law I mean we have a
  • 00:24:30
    ton of we have a ton of students that
  • 00:24:32
    are very interested in going into law
  • 00:24:34
    and the the justice department last year
  • 00:24:36
    changed their or was a year before they
  • 00:24:39
    changed their like summer internship
  • 00:24:40
    program where they weren't hiring
  • 00:24:42
    pre-law or economics they were only
  • 00:24:44
    hiring data analysts which was like a
  • 00:24:45
    big demand signal for like where they
  • 00:24:47
    thought the field was going to go so
  • 00:24:48
    they're bringing in all these data
  • 00:24:49
    analyst Majors into the justice
  • 00:24:52
    department to help them manage their you
  • 00:24:54
    know their projects and so I I think of
  • 00:24:57
    it as that way it's just like a you just
  • 00:24:59
    can't miss kind of foundation and then
  • 00:25:02
    you can make the choice about where you
  • 00:25:03
    want to go and that the last part I'll
  • 00:25:04
    say is that we actually built that into
  • 00:25:07
    the degree so we have we have nine
  • 00:25:08
    concentrations in different areas
  • 00:25:10
    everything from public policy and
  • 00:25:11
    Leadership from economics to accounting
  • 00:25:13
    to Public Public Health Neuroscience
  • 00:25:17
    astrophysics like we have all of these
  • 00:25:19
    concentrations that are part of the
  • 00:25:21
    degree that the students once they get
  • 00:25:22
    through our core then they just jump in
  • 00:25:24
    to a concentration may be do many of
  • 00:25:27
    them which is really cool you you're
  • 00:25:28
    part of the data science program and
  • 00:25:30
    then you can concentrate in Neuroscience
  • 00:25:32
    when then you're learning more about
  • 00:25:34
    clinical data CL resarch data all all of
  • 00:25:38
    that kind of stuff yeah that's exactly
  • 00:25:40
    right yeah and then the idea is that um
  • 00:25:42
    and this was designed by The Faculty
  • 00:25:44
    from those departments is that they
  • 00:25:45
    wanted students to take a few of the
  • 00:25:47
    courses because they're going to come
  • 00:25:48
    with these data science machine learning
  • 00:25:50
    skills take a few entry level courses so
  • 00:25:52
    they'll be
  • 00:25:53
    conversational in that field and then
  • 00:25:56
    they can take them and put them into
  • 00:25:57
    their lab groups so that's a combination
  • 00:25:59
    of like this basic understanding of the
  • 00:26:00
    field and then also like two and a half
  • 00:26:03
    years of data you know curriculum it'll
  • 00:26:06
    be pretty valuable yeah I love it yeah
  • 00:26:09
    it's amazing and you know I was I was
  • 00:26:11
    expecting you you would say that because
  • 00:26:13
    I I do agree and I feel the same way as
  • 00:26:16
    I mean even when you're thinking machine
  • 00:26:18
    learning AI Core data science skills
  • 00:26:21
    always are of prime importance uh yeah
  • 00:26:24
    yeah I mean there's there's a lot of
  • 00:26:26
    abstraction that you know companies like
  • 00:26:28
    AWS and Google cloud and all of them are
  • 00:26:30
    doing where now you can take all of your
  • 00:26:33
    data and just you know magically like
  • 00:26:34
    just bump it against a model and it it
  • 00:26:37
    spits back some result but understanding
  • 00:26:41
    the core of that understanding how data
  • 00:26:43
    actually works uh yeah that's that's
  • 00:26:45
    that's super important so you know we're
  • 00:26:47
    always looking for data science uh
  • 00:26:49
    graduates uh still even if we're hiring
  • 00:26:52
    machine learning Engineers or anybody
  • 00:26:54
    else so so this is good um yeah well let
  • 00:26:57
    me know I got a I got a small Army
  • 00:26:58
    they're all coming out but yeah I uh
  • 00:27:00
    I'll let you help me make the pick
  • 00:27:02
    that'll make my interview process easy
  • 00:27:05
    there you go I'm happy to do that yeah I
  • 00:27:07
    think it's I think it's fun and then the
  • 00:27:09
    nice nice part about AR degrees and many
  • 00:27:10
    of the other ones is that it really does
  • 00:27:12
