YT YouTube2Text

YouTube2Text → Transcripts → YouTube transcript (XNqpySV97IU)

YouTube transcript (XNqpySV97IU) — Transcript

9,781 words · 1,459 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:10I'm excited to introduce our speaker
0:13aditia uh so just before we get into the
0:15webinar I wanted to just say a little
0:17bit about adtia um adtia chapali is a
0:21machine learning engineer and product
0:24lead he currently works at Microsoft
0:26where he helps build gen products that
0:28are used by millions glob globally he's
0:30also an adviser to several silicon valy
0:33startups and Fortune 500 companies
0:36aditia lectures on product management
0:38and AI both at Stanford School of
0:40Engineering and Center for professional
0:43development Adida is here as part of a
0:46new course as well that just launched on
0:48July
0:4929th and if you want to learn more about
0:51the course there are links to the course
0:54listed here within the platform so
0:57without further Ado I'm going to turn it
0:59over to to aditia to start our
1:03webinar awesome thank you very much Jess
1:07uh I'm super excited to be here and the
1:10title for today's webinar is identifying
1:14AI opportunities like a strategy for
1:16Market success but really what I want to
1:19do
1:20is bust a lot of myths with you that
1:24people seem to have about Ai and gen
1:27Ai and one of the most common set of
1:30questions that we get around gen is how
1:34do I personally succeed in gen and AI
1:38you know how do I get in and if I'm in
1:39there how do I succeed as a business or
1:42technical professional and we've done a
1:44lot of research on that topic as well
1:46and so I'm super excited to share those
1:48bits of questions with you and obviously
1:51there's a Q&A section where you can post
1:53questions and we'll stop very
1:56regularly uh to get some of those
1:58questions as well as we'll have points
2:01throughout this presentation where I'll
2:04ask you questions just uh to understand
2:07if you are understanding the process so
2:09far Okay cool so uh a little bit about
2:13the information in this webinar it's
2:16taken from the course that Jess
2:18mentioned that me and a Stanford
2:20professional
2:22teach to get the information in this
2:24course we spoke to 50 plus Executives uh
2:28we did a user research survey 300 plus
2:30users and we've put a lot of thought
2:33into making sure that these answers are
2:34correct that they're valid and sort of
2:37backed by a lot of
2:38data so let's dive into it so the one of
2:41the first things I want to talk about is
2:44this
2:45curve
2:47where I want to talk about a phenomenon
2:51with you specifically you can post your
2:55guesses in the chat about what you think
2:58this phenomenon might be
3:00and
3:02so this is
3:04the I'm going to give you the answer the
3:07first version of this answer right now
3:09this is the internet
3:11so initially people thought it's a
3:14bubble where they said you know is it a
3:16bubble or is it going to create real
3:18value there's a lot of discussion and
3:20then people started saying oh we have
3:22other priorities we've built an internal
3:25Tool uh you know to compensate or to
3:27deal with this uh work and then people
3:31said oh you know companies started
3:33getting disrupted new ones started
3:34getting created and
3:37then we're here right now in this new
3:40process and exactly this is Gen this is
3:42basically gen is following the exact
3:45same curve process as the internet
3:47literally what bit by bit people
3:49initially thought it was a bubble and
3:51then they had some discussion around it
3:54now we're past that with J we know it's
3:56clearly going to create a lot of value
3:58but a lot of companies now are dealing
4:00with it by just building internal tools
4:02and not necessarily building user facing
4:04products and those people are losing
4:06those same people who did that with the
4:07internet have lost essentially and the
4:10people who took advantage of it
4:12disrupted those previous companies and
4:15what's more astounding is the internet
4:17went on to create a thousand times more
4:20value Than People estimated in 1999 so
4:24at the peak bubble people thought that
4:27you know this is as big as it's going to
4:29get and now almost let's say 20 25 years
4:33later we have created a thousand times
4:37more value than that and so we expect
4:39that same curve to happen with ji and so
4:42we can use a lot of those same lessons
4:45specifically we're still in the early
4:48days of the internet so because it's a
4:50new platform nobody knows who's going to
4:52win and the tech is improving extremely
4:56rapidly such that users themselves don't
4:59even have solid preferences yet so the
5:01stuff we're going to tell you is very
5:04early research but we have a strong
5:06sense that it's correct but it's super
5:07evolving and the most exciting part
5:09about all this is the game is yours to
5:11capture so I would almost say we're
5:14gonna tell you some secrets in this
5:15webinar because this data has been let
5:18say locked up for people in our course
5:20or or you know webinars or executive
5:23workshops that we've done and now we're
5:24releasing it and people have used these
5:26secrets or these myths to create
5:29millions of dollars of value and so
5:32we're super excited to share that with
5:33you today but that's the premise of this
5:37webinar so the first thing we're going
5:38to dive into is the actual Tech behind
5:42gen at this point people are very
5:44familiar with what gen can do in general
5:47but there's still a few things people
5:49get
5:50wrong number one note that geni or
5:54generative AI is a marketing term it's
5:56not an actual technical term generative
6:00AI is actually composed or refers to
6:02various different types of models like
6:04generative adversarial networks or
6:06Transformer models or variational Auto
6:08encoders basically what that means
6:11is if you're a non-technical person like
6:14you're a business professional don't go
6:16around saying you work in gen we see
6:19this happen all the time uh with
6:21business professionals
6:23especially nobody's a Nobody calls
6:26themselves a gen engineer they call
6:28themselves an ml engineer er a machine
6:30learning engineer so you shouldn't call
6:32yourself a gen business professional
6:35necessarily you should be calling
6:37yourself a machine learning professional
6:39or some sort of technical professional
6:42in some sense or a ml business
6:45professional definitely definitely not a
6:46generative AI business professional
6:48unless you know you're doing it for
6:50marketing sake if you're going to an
6:52engineering team just don't use the word
6:54gen use the word machine learning
6:57okay there's various types of gen
7:01applications and I will be using the
7:03words and terms gen going forward
7:05because obviously we all understand what
7:08it
7:09means the first type is a collaborative
7:12type of generative a application so this
7:15is every idea is Amplified and every
7:18possibility is explored while it while
7:20you work so you have an assistant that
7:22works along beside
7:23you's personalized gender of AI which
7:26everything is created for you so
7:28everything will become a market of one
7:30and then there's proactive generator
7:33doesn't action for you those are the T
7:35three types and why those are the three
7:36types is because if you project all of
7:39these going forward you can see a world
7:42where gen surrounds you in everything
7:44you do so now you open up Tik Tok or
7:47Instagram and you see all these creators
7:48in the future that could potentially be
7:50all done with generative AI
7:52collaborative you potentially have a lot
7:54of team members that work with you in
7:55the future that could be all generative
7:57AI assistants and proactive something is
7:59done for you you have to go out and do a
8:01list of chores that's all done for
8:03you and so there's a bunch of enabling
8:06Technologies behind this which we won't
8:07dive into like text image audio video
8:10but just know that when people talk
8:13about generative AI now people think
8:15about oh I'm gonna do I'm doing like a
8:17writing chatbot or something like this
8:19but the most
8:22significant sort of advancements are
8:25sort of five 10 years ahead of us
8:29um cool so the first paradigm shift I
8:32want to talk you about with the tech
8:34associated with Gen is previously you
8:37needed a lot of data and infrastructure
8:39and algorithms to build your own model
8:41so this is something like a Facebook or
8:43a Tik Tok they had all these data
8:46centers and all these data scientists to
8:47make an amazing recommendation algorithm
8:50now those models come fully functional
8:54out of the box and people don't realize
8:56how drastic of a change this is not only
8:59is generative AI really amazing and
9:01really smart open Ai and all these
9:03people have actually made it public for
9:05everybody that's the equivalent of
9:06Facebook saying you can use a
9:08recommendation algorithm just give us
9:09your content and we'll rank it for you
9:12and that's amazing so that's a second
9:16change that's happened and because of
9:17that
9:20or uh you know in part of this we see a
9:24lot of more opportunities and sort of
9:27investment opportunities open up and
9:29we're going to talk really quickly about
9:31how to access these models and then
9:33we're going to talk about those
9:34investment opportunities in a second and
9:36how you can potentially make a lot of
9:37money and how people have potentially
9:38used this webinar to sort of really
9:40Advance their goals and you know either
9:42make money or get into their own
9:44careers the first one is to call a model
9:47Creator so the first one is you can call
9:49something like an open AI or a deep mind
9:52with an API the second one is you can
9:55host your own open source model and you
9:57know there's plenty of Open Source
9:58models plenty that arrival the more
10:01close sourced versions like llama 3.1
10:05and the amazing part about this is that
10:07these resources are democratized like
10:08never
10:09before you know if you take one thing
10:12away from the session take away that
10:14most people aren't recognizing the
10:16massive paradigm shift that free
10:20intelligence is providing to people
10:22completely free intelligence that you
10:23can host by
10:25yourself and that opens up a ton of new
10:28investment opportunities which which
10:29we're going to talk about in a second
10:31but we're going to take a quick pause
10:33does anybody have any questions so far
10:36uh about the technical section you can
10:39post them in the chat and we'll wait a
10:40few seconds if you
10:49do so ad Dida one question that did come
10:53up um in general that we have is do you
10:57need to know the technical details
10:59generative AI to be a good
11:01leader that's awesome yeah so
11:05um this is an amazing question which
11:09we're actually GNA cover in depth in a
11:11few sections so I'm not going to talk
11:13about right now because
11:17um we are going to talk about it in more
11:21detail in the personal section
11:23specifically in how you can succeed as a
11:25professional in really succeeding in
11:28general AI or in this new post AI World
11:33um Okay cool so um someone else I don't
11:37entirely understand the question of can
11:38you add more details what for staging
11:40open source model means could you
11:42describe that in a little bit if you if
11:44you describe that question a little bit
11:45more I'm happy to answer it um okay cool
11:50uh and for yeang and Yousef we'll get to
11:52those two questions in a bit as
11:54well and just you can Mark those
11:57questions if I don't answer them later
11:58for youf and Yan and Allan them at the
12:00end cool um let's talk really quickly
12:05about some types of gen companies and
12:08specifically we want to talk about where
12:11the most money is and where people are
12:12going to make the most money and so
12:16um there's three types of
12:19major companies that take advantage of
12:21geni the first one was we called takers
12:24they just take the model and they put a
12:27user interface in front of it
12:29this is the best place to start the
12:32second one is a model customizer which
12:35is a customer that takes a model adds
12:37their own data into it and then
12:38customizes the model this is the best
12:41long-term place to be and then there's a
12:45final model Creator who actually goes
12:48out and create the models and we're
12:50gonna talk in a second about how
12:52everybody's wrong about where most of
12:55the money is in these cases but first
12:57we're going to layer these companies in
12:59a second and so
13:03um give me a second okay cool the first
13:06one is you have to to understand where
13:09those companies make a lot where you can
13:11make a lot of money with those companies
13:12or who's going to make a lot of money we
13:14first have to understand the technical
13:16stack of bit and the industry stack so
13:19just know at the bottom there's compute
13:20Hardware these are comput these are
13:21companies like Nvidia that make the
13:23chips Cloud platforms that sit on top
13:26like Google and you know Amazon and and
13:30Azure and then on top of those there's
13:33people who create the models using Azure
13:36and all those data services and then
13:38people who take those and then there's
13:40people who create those models and make
13:42them open source and then people who
13:45customize those open source models or
13:47sort of take them and put them in front
13:49of users that's important to note
13:52because I'm going to ask you a question
13:54in a second and you can post your
13:55answers in the
13:56Q&A um of who do you think you should
14:00invest your money into and we're going
14:02to talk about how most people get this
14:03question wrong but we still want your
14:06answers
14:09so what company would you invest $1
14:12million into mistal AI which is on the
14:16left is a model Creator so they make
14:21models and they've recently raised a lot
14:23of money and they've made Cutting Edge
14:26models and they're doing really well an
14:28amazing
14:30tomson reuter's West law is a legal
14:34database as in it holds a bunch of legal
14:39cases and who won what and how they won
14:43and their results and Associated
14:45information if you had a million and
14:47they're super old company if you had a
14:50million
14:51dollars what company would you invest
14:53into and we're gonna we're going to wait
14:55a few minutes and and get people's
14:57questions and get people's answers
15:00in the
15:11chat okay
15:13so maybe I gave it away too much because
15:16most people are saying Westlaw and that
15:19is correct Westlaw is the one you should
15:21invest in and we're g to talk about why
15:25in a
15:26second uh and
15:29the myth is that model creators will
15:33make more money actually it seems like