    Center on like this you know on solving
  • 00:27:14
    problems you know so it's like the thing
  • 00:27:16
    that the students we try and give them
  • 00:27:19
    over and over again is this like
  • 00:27:20
    intuition about the way things are going
  • 00:27:23
    to work out like you can develop
  • 00:27:25
    intuition through like practice over
  • 00:27:27
    time
  • 00:27:28
    and so even you know people that are
  • 00:27:30
    just like jumping their data into Google
  • 00:27:32
    spreadsheets or into some type of AI
  • 00:27:34
    cloud system like the hard part is like
  • 00:27:36
    where do those instincts get developed
  • 00:27:38
    that help facilitate like how how they
  • 00:27:40
    should use those results and it ends up
  • 00:27:42
    and even there's like hard quantitative
  • 00:27:45
    people you know typically the people
  • 00:27:47
    that are best at their jobs are the ones
  • 00:27:48
    that are like can see the mistakes
  • 00:27:51
    coming you know and that's because they
  • 00:27:52
    made them before you know and so that
  • 00:27:55
    that part of it is is tough and it takes
  • 00:27:57
    a little while to get there for that but
  • 00:27:59
    usually that's the thing that puts
  • 00:28:01
    people over top is just kind of this
  • 00:28:03
    General experience and that's we're
  • 00:28:04
    hoping to get like with the
  • 00:28:05
    concentration so they can kind of do
  • 00:28:06
    more projects and develop that that
  • 00:28:08
    muscle a bit yeah but do you think like
  • 00:28:11
    a decade a few decades from now let's
  • 00:28:14
    say we will lose that intuition around
  • 00:28:18
    learning around data because now that
  • 00:28:20
    we're developing all these you know
  • 00:28:23
    large machine learning models that can
  • 00:28:25
    do so much of the thinking themselves I
  • 00:28:27
    mean now you
  • 00:28:28
    machines writing code they can write an
  • 00:28:30
    entire piece of software themselves uh
  • 00:28:34
    how how do you think that's going to
  • 00:28:37
    shape us as humans who think and you
  • 00:28:39
    know what does the data science program
  • 00:28:42
    become 20 years from now well 20 years
  • 00:28:45
    is hard because it's like you know I
  • 00:28:47
    don't know you know I the way the
  • 00:28:49
    technology is moving now like no no one
  • 00:28:52
    no one can predict you know like what's
  • 00:28:54
    going to happen you know but there are
  • 00:28:56
    these you know
  • 00:28:58
    you know what we do know like if you
  • 00:29:00
    consider trend lines the way things are
  • 00:29:02
    going like the emphasis in the need and
  • 00:29:04
    the way that companies across the
  • 00:29:06
    Spectrum are using and leveraging data
  • 00:29:08
    as like a just normative strategic
  • 00:29:10
    advantage or strategic like requirement
  • 00:29:13
    keeps going up like if you're not doing
  • 00:29:15
    that then you're Le you're gonna you're
  • 00:29:17
    going to have market failure right I
  • 00:29:18
    mean it just the advantage of being able
  • 00:29:20
    to do that is so important that that
  • 00:29:23
    demand doesn't feel like it's going
  • 00:29:25
    doesn't feel like it's going anywhere
  • 00:29:26
    now I do think like the a automation is
  • 00:29:29
    going to be definitely be like a real
  • 00:29:31
    thing you know and I a lot of the code
  • 00:29:34
    production like again it it's compared
  • 00:29:36
    to and not
  • 00:29:38
    to anyway it's compared to someone
  • 00:29:40
    that's just strictly like you said
  • 00:29:41
    developing pure sweets of pure um
  • 00:29:44
    scripts of software for deployment you
  • 00:29:46
    know data data scientists you kind of
  • 00:29:48
    hacky coders you know because the idea
  • 00:29:50
    is like we're not necessarily developing
  • 00:29:52
    a full St like a full standup piece of
  • 00:29:55
    software or leveraging the tools in
  • 00:29:57
    order to be able to Sol some problem you
  • 00:29:59
    know and to a certain extent if like
  • 00:30:00
    that automation of that like getting to
  • 00:30:02
    the point where you can iterate like
  • 00:30:04