15:34it's 50-50 Westra mistol but oh actually
15:37it seems like most people are saying
15:39mistol and it's actually wrong uh the
15:42myth is that model creators will make
15:44all of the money and that's wrong in
15:47fact the reality is models are becoming
15:50customized and so these model creators
15:53are actually sort of competing against
15:56commoditized competing models and
15:58they're sort of they're not struggling
15:59to make money but they don't there's not
16:00a lot lot of money to be made there in
16:04fact um you know we're GNA really talk
16:09about how this
16:11democratization is making it such that
16:13these gen models
16:15are providing these really accessible
16:18Foundation models to app developers such
16:21that these app developers can unlock 10
16:24times more businesses without running
16:25any significant investment in their own
16:27data science centers and and because of
16:29that we're going to talk about how this
16:31affects the profit of each layer and so
16:35we're going to talk about compute
16:36Hardware the hardware folks like Nvidia
16:40are super defensible and they're going
16:42to make a ton of money with that cloud
16:45platforms will probably provide the same
16:48sort of profitability profiles as
16:50they've previously provided with Cloud
16:52so generally so quite high but not
16:55insane as compared to actually model app
17:00users so these takers and these
17:02customizers we were talking about
17:04earlier
17:06specifically these takers and
17:08customizers were finding or actually
17:10making the most amount of money and
17:14we're gonna spend a few seconds
17:16explaining why so
17:20um number
17:23one the biggest myth is
17:27that you these rappers which
17:30are these apps that take chat GPT and L
17:33large language models and put a user
17:36interface in front of them don't make a
17:39lot of money so an example for this
17:40could be perplexity or uh there's
17:44various ones like Julius and things like
17:46this that customize chat GPD for data
17:47scientists or for researchers or whoever
17:50in fact these apps are making a ton of
17:52money and this is the this is why chat
17:54GPT or these llms are only used
17:57regularly by by let's say maximum around
18:00a 100 million people a a month that
18:03means there's still 8.9 billion people
18:06in the world who don't use chat GPT or
18:07these llms regularly which also means
18:11that making chat GPT or these llms
18:13accessible to those 8.9 billion people
18:16is extremely lucrative like extremely
18:19lucrative opportunity and that's what
18:20these rappers are doing and they're
18:21making a ton of money doing that that's
18:24number one number two people have
18:26pointed out that data is
18:29the real Oil we're going to talk about
18:32in a second how that's not entirely true
18:35data is quite significant but what's
18:37even better is another type of
18:40competitive advantage that large
18:42companies have which I won't give away
18:44just yet because we're going to ask it
18:46is a question in a few seconds but if
18:50there's any questions on this we'll
18:51answer this at the end of this section
18:53but I'm going to talk enough that's
18:54enough about the profitability section
18:57I'm going to now ask you a question
18:58which we just talked which I just said I
19:00would ask which is what is the biggest
19:03Moe in generative AI so that could be a
19:06newer or better model which includes
19:08taking a model and making it better with
19:10customized data having more gpus having
19:14better user experience or having better
19:18distribution which means you know
19:20accessing a large number of users at
19:23once so post your answers quickly in the
19:26chat and then we'll talk about um
19:30who's
19:45right okay
19:47nice a lot of good answers coming
19:52through okay the split seems seems to be
19:55mostly between user experience and
19:58distri tion with some people answering
20:01other P other
20:02places oh sorry somebody asked what is a
20:05mode a mode is a competitive Advantage a
20:07mode is something that you know allows
20:09you to differentiate yourself with with
20:11your competitors okay so the answer seem
20:14to be split between user experience and
20:16distribution mostly with some people
20:17saying gpus the people who said
20:19distribution are correct so congrats to
20:22those
20:23folks why that is is
20:26because you know people talk about poen
20:29data being the new oil that used to be
20:32true when you had to make your own ml
20:35models and your own AI models it's not
20:39it was necessary that you needed a lot
20:41of data because you essentially
20:42constructed these models from scratch
20:44but now that's not true anymore you
20:46don't need you don't necessarily need
20:48better models or better data because
20:50these models are so good out of the box
20:53so that's number one data is not
20:55necessarily the biggest killer user
20:58experience
20:59is really valuable if I had to give the
21:00second best answer here it would be for
21:02sure user experience but it's still not
21:05the best because even if you have a
21:07subpar user experience if you have
21:10access to a large number of users that's
21:13actually also fine because our research
21:16has found that people will take even a
21:19slightly worse user experience if the
21:22results help them it doesn't really
21:25matter if the user experience has to be
21:26amazing as long as the results are
21:27somewhat helpful and that's what
21:29distribution offers here and so
21:31distribution is really the key to
21:33winning here and we're going to talk
21:35about how that affects the types of
21:38companies that are going to win and so
21:39when people think about hey making these
21:42you know if you're a startup or if
21:43you're a large company let's say your
21:45startup you know previously potentially
21:48the world the way to win would be hey
21:51how do I like um you know gain the best
21:54model or gain the best data or make the
21:56amazing user experience that people
21:58always come back and now sort of the
22:00game is how do I get this into the hands
22:02of everybody as quickly as possible in
22:04this new market this rest of this 8.9
22:07billion people so that they can start to
22:09use my product that's the that's the new
22:11people are way of
22:12winning Okay cool so let's gonna let's
22:16Bust The Myth versus reality
22:19then the previously like we just talked
22:23about the biggest advantages used to be
22:25talent and infrastructure and model
22:27stuff that big tech companies used to to
22:29have because you needed that to make
22:31your own model now the biggest
22:33advantages are distribution like we
22:36talked about user experience with a
22:37second and then data and so what that
22:40means is if we look at the Moes they've
22:44been flipped completely with Gen so now
22:48the strongest Moe is distribution and
22:50the weakest most is algorithm which is
22:53amazing so you can see here distribution
22:55user experience and then data for the
22:57people who guess user experience and dat
22:58data and
23:00so uh the paradigm shift is you
23:05previously needed a lot of money a lot
23:08of models and thousands of employees to
23:12win and now all you need is just really
23:15good
23:16uiux some unique data and 10 employees
23:19and those three things will enable you
23:21to get good
23:23distribution okay
23:26uh oh
23:30well between these two