    discover iterate discover iterate
  • 00:30:06
    discover gets easier then that's great
  • 00:30:08
    you know you're you're moving you know
  • 00:30:10
    from a typewriter to a word processor
  • 00:30:12
    like and further up and we already embed
  • 00:30:15
    you know all the you know the the AI
  • 00:30:17
    machines into the you know into into our
  • 00:30:21
    courses you know it's almost impossible
  • 00:30:22
    not to especially if you're using like
  • 00:30:24
    vs code you know where it's just like
  • 00:30:26
    organically embedded
  • 00:30:28
    uh co-pilot is just there and it works
  • 00:30:31
    really well and so the hard part about
  • 00:30:34
    that at least in our experience I think
  • 00:30:36
    I don't know and this is a
  • 00:30:37
    generalization but once you get over a
  • 00:30:38
    certain amount of lines in the code at
  • 00:30:40
    least relative to the things that we're
  • 00:30:42
    doing it the code generator starts to
  • 00:30:44
    kind of break down and so in some ways
  • 00:30:47
    like what it demands that you do is
  • 00:30:50
    break down your problem into smaller
  • 00:30:52
    pieces and then solve each piece and if
  • 00:30:54
    the code generator is creating you code
  • 00:30:56
    for that little piece
  • 00:30:58
    you that's fine I mean I I you know to
  • 00:31:01
    me I don't think it really matters
  • 00:31:02
    you're GNA be exposed to code so much in
  • 00:31:04
    doing it like it'd be a you don't almost
  • 00:31:06
    have to ignore it not to really
  • 00:31:07
    understand it and so in some ways it
  • 00:31:10
    almost encourages students to think more
  • 00:31:13
    strategically about the things they're
  • 00:31:15
    doing as compared to usually the
  • 00:31:17
    instinct is to just jump in and start
  • 00:31:19
    coding right away without really
  • 00:31:21
    thinking about what are the things you
  • 00:31:22
    need to solve in what order and because
  • 00:31:25
    the limitations you know kind of this is
  • 00:31:27
    not always true but because of the
  • 00:31:29
    limitations are some of the things that
  • 00:31:30
    the Google with the co-pilot can do it
  • 00:31:33
    actually sometimes helps you know
  • 00:31:35
    because they have to stop and be like
  • 00:31:36
    okay this is too much script like I
  • 00:31:37
    don't really understand what's going on
  • 00:31:39
    here they have to go back create smaller
  • 00:31:41
    problems and solve those and then move
  • 00:31:43
    to the next which is what exactly what
  • 00:31:45
    we want them to do you know is that
  • 00:31:47
    iteration is the is the point yeah so I
  • 00:31:50
    don't know 20 years from now I can't
  • 00:31:51
    answer I don't know I'll probably you
  • 00:31:53
    know yeah it's at this point it's hard
  • 00:31:55
    to answer what what's going to happen
  • 00:31:56
    three years from now it's that's right
  • 00:31:58
    three months I don't know I don't know
  • 00:32:00
    yeah it's uh it's crazy but I I do agree
  • 00:32:04
    with you I think that 100% like whatever
  • 00:32:07
    we're doing now is not what we'll be
  • 00:32:09
    doing like even a few years from now
  • 00:32:11
    things will change so fast I'm sure of
  • 00:32:13
    it um which is part of the reason I I
  • 00:32:15
    like the field so much because I just
  • 00:32:18
    it's a real challenge I mean if you want
  • 00:32:20
    to stay in it you really do I mean if
  • 00:32:22
    you step out of that stream even for a
  • 00:32:23
    little while like the cost is expensive
  • 00:32:25
    to get back up yeah can get irrelevant
  • 00:32:30
    pretty quickly if you're you go on a
  • 00:32:33
    vacation for a couple of months and
  • 00:32:35
    you're at this point come back sorry
  • 00:32:38
    you've been replaced like oh uh yeah I I
  • 00:32:42
    I like I said I think you know some of
  • 00:32:43
    these things like um experience and
  • 00:32:45
    intuition really do like that becomes
  • 00:32:48
    like but if you're not using the latest
  • 00:32:49
    tools to drive that then that that's
  • 00:32:51
    also a problem bringing all that
  • 00:32:53
    together you speak about blurring the
  • 00:32:55
    lines between institution and the
  • 00:32:58
    workforce yeah right and that's what
  • 00:33:00
    that's what you're aiming to do with the
  • 00:33:02
    program there you're you're looking at
  • 00:33:04
    the latest greatest Technologies you're
  • 00:33:05
    working on llm so uh how successful do
  • 00:33:10
    you think you've you are at being able
  • 00:33:13
    to blur that line because that's
  • 00:33:15
    something that a lot of Institutions are
  • 00:33:17
    in in the hunt for right and it's just
  • 00:33:19
    hard for them to cope up with the pace
  • 00:33:21
    of Technology pace of the industry I I
  • 00:33:24
    will say it's a challenge uh and I think
  • 00:33:26
    there's some cultural stuff around what
  • 00:33:28
    it means to be a faculty member that
  • 00:33:29
    makes it hard um because there there are
  • 00:33:32
    things about uh you know being a kind of
  • 00:33:36
    uh a more seasoned faculty member that
  • 00:33:39
    uh that are normalized you know that
  • 00:33:41
    were in our jobs for a long time you
  • 00:33:43
    know that the experience plays but
  • 00:33:45
    having about again being being able to
  • 00:33:47
    hire so many people in the school we
  • 00:33:49
    have a ton of younger faculty and you
  • 00:33:52
    know the nice part about that is that
  • 00:33:54
    often they they teach up you know if you
  • 00:33:56
    have a community where you know
  • 00:33:58
    someone's been here for a while they're
  • 00:33:59
    really good at a certain thing but like
  • 00:34:00
    you said maybe they stepped out of the
  • 00:34:01
    stream for a little bit like we can
  • 00:34:04
    organically at least right now because
  • 00:34:06
    of the Youth of the school like
  • 00:34:08
    cross-pollinate that in a pretty pretty
  • 00:34:10
    healthy way you know like from my
  • 00:34:12
    experience this is like the best way to
  • 00:34:13
    do your research program to like teach
  • 00:34:15
    in a class like fundamentals and then it
  • 00:34:17
    goes the other way and they're like well
  • 00:34:18
    I'm using this really Innovative way
  • 00:34:20
    this technique this tool this approach
  • 00:34:22
    that maybe you hadn't thought about
  • 00:34:23
    because you've been using you know our
  • 00:34:25
    studio for the last 15 years so like it
  • 00:34:27
    it's nice so it like kind of it evolves
  • 00:34:30
    both ways I we also try and like
  • 00:34:32
    whatever possible like the motto of the
  • 00:34:34
    school is like the school Without Walls
  • 00:34:36
    to try and bring industry in because
  • 00:34:39
    it's not if you accept the reality that
  • 00:34:41
    industry is going to be ahead of you and
  • 00:34:42
    they probably will they almost at least
  • 00:34:45
    in data science I it's see I don't see a
  • 00:34:47
    Frontier where they're not like at least
  • 00:34:50
    equal or running ahead just given how
  • 00:34:52
    fast they can move um and try and
  • 00:34:55
    leverage that not as a weakness but as
  • 00:34:57
    like asset you know that you have this
  • 00:34:59
    population of people out there that you
  • 00:35:00
    know maybe they graduated from UVA
  • 00:35:02
    they're really into data science you
  • 00:35:03
    know they want a space to work they like
  • 00:35:05
    being around students like just own that
  • 00:35:07
    and bring it in I think a lot of schools
  • 00:35:09
    are doing that and we're we're doing
  • 00:35:11
    that too so we you know we have we have
  • 00:35:13
    a I have a a a group of of just like
  • 00:35:15
    pure data scientists like three or four
  • 00:35:17
    years out of school or or less that are
  • 00:35:20
    that are our curriculum Advisory Group
  • 00:35:22
    that advise the undergraduate program so
  • 00:35:24
    they're either graduate from our program
  • 00:35:25
    or just colleagues I've met that are
  • 00:35:28
    that are you know just they're they're
  • 00:35:30
    they're keyboard you know they're
  • 00:35:31
    they're doing the work you know and
  • 00:35:32
    that's the level that you want you know
  • 00:35:34
    to like really understand like what is