23:32companies the answer is already
23:34highlighted here but between these two
23:37companies IBM and JP Morgan if you had
23:41to guess which company was better
23:43positioned to succeed in geni it would
23:46be JP Morgan which is kind of
23:48interesting because IBM has a lot of
23:50talent it has a lot of money it has way
23:51more gpus it has way it has all these
23:54things but JP Morgan has a wider
23:58distribution to actually more end users
24:01and then it can potentially build a
24:02flywheel with the data and the user
24:05experience in that iterative model so JP
24:08Morgan is the better position to win in
24:09generative Ai and in in this case you're
24:12not that I'm giving you any stock advice
24:14or investing advice but most of the
24:17market doesn't understand this and so
24:19most of the market thinks oh IBM is
24:21going to win so whenever they make a
24:23generative AI announcement or whenever
24:24generative AI comes out IBM stock goes
24:26up but in fact it should be JP Morgan
24:29stock that's going up in fact if
24:30anything JP Morgan stock or its
24:32equivalents are undervalued compared to
24:34their potential for generative AI again
24:37this is not any investment advice I'm
24:38just telling you that the M this is a
24:41mistake a lot of people
24:43make Okay
24:46cool so that means just to recap that
24:50the key AI winners used to be big tech
24:54companies because they had a lot of uh
24:56money a lot of models these like very
24:59specialized employees now the key
25:02generative AI winners are actually
25:04non-te companies and that's they
25:07themselves don't realize this and the
25:09mistake that they make is they keep on
25:10building internal tools and things that
25:12don't really take advantage of this new
25:16world and hopefully if you're part of
25:19those companies you'll start realizing
25:20that you are actually better positioned
25:22to succeed than these tech companies
25:24because the real values in integrating
25:27AI to enhance exist existing systems
25:29because you already have a lot of
25:30distribution and you can pump generative
25:33AI experiences through those so they
25:36have a lot of data and they have a large
25:38user
25:40base and so of these three compan of
25:43these four companies let me
25:45explain the last question before we go
25:48to qu before we I take some questions on
25:51who would you think is be best
25:53positioned to succeed in generative AI
25:56chat
25:56base is a company that takes
26:01tppt and puts it on a website that's it
26:04that's all it does and you can ask
26:05questions about the website through that
26:07chat bot coher makes models they make
26:11amazing models like some of the best in
26:14the Enterprise world but they make
26:16models Riz
26:18GPT allows you to test uh essentially
26:22allows you to improve how you talk to
26:25other people in some cases it's for
26:26dating in some cases it's just to have a
26:30conversation with a chat bot uh and
26:33perplexity allows you to search and get
26:37better results with generative AR so
26:39which of these companies you think is
26:40best position to
26:41succeed and people are asking where's
26:43the chat button I believe that's just
26:45your I think the Q&A box is the chat
26:48button so just post your answers in
26:49there of which four uh which of the four
26:52companies you think is best position to
26:54succeed
27:06okay so the answers mostly seem to be
27:07split between chat-based perplexity and
27:10Riz
27:11GPT
27:12[Music]
27:14um and you know what and most most
27:17people are saying
27:19perplexity uh
27:22so the real value is uh nobody said
27:26cohere actually a people are now saying
27:28cohere the only wrong answer is coher
27:31the rest are all well positioned to
27:33succeed because coher and actually we
27:36made this slide a while back and now
27:37coher is actually laying off employees
27:40they're cutting back a lot of deals so
27:41you're actually seeing some of these
27:43cracks happen because and we're not
27:46commenting on a specific company in this
27:48case we're just commenting on the
27:49general
27:50Trend and I this is again not any
27:53investment advice but the rest of these
27:55companies have these ility to put really
27:59good user experiences in front of these
28:00models and are getting a lot of
28:03distribution in this case let me explain
28:06really quickly why all of these could
28:08work well Riz GPT got sold for I think
28:10millions of dollars chat GPT chat base
28:13sorry he's making millions in ARR and
28:16perplexities has I think 100 million
28:17users or something so all of them are
28:19doing quite well I I can't really
28:20compare if any of them are doing better
28:21than the other I think more people are
28:23familiar with perplexity because it's a
28:25b Toc product okay so we're gonna pause
28:29the end of this section and we're going
28:30to take some questions before we go to
28:32personal tips uh
28:35cool great so first question do you
28:40think the current models are good enough
28:42I agree you don't need large data but
28:45models are not specialized to a specific
28:47task and nowadays big companies need to
28:50specialize models for a specific task
28:53for example due to EU act privacy issues
28:56ETC
28:58Okay cool so um this okay so this is a
29:01good question so number one this is the
29:05mistake a lot of people make so they
29:07often think
29:09that they need to do a lot of like they
29:12need to update a lot of their models a
29:14lot with their own data to really
29:16differentiate themselves and research
29:19from our course s basically highlights
29:23that that's not necessarily the case if
29:26you implement even the most basic gen
29:29functionality into your uh app or
29:32website or experience somewhat
29:34seamlessly that itself has a lot of
29:36value let me provide an example there's
29:39a company called
29:40CarMax that in which you can sell and
29:43buy
29:44cars and all they did is they didn't do
29:48a chat bot that's customized to help you
29:51buy cars or take a model and helping you
29:54determine between cars or anything like
29:55that all they did was they sumarized car
29:59reviews that's all they did and it is
30:02one of the most helpful generative AI
30:04features that's been launched and it's
30:06one of the most helpful features that
30:07they themselves have launched and it
30:09there's no customization there needed
30:11and in fact what we found is a lot of
30:14companies get stuck in this
30:15customization refine tuning process when
30:17it's not really necessary the best
30:19companies we found succeed when they
30:22distribute just a model out of the box
30:25find a user scenario that somewhat works
30:27and then based on those signals they
30:28then start to fine-tune it afterwards
30:30and in terms of fine-tuning models being
30:32better than General Foundation models uh
30:36that is technically true but what we've
30:38also found from our research
30:40is um in most cases even if you find
30:43tuna model with a lot of data the next
30:45iteration the next Frontier iteration of
30:48the models often beats fine-tuning even
30:51the fine tune models for your specific
30:52use case and except in really specific
30:56Parts where you have really data like
30:58Medic in the medical field or legal
31:00field um but yeah Okay cool so next
31:03question
31:04Jess yes so do tech companies like
31:08Google Apple that have large
31:10distribution networks win as well would