  • 00:35:36
    the latest thing and so I bring them in
  • 00:35:39
    every you know every semester and just
  • 00:35:41
    ask questions like where where are we at
  • 00:35:43
    you know is are we keeping up and it
  • 00:35:45
    helps so yeah not not meaning to be
  • 00:35:50
    controversial here but oh yeah would you
  • 00:35:53
    would you rather prefer
  • 00:35:57
    dealing with a human or a
  • 00:36:02
    machine and would would that preference
  • 00:36:05
    change based on certain things you want
  • 00:36:07
    to do achieve yeah uh where you at with
  • 00:36:11
    that that's a good question so it
  • 00:36:14
    certainly depends on the context you
  • 00:36:17
    know that there are there are certain
  • 00:36:18
    things where if you just need an answer
  • 00:36:20
    fast I mean geez so much easier just to
  • 00:36:22
    have a nice machine that could search
  • 00:36:23
    and bring that answer for you instead of
  • 00:36:25
    having to rely on humans to respond or
  • 00:36:27
    pinging being or whatever like social
  • 00:36:29
    dynamics happen to be present right you
  • 00:36:31
    know if things are complicated if there
  • 00:36:33
    if there is history involved and there
  • 00:36:35
    are personalities and all and of course
  • 00:36:37
    it's better you know to have you know
  • 00:36:39
    humans there as much as possible I think
  • 00:36:41
    the complexity of any situation you know
  • 00:36:44
    kind of dictates like where you where
  • 00:36:45
    you fall on that I I will say like we're
  • 00:36:48
    we're like um you know we're working on
  • 00:36:51
    the an intelligent uh agent to embed and
  • 00:36:53
    I've done it in other classes but into
  • 00:36:55
    our machine introduction to machine
  • 00:36:56
    learning class as a as a as almost like
  • 00:37:00
    a just like a you know a study a study
  • 00:37:03
    buddy or like a tutor or like you know a
  • 00:37:05
    reference uh reference agent like all
  • 00:37:07
    those things like not not one that'll
  • 00:37:09
    like generate questions for you and like
  • 00:37:11
    you can you can rep on it but one will
  • 00:37:13
    be like okay that's located in lab seven
  • 00:37:16
    and also by the way it's in chapter
  • 00:37:18
    three of the book that Brian recommended
  • 00:37:19
    like that like tailor tourus a conent in
  • 00:37:21
    our
  • 00:37:22
    class and part of that project um is
  • 00:37:25
    what we're exploring is like embedding
  • 00:37:27
    that agent like how does that change
  • 00:37:29
    like the Dynamics associated with just
  • 00:37:31
    the interaction between like student ta
  • 00:37:35
    and teacher like in this in this what
  • 00:37:39
    you can think of it is like in a
  • 00:37:40
    traditional kind of model where you have
  • 00:37:42
    like your like um you know on Prem like
  • 00:37:47
    bought it system which is like
  • 00:37:49
    Blackboard or canvas or whatever so
  • 00:37:51
    there's that so they interact with that
  • 00:37:53
    and email is part of that so there's a
  • 00:37:54
    technology layer that's on the outside
  • 00:37:56
    but inside the team
  • 00:37:58
    you know there's these natural
  • 00:37:59
    interactions that that typically occur
  • 00:38:01
    between like professor and student and
  • 00:38:04
    ta so what happens when you drop an
  • 00:38:06
    intelligent agent in there like do they
  • 00:38:08
    leverage that more like do they leverage
  • 00:38:10
    it between hours of 10 and you know 3:00
  • 00:38:13
    a.m when I'm not going to answer emails
  • 00:38:15
    right but they can do that because that
  • 00:38:16
    might be when they study you know I
  • 00:38:18
    don't know exactly yeah and then how
  • 00:38:20
    does it work when they're just in a team
  • 00:38:22
    by themselves like work because we do
  • 00:38:23
    everything's Active Learning right so
  • 00:38:24
    they're like a team do they pull up the
  • 00:38:26
    agent as a reference point to go back
  • 00:38:28
    and forth about it so I'm I'm interested
  • 00:38:31
    in the quality of the agent responses
  • 00:38:33
    like that that that's an important part
  • 00:38:34
    of it but the more interesting part I
  • 00:38:36
    think like I think that we can solve
  • 00:38:38
    that problem is like how it actually
  • 00:38:39
    changes the Dynamics in the classroom in
  • 00:38:43
    these like
  • 00:38:44
    microcosm uh little little teams and the
  • 00:38:47
    the communications between the between
  • 00:38:49
    the the different players so I think
  • 00:38:51
    that part of it kind of gets to where
  • 00:38:52
    you're like what they might some
  • 00:38:54
    students that might totally prefer not
  • 00:38:55
    to talk to the teacher for the TA right
  • 00:38:58
    they go straight to the straight to the
  • 00:39:00
    agent why not it might be more efficient
  • 00:39:01
    like you said all the reasons I was you
  • 00:39:03
    know I was very intrigued at the latest
  • 00:39:05
    uh Tesla event where they were showing
  • 00:39:08
    those Robo taxis but then there were a
  • 00:39:10
    bunch of like humanoid machines that
  • 00:39:13
    walked out and I was like bluring the
  • 00:39:17
    lines between a machine and a human it
  • 00:39:19
    was very very interesting for me to see
  • 00:39:22
    I didn't expect that in 2024 but that
  • 00:39:24
    was interesting so you know the pace of
  • 00:39:26
    just even blurring the lines between a
  • 00:39:28
    human and machine like once we get to a
  • 00:39:31
    point where you're talking to a machine
  • 00:39:32
    but you cannot really identify whether
  • 00:39:34
    it's a machine or a human we're kind of
  • 00:39:36
    close to that I feel yeah that's the
  • 00:39:39
    Turning test right you just can't tell
  • 00:39:40
    the difference and so yeah I I uh agree
  • 00:39:44
    yeah so we'll see and then it's if if
  • 00:39:46
    you know or don't know like that's a
  • 00:39:47
    that is a condition but if you know do
  • 00:39:50
    you make the same choices like that's
  • 00:39:52
    also a condition like worth exploring
  • 00:39:54
    like it's interesting to see and there's
  • 00:39:55
    a bunch of you know then it gets into
  • 00:39:57
    you know trustworthy Ai and like how you
  • 00:39:59
    build explainable systems or trustworthy
  • 00:40:01
    systems that people can rely on things
  • 00:40:03
    like that and that that that's a field
  • 00:40:05
    that's going to grow and grow
  • 00:40:06
    exponentially certainly amazing well Dr
  • 00:40:09
    Wright thank you so much for sharing all
  • 00:40:12
    that information I'm hoping this will be
  • 00:40:15
    helpful to students who are thinking
  • 00:40:16
    about the future this will be helpful to
  • 00:40:18
    faculty and and administrators who are
  • 00:40:20
    thinking about you know hey I want I
  • 00:40:22
    want to get a flavor of technology in my
  • 00:40:24
    life CU you know everybody is out there
  • 00:40:26
    doing technology and with all these llms
  • 00:40:28
    and Ai and machine learning I'm kind of
  • 00:40:31
    left out so I think you're you're you're
  • 00:40:33
    like a Guiding Light to to those kind of
  • 00:40:35
    people in terms of you know getting in
  • 00:40:37
    that direction so thank you so much for
  • 00:40:39
    sharing all those insights yeah happy to
  • 00:40:43
    do it it was a lot of fun and uh yeah if
  • 00:40:45
    anybody wants to reach out or has
  • 00:40:46
    questions feel free to just uh shoot me
  • 00:40:49
    an email if they want it's Brian Wright
  • 00:40:51
    virginia.edu I'm always happy to chat or
  • 00:40:54
    talk if people uh want some advice
  • 00:40:56
    amazing thank you so much I will
  • 00:40:58
    actually we'll make sure we add that
  • 00:40:59
    email to our description and people can
  • 00:41:01
    reach out so that will help but but
  • 00:41:04
    thank you and thank you to our listeners
  • 00:41:06
    for joining another episode of edu
  • 00:41:08
    unlocked with me Ashish Fernando and our
  • 00:41:10
    special guest for today Dr Brian Wright
  • 00:41:12
    from UVA thank you everybody
タグ
  • data science
  • education
  • AI
  • career transition
  • curriculum design
  • machine learning
  • large language models
  • academia-industry connections
  • technology in education
  • ethics