31:13they win as
31:14well yes so this is again again a key
31:17question so this they would win so
31:20number one they obviously have a large
31:22amount of distribution I work at
31:24Microsoft I understand that you know we
31:26have access to a billion users you know
31:28we when we build products we can
31:30immediately publish them to a large
31:33audience but this is the advantage that
31:36non-te companies
31:39have all these tech companies can't
31:41train to talk when we talk about you
31:43know really specific data all these uh
31:47these banking legal healthc care
31:50companies and all these places have all
31:52this data and access to a set of users
31:54that we these tech companies don't have
31:56access to necessarily these tech
31:58companies don't have banking products or
32:00medical products or things like this and
32:02so these companies have access to
32:05different users who use their products
32:08for a different use case and th you can
32:09almost think of them as a different set
32:11of users have access to a lot more
32:13unique data and a different user
32:16experience and a different UI
32:18so you know these tech companies are
32:21positioned to win but these non-te
32:22companies are positioned to win like
32:24never before um and that's what I would
32:26say and in fact if they execute really
32:29well like if JP Moran really executes
32:32really well they could potentially even
32:34perform better in generator VII than a
32:36lot of these standard tech companies
32:37because gener works best when there's
32:39other context involved like banking
32:41context or medical context or whatever
32:47Jess so um how can um startup companies
32:52leverage this and how can they find the
32:53right opportunities in the space because
32:55we I know you were talking a lot about
32:57bigger companies so uh the questions
32:59around like startup companies and
33:01leverage leveraging yeah so um this is a
33:05great question because startups are also
33:08extremely well positioned to win in this
33:09space again you know people say hey like
33:11you know if distribution is how you win
33:14then the large companies have everything
33:15it takes to win but the real issue here
33:20is these large companies aren't moving
33:22on this opportunity some tech companies
33:24are obviously but especially these
33:26non-tech companies they're moving super
33:28slowly like extremely slowly and so what
33:32you can do as a startup is Target these
33:35companies these industries like law
33:38Finance health care that have these sort
33:40of unique modes or these unique sets of
33:42data and build really good experiences
33:46or build sort of uh experiences that
33:49either help those professionals or help
33:51those end companies a company that's
33:52gaining a lot of hype is a company
33:54called Harvey which does this for in the
33:56legal profession where they help lawyers
33:59do a sort of customized chat GPT for Law
34:04and you know there's plenty of versions
34:06of these in healthc care and all these
34:08places basically the advantage for
34:10startups is there's tons of
34:12opportunities for these non tech
34:13companies they're not moving at all
34:16they're super slow and so as a startup
34:19you can quickly move and take advantage
34:21of it okay like for example let me let
34:24me tell you right now Westlaw has
34:26amazing dat data amazing access to a
34:29large number of users is doing nothing
34:31with g at least nothing substantial
34:33Harvey's coming in complete startup and
34:36they're sort of completely disrupting
34:37the space and they're started by some
34:39small 20 year olds like some young 20
34:40year olds Okay cool so we'll move on any
34:43other questions we'll answer at the end
34:45uh to because we want we have a lot to
34:47get through okay cool we're going to
34:49talk about personal tips and this is a
34:52bit of a jump because so far we've
34:53talked about you know where do you
34:54succeed as a company where is the most
34:57money
34:59but I want to take a bit of an aside
35:02because this is the most common question
35:03we get from
35:05professionals how do I succeed what do I
35:08need to do and especially we get this
35:11from non-technical professionals so we
35:12get this from more business-oriented
35:14professionals and so I want to talk
35:17through you know how we've seen people
35:18succeed and the curriculum what we're
35:20going to put out has helped people
35:24increase their compensation or their
35:25success significantly you know I think
35:27speak to one person who doubled their
35:30income to $500,000 or more than double
35:33their income significant increased their
35:34income to $500,000 because they sort of
35:37went down and took this curriculum quite
35:38seriously so you
35:41know
35:42that let's dive in the first change is
35:47non-technical people used to not be able
35:50to build because you had to learn how to
35:52code and do all the system architecture
35:53and all these things now non-technical
35:57people can build and I'll explain why
35:59and there's different sort of there's a
36:02beginner to advanc level of
36:04building but let me explain how your job
36:06as a non-technical person can
36:09change previously let's say you're a PM
36:12and you don't necessarily have to be a
36:13PM you could be any sort of business
36:16professional you can start to act as a
36:19data scientist way more than before
36:21because now what a data scientist does
36:23or an applied scientist is they don't
36:25make new models what they do is they
36:27give chat GP these llms new data and ask
36:30it to you know be fine-tuned essentially
36:33and it's as simple as uploading an Excel
36:35file and or they tune prompts so they
36:41they basically fiddle with the English
36:43language to make sure that they get the
36:45results that they
36:47want and it's an reasonably easy skill
36:49to contribute and if you pick it up and
36:51you can start going to these engineering
36:53teams or going to these technical teams
36:54or start doing it yourself you can start
36:56saying hey I want to build this app
36:58where I have this idea and look chat PT
37:00does it super well let me give you an
37:03example we worked with a banker who or
37:07we didn't necessarily work with this
37:08Banker but we worked with this bank and
37:11the banker in some Regional Branch put
37:14in a customers information to for a loan
37:17approval into chat GPT it was a company
37:19approved instance
37:22and chat GPD provided a great answer as
37:24to why this person should get the
37:26approval and they took that version and
37:30they they they provided the approval to
37:32the customer for the loan and then you
37:33basically use chat GPT for the rationale
37:36and it was so helpful that the company
37:39uh sort of approved this use case for
37:41everybody to use and now this Banker is
37:44actually leading this initiative across
37:46the bank on launching generative AI so
37:49their sort of career rent from being a
37:51regional sort of rank representative to
37:54now leading this companywide technical
37:58initiative and it's all because they
37:59were just fiddling with prompts and
38:01there's such a demand from these
38:03companies that even some small efforts
38:06really are appreciated you know we talk
38:08to leaders all the time who say you know
38:10our employees don't do enough of this
38:11and we want to see more and so another
38:15thing is if you're a business
38:17professional one of the things you have
38:18to go ask is you know hey can is this
38:21even feasible you know if I want to
38:22build an app how long would it take you
38:25know let's like you know how many
38:27Engineers would we need and you have to
38:29consider all of those estimates now you
38:30don't need to do any of that now you can
38:32go to chat GP and get the answer
38:34yourself you can just say hey like can
38:35you do this if so like you know what do
38:38you need all these things and it
38:39provides the answer for you we'll talk
38:41about this in a little bit more
38:42technical detail in a second or in a few
38:46slides and uh yeah and we so we'll talk
38:50about it in a little bit more detail but
38:52if you're in a tech adjacent role how
38:55you communicate business requirements
38:56completely Chang changes now you can
38:59come come in with examples of prompts
39:01and output at content you can be the
39:02leader in how generative AI limits or
39:05expands your scenario so you can say hey
39:06you know what like you can be the leader
39:08in your
39:09organization for what gender of AI can
39:12or can't do and how feasible it is to
39:15get it to do a certain piece of the app
39:19then you're GNA research data and
39:21privacy requirements beforeand we're
39:22going to talk about it in a few seconds
39:24how more specifically about how you can
39:26do these things specific specifically to
39:28advance your
39:29career so what can a PM do and in this
39:33case a PM is a non-technical person
39:36Bally a business professional we went
39:38out and asked 50 Executives and product
39:42leaders uh basically Business Leaders
39:46about how in this new world a product
39:51manager or a business professional
39:53specifically can expand their respect
39:55and legitimacy and grow the
39:58organization the first thing is they
40:00need to understand the tech in depth and
40:02we're going to talk about it in a second
40:04what they mean by that and how much Tech
40:05you need to understand the second thing
40:07is understanding product Vision the
40:09third thing is can or does a PR and a PR
40:12is a poll request so if you can start to
40:14code even a little bit even with
40:16prompting that's a earns a lot of
40:20legitimacy and then the fifth one is
40:22gathering and defending requ or the
40:23fourth one is gathering defending
40:24requirements we did this survey three
40:26years ago ago or we've been doing this
40:29survey consistently this year
40:32understanding that Tech in depth was by
40:34far the biggest job so how important
40:37it's it used to not be super important
40:39to understand the tech now it's
40:41extremely important leaders are really
40:44searching for business professionals who
40:47understand a lot more of the tech uh and
40:49you know that's becoming increasingly
40:52important and so and it's in fact the
40:54most important thing so how do the best
40:56ml PM so the ml business professionals
40:59grow and succeed obviously the first
41:02advice was join a fast growing company
41:04the second advice is become
41:06technical then it's become a domain
41:09expert and then it's build more side
41:11projects okay so basically there's three
41:15paths to get into generative AI get
41:18technical get Niche domain experience
41:22and ideally you do both of the three the
41:26most common question then get is you
41:28know which one should I do should I go
41:30out and get domain experience like learn
41:31about a specific industry or should I
41:33get technical and most often the advice
41:35we give for our business professionals
41:37and the advice we see works is get
41:41technical and let me spend a few minutes
41:45explaining why that
41:47is most business professionals already
41:50understand a little bit about their
41:51domain a little bit about the industry
41:53like even if you're an accountant let's
41:55say you understand how accounting works
41:57you don't necessarily understand how a
41:59specific industry works but you
42:00understand how accounting works if
42:01you're in manufacturing obviously you
42:02understand how manufacturing
42:04works and so what we
42:08find uh is
42:11that a lot of people opt for getting
42:15more deeper domain
42:17expertise and number one when we see get
42:19deeper domain expertise here it's still
42:21with a mix of generative AI so you still
42:23have to say hey I I I know manufacturing
42:28really well but I know manufacturing the
42:29context of AI so I can tell you really
42:32specifically where AI is the most
42:35helpful what use cases have been tried
42:37before have not been tried where has AI
42:39been successful where has AI not been
42:41successful where are some Primal from
42:43from initial customers I can get to try
42:46out some initial use cases for
42:47generative AI in in a specific
42:51industry and because a lot of people opt
42:54for this domain expertise route it's
42:56actually a really competitive because
42:58some people have been in this industry
43:00for 30 40 years and they know so much
43:02they know everybody in this space it's
43:04really hard to get competitive you have
43:06to know an extreme level of knowledge on
43:09the technical side a lot of business
43:11professionals shy away from it they say
43:13hey I don't know how to code I don't
43:14really understand that world and we find
43:17that if those people try to get
43:19technical and they get more technical
43:22they sort of become unicorns because
43:24they are one of the few in the world who
43:26can bridge the gap between the technical
43:28side and the industry
43:30side and I'm going to explain in a
43:32second what get technical means because
43:34there's there's beginner intermediate
43:35and advanced but the path that is more
43:39valuable we often find is get technical
43:41the ideal is you have whatever 10 20
43:44years of domain expertise and you get
43:46super technical then you're a unicorn
43:48you know these are the types of people
43:50that we've seen make $500,000 a million
43:52dollars in a few instances because
43:55startups and Industry leaders and
43:56companies
43:57pay so much money for this
44:00advice and I can explain what examples
44:03of advice we mean by that but okay let's
44:05go talk about how to get technical so
44:09there's three stages like I said
44:10beginner intermediate and
44:12advanced if you're in this session You'
44:15already probably passed the beginner
44:18section of how to get
44:20technical because in specifically
44:22related to generative AI you just have
44:24to understand what are the rough tools
44:27what are the rough implications what can
44:29it roughly do at a high level and that's
44:31pretty much it the intermediate advances
44:34really where most of the money starts to
44:36come in uh for a lot of these people
44:38especially at the advanced
44:40level so at an intermediate level what
44:44we mean by how to get technical is you
44:45get extremely good at prompting and this
44:48takes about four to six weeks we've
44:50found and this is like number one the
44:53basic level is you know understanding
44:55how to use chat PT at a pretty deep Lev
44:57level intermediate is using system level
44:59prompting and multi- prompt examples so
45:02really understanding how to coach it to
45:04get a set of results that are really
45:07valuable and you you and regular and
45:10consistently what you do with Advanced
45:13is you get achieve consistent results
45:15with multi-channel prompting we'll
45:17explain what that means in a second
45:19but uh think using things like Json
45:22formatting and Checker llms let me give
45:23you an example like let's say the llm
45:26like Chach PT you're say you're using
45:28doesn't give you the right result the
45:30first time you say hey can you give me a
45:32a a
45:33workout uh regime and it gives you a
45:37like a sort of wrong workout regime you
45:39have another llm check the response with
45:43the specific things you're looking for
45:44like saying hey I told them it had back
45:47pain I had back pain is it recommending
45:49squats to me because if so it's a bad
45:51workout program so you know redo it so
45:54you have a checker llm set up that
45:56checks that for LM and knowing how to
45:59understand Json formatting and
46:00structured outputs from opening and all
46:02these places and doing things like Chain
46:04of Thought which is like prompting llms
46:06to talk about how they get their
46:07reasoning right and you know we won't
46:09talk about it in more detail but that's
46:10the advanced stage and if you really
46:13want to understand how these companies
46:16are making so much money this is where
46:18most of it is and especially in these
46:20rapper ones and
46:23so this is where if you also learn this
46:26you you would also make you would also
46:28start to access those sort of Worlds the
46:31second one that's important to
46:32understand is data boundary and
46:33Associated systems architecture so you
46:37know really quickly and we won't spend
46:39too much time on this because we have so
46:41many other things to cover this is just
46:43understanding how you can call gender of
46:44AI in your own company or personal
46:47boundaries for example we work with so
46:49many companies who think that they can't
46:52use gender of AI or they can't use these
46:54closed models from open Ai and all these
46:56places because Oh They'll break the EU
46:58act or they'll break these privacy acts
47:00and that's just not true all of these
47:02Cloud companies if they work on cloud
47:05run these models Within These companies
47:07data boundaries they just have to call
47:09it within their data boundary API but
47:11it's
47:12astounding how many companies don't know
47:15this
47:16like you know I think we let's say if we
47:18work like 99% of the people we work with
47:20don't know this which is you know
47:23astounding and so they go through all of
47:26this infrastructure sure to host their
47:27own model but they just need to know
47:29this so if you and the few business
47:31professionals who know this provide so
47:33much value because they're saving
47:34millions and thus they're rewarded with
47:37an Associated compensation package so
47:39just knowing data boundary and
47:40Associated systems architecture how to
47:42call generative AI in your own beta
47:44boundaries understanding data flow
47:45diagrams understanding Storage storage
47:48systems is really helpful Advanced is
47:52really where we get to those really big
47:54numbers so there's two paths you can
47:57take if you're a big or or worker or we
48:00find this also works with technical
48:01adjacent workers dive into systems
48:04architecture really deeply so what you
48:07want to be able to do is you need to
48:09practice communicating these technical
48:10Concepts to non-technical and Technical
48:13people really deeply and what that means
48:15is you need to understand like what is
48:18you know eup what is eup like what is
48:21personal information what are things
48:23like caching what are things like stuff
48:25like this so you can dive into systems
48:28architecture you don't necessarily have
48:29to code and where to put in generative
48:31AI super valuable these people are paid
48:34an incredible amount of money for solo
48:37workers or non- Tech adjacent folks
48:38maybe you're not in the discussions with
48:40those systems Engineers all these
48:42people these people make a lot of money
48:44by using low code or no code tools with
48:48gender VII capabilities to either
48:50automate parts of their jobs or start to
48:53build so some of these people we've seen
48:56learn to build apps without ever
48:58actually learning how to code they don't
48:59even understand what chat GPT outputs
49:01they just prompt chat GPT or these llms
49:04to give them code or give them an app
49:06and copy paste that code into you know
49:09an app or some sort of browser and have
49:11it work and so these people also make a
49:14lot of money because you know we've seen
49:15for example a nurse who made a patient
49:18intake tool not by learning how to code
49:21by just prompting chat gbt to make that
49:24app and then sort of that app made a
49:26significant amount of money okay so
49:29we're going to take a quick pause on
49:30that personal section uh just any
49:34questions yeah so we'll do one question
49:36here because I know we're getting closer
49:38toward the time of the top of the hour
49:41so we'll do one question and then move
49:42forward so we have time a little bit at
49:44the end for some more Q&A so one
49:47question is I see we're getting a lot of
49:49questions from people with specific
49:51problems they want to solve with
49:52generative AI so what would you
49:54recommend as some good first steps
49:57and how should they evaluate the
49:58opportunity and decide about the best
50:00next steps for their pain points and
50:02ideas you know what this is great
50:05because it works well into our later
50:07sections of the most common mistakes
50:08people make and you know how you should
50:10go about evaluating what is really a
50:12good idea or bad idea so this works well
50:14so let's Dive Right In um so the SE
50:17second section should should answer this
50:20you know what really quickly we're going
50:22to cover how generative AI models
50:26business models are different and this
50:27is the first thing people sort of forget
50:29we're going to go through it really
50:30quickly um before we talk about how to
50:33evaluate gender of idas because number
50:36one most people the thing people get
50:38wrong is they think of gender of AI like
50:40a typical subscription model where they
50:43say hey there's a typical subscription
50:45price there's a total gen that it's a
50:48typical subscription price and typically
50:50software would always operate within
50:53that sort of subscription price like if
50:55you charge $ eight a user you typically
50:57wouldn't incur a cost every time
50:59somebody uses that new product but with
51:02generative AI that cost goes up with
51:06usage and so that means and it often
51:08looks like this where generative AI goes
51:11up exponentially as usage goes up so
51:14often a lot of SAS or softwares or
51:17service business models don't really end
51:19up working out so that's why you'll see
51:21companies potentially like perplexity
51:23that actually lose a lot of money
51:25because even though they charge users
51:27for their usage especially if you go to
51:29the Premium plan they quickly start to
51:32break that cost period pretty soon so
51:36just remember that as you start to
51:37investigate these ideas know that a lot
51:40of the dollar per seit uh or dollar per
51:43person or dollar per month business
51:45models are breaking with G of AI
51:48okay so we're going to talk about now
51:51answering the question prioritizing
51:53generative AI opportunities how should
51:54you think about
51:57uh how which generative AI product or
52:00project to pick to pick and we're going
52:01to give you some some quick
52:04tidbits so number one let me give you
52:08three of the most common options people
52:11come to us with build an internal facing
52:14tool because they want to build up the
52:15expertise and capabilities while helping
52:17increase efficiency
52:19internally a user facing feature or
52:22buying a gen chatbot software or some
52:25other software in white lab it most
52:27people build the internal facing tool
52:30and they're wrong to do it what you
52:32should be be building always is a user
52:35facing feature almost never an internal
52:38tool you know it's just not recommended
52:42and why that and we're going to talk
52:44really quickly about how common this
52:46problem is you know we did an analysis
52:48of the people who've implemented
52:49generative AI product products and
52:53almost everybody either does a chatbot
52:55or just internal tool 95% of people do
53:00an uh chatbot or internal tool so that's
53:0460 plus 35% and when we ask people you
53:08know how useful is your internal tool
53:1175% said not useful 25% said somewhat
53:15useful so basically 20 95% of people
53:17don't find it useful so if you're
53:19building an internal chatbot don't do it
53:21or internal tool don't do it it doesn't
53:24help you build the skills that you need
53:27to build the skills that you need let's
53:29just walk through this really
53:30quickly what you should do is keep
53:33whatever feature you're doing in private
53:35preview avoid chat Bots or free form
53:38interactions because it allows people to
53:41put in inappropriate information and and
53:43potentially unethical
53:45pieces and you get into a lot of legal
53:48trouble if you give the wrong answer
53:50always do in product
53:53features and just test regularly and
53:56what does allows you to do is it allows
53:57you to build the capabilities you need
54:00to build great user facing features and
54:02the research from our course highlights
54:03that companies who focus
54:07on user facing
54:13featur so that's number one
54:16um but the basically don't build an
54:19internal tool if you're picking a gender
54:21to a opportunity almost anything or a
54:25chat box also doesn't really work super
54:28well though both of those are pretty
54:30tough to iterate and work with but if
54:33you focus on spec almost any other
54:36problem it works out well and we're
54:38going to quickly talk about how to
54:40implement those really quickly and then
54:41we'll take some
54:44questions
54:46um the thing that most people get wrong
54:49is when they go out and create these
54:51generative AI products they create
54:55products for creation so they'll say hey
54:59a product that helps people write emails
55:00and essays within just a few prompts is
55:03what people love versus a product that
55:05helps people read documents and emails
55:07easily and summarizes key
55:09Concepts but research from our course
55:12shows that users love the second option
55:14and they hate the first
55:17option so when you're building a tool or
55:20a generative AI feature prioritize
55:22content creation or sorry do not
55:25prioritize content content creation
55:26that's not recommended what you should
55:28prioritize is content consumption which
55:30means making it easier for people to
55:33consume content and understand things
55:35better so and why we think that people
55:38often do the content creation side is
55:42because the first wave of generative AI
55:44products was this sort of naive back and
55:46forth talking to chat GPT you know can I
55:49get an answer can I get this could I get
55:51you know could you answer this question
55:52for me a request response setup but
55:55users are getting tired of that in fact
55:59what people want now is they want to
56:00take a lot of information and just take
56:02the key insights out of it through gener
56:05generative AI in fact research from our
56:08course shows that content consumption is
56:11on average useful to very useful on a
56:14scale of one to four it's three and a
56:16half on how useful it was content
56:19creation was just a one or two so people
56:24love content consumption for gener of AI
56:27they hate content creation so let me
56:29give you some
56:31examples like instead of creating a
56:33legal document or helping having lawyers
56:35create legal documents you should
56:36analyze historical case outcomes to
56:39advise on legal strategy so instead of
56:41creating personalized documents or or
56:44proposals for potential clients you
56:46should leverage market trends and sort
56:48of refine sales pitches and you know
56:51instead of generating product specs or
56:53generating product emails aggregate user
56:55feed back and tell product people like
56:58what they should be building next we
57:00won't go through too many of these
57:01examples but there's they're here for
57:04you you know some of the best generative
57:05AI examples we've seen companies that
57:07work really well are companies that
57:10listen to sales calls or customer
57:13service calls and tell people how they
57:15can perform better which is really
57:17exciting versus for example creating a
57:20customer service agent
57:23okay so um we won't walk through more we
57:27won't walk through all these examples
57:28the final tip we're the second final tip
57:30is don't build a chatbot instead build
57:34AI features natively into a product
57:37users hate building it using or working
57:40with a chatbot you know most people
57:43aren't willing to use a chatbot almost
57:4699% of people are not willing or only
57:48somewhat willing to use a chatbot and if
57:51they do 80% find it not useful and this
57:53is the modern version of gen chatbot
57:57versus if you build a gen feature into a
58:00product 90% are pretty excited to use
58:03that product and almost 55% find it
58:07useful so this is you know 10 times more
58:10useful for
58:11users Okay cool so we definitely answer
58:13that question in a little Speedy format
58:15because we want to leave time for some
58:17questions at the end so
58:19Jess yeah so I think because we are a
58:22minute away from ending at the top of
58:24the hour um unfortunately we don't have
58:27the time right now for some more
58:28questions but um is there any tips or
58:32takeaways that you want to share before
58:34we end the webinar with
58:37folks uh nope I think that's that's if
58:40if you want any if you want any of your
58:41questions answer just message me on
58:43LinkedIn and I'm happy to answer them
58:45there just one
58:47onone great well thank you so much for
58:51joining us everyone thank you aditia um
58:53and just real quick as a reminder our
58:55new course Adida has launched and if
58:57you're interested in learning more
58:59please refer to the link here um on the
59:02platform and feel free also to reach out
59:04to our team via email and we also have
59:07some other generative AI courses that
59:09are launching and we can help you choose
59:11from some of those courses as well so as
59:13mentioned the recording of the webinar
59:15will be sent to everyone within the week
59:17thank you so much everyone have a
59:19wonderful rest of your day

About this transcript

This page contains the full transcript of YouTube transcript (XNqpySV97IU) , generated from the public captions YouTube serves with the video. The transcript has 9,781 words across 1,459 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.