In this episode we talk with the Amina Al Sherif who works as an AI/ML specialist at Google Cloud. In this fascinating conversation we talk about Aminas journey starting off as a computational linguist to working as a spy- yes you heard that right a SPY, to being an AI/ML specialist. We also talk about AI ethics, AI literacy, when to use GenAI, when NOT to use GenAI, how she keeps up with everything happening in the GenAI space, we also talk about her next fiction book, her secret to generating creative ideas, all this and more - you don’t want to miss this jam packed episode with Amina. Sit back grab your favorite drink and enjoy the show

Amina’s background and her journey (02:02)

Project Maven (07:10)

AI Ethics (12:36)

How to keep up with everything changing in the GenAI space (18:00)

If she could rewind time - how could this technology be introduced to the public (22:31)

Interesting projects in the AI/ML space (29:21)

What is one thing most misunderstood about how GenAI works (33:33)

Hypothetical predictions of how GenAI could end up being (37:30)

How should one get started off in the AI/ML space (41:01)

Life as an author and managing different responsiblities (45:27)

How to get the creative juices flow (53:22)

How to overcome writers block(55:52)

LinkedIn - Amina Al Sherif

Amina’s books - her own bookstore or Amazon

Full Show notes :

00:00.00 Anand Kumar (host) Hey Amina welcome to the show.

00:02.73 Amina Thank you for having me Anon I appreciate it.

00:08.24 Anand Kumar (host) Awesome! Before we get started I mean I know you have a very extensive and interesting background on you know how you got to where you are today if you can give the listeners a background of your journey that’ll be awesome. Okay.

00:25.12 Amina Yeah, so I assume when you mean where I got to today. It’s you know, being an ai and machine learning lead at Google, so I’ll go on that assumption. So I started my undergrad.

00:38.46 Anand Kumar (host) Yeah.

00:41.82 Amina In computational linguistics but I was really interested in the linguistics part and less interested in the computer science piece of it. Until I joined the military here in the United States and the military basically told me that there was no way I was going to get a security clearance. That I needed to become what I wanted to do which was be a spy. I spoke and speak several languages. So that’s what I wanted to do and you know since I was young who doesn’t want to be a spy and so they told me hey we based on your background I’m originally egyptian.

01:15.50 Anand Kumar (host) Ah.

01:20.20 Amina And based on your background. We don’t think you’re going to get a security clearance and so we’re going to shove you over to the it and networking side of the army because you know that’s where we put the people. So not not a very.

01:36.59 Anand Kumar (host) So.

01:37.14 Amina Glorious turn of events for me. Certainly definitely not where I saw my career going but it’s it’s funny how things work. I end up going to the first iteration of the cyber training at.

01:40.96 Anand Kumar (host) I.

01:53.60 Amina The us army cyber school in Fort Gordon and from there kind of got launched into my my technical career I was dealing with a lot of really large data sets. You know surprise surprise I did end up actually being a spy but I got to do technical things at the same time.

02:08.58 Anand Kumar (host) He.

02:13.28 Amina And I dealt with a lot of data in my job and there wasn’t a whole lot of automation put into place especially where I was working at in the special operations field and so that’s when I started to kind of explore coding basically out of necessity. Not really because I wanted to learn. But I was a 1 person shop and so I had to kind of automate myself to a certain extent or else I wasn’t going to get the job done. Fast forward a couple years from there I actually broke my back on a jump during my military service and ended up in bed for three months

02:38.97 Anand Kumar (host) And.

02:47.42 Anand Kumar (host) Go.

02:52.10 Amina And so of course the first thing that came to mind was what am I going to do with my brain while I sit here like a vegetable in in bed recovering from a broken back and so I did that classic kind of.

02:57.36 Anand Kumar (host) So.

03:06.66 Amina Well I’m going to join a boot camp and learn a little bit more about this coding thing because it seems interesting I could do some fun stuff with this and the next thing you know I graduate from the coding boot camp on a whim apply to Google this was back in 2016 and somehow made it into Google now my journey with machine learning started up at that point because I was launched the first project that I was launched into as a generalist ceo or sales engineer was a project known as project Maven at Google. Um, and publicly because there was a lot of public. A lot of public exposure on that project but the basic premise of it that the public knows and Google knows is that Google was building Ai for drones. For the department of defense here in the United States

00:02.87 Amina Um, okay, so yeah, but the last part of my journey getting to the machine learning piece of um, working machine learning at Google was as I mentioned joining project Maven and suddenly I was launched into the world of computer vision and um.

00:22.82 Amina I’m a fully taught self-taught machine learning engineer at this point. Um, now of course recently with ll lens becoming super popular. Um I think Google kind of folded in on itself and was looking for people who had experience in that field. And being a computational linguist that’s kind of you know one of the few things that we are experts at is understanding how language interacts with um, really electronic systems of any kind. It doesn’t have to be code. Um and it doesn’t have to be machine learning. Um.

00:53.98 Anand Kumar (host) I.

00:56.67 Amina Yeah that’s how I got to where that’s like the the longer maybe version of how I got to to where I am today I’ve been at Google off and on for the last six years and yeah, I’m still in public sector but I work with um, well frankly, all the customers that we have. Under our umbrella so federal defense state and local governments education I kind of work with all those those different kinds of customers.

01:25.32 Anand Kumar (host) Awesome! Ah, thank you for that over you mean and that that is awesome that there’s almost so many things I want to Dr down on and talk about but the first thing I’ll start off with saying I personally I don’t know any I have not had like you know a real life friend as a sp old. It’s great to know somebody in real life who’s who is actually a spy so. That’s that’s very very cool. Ah and and also you mentioned that you started off as a you know, computational linguist and then it’s taken like a whole you know circle and your back and in the llm world right? So that that must be like you know. Pretty interesting for you because it came back a whole circle and now you’re back in sort of where you started off initially.

02:10.45 Amina Yes, yeah, it is funny how everything did kind of circle back to that. Um without me really meaning to but um, yeah, it’s funny. How things come around that way.

02:15.83 Anand Kumar (host) Oh. Awesome! And and you had mentioned about project Maven would you mind is is that something that we can talk a little more in detail about project Maven and what that was.

02:30.19 Amina Yeah I mean from a personal standpoint. Absolutely I can I can talk about my experiences there. There will obviously be a few things that I can’t hear but I’ll go as I’ll go in areas where I can go I Guess let’s let’s just put it that way.

02:46.40 Anand Kumar (host) You? Okay, awesome. Yeah, if you can just give us like a little bit of background on what the project mave is and but you know what should people know about it and what was what happened to project mave now. So.

02:59.64 Amina Um, yeah, so this is all public information. It’s been out in the news with various you know as news goes interpretations of what project Maven was and what Google’s role was in and ah in the project. But um. Basically the premise of the project was Google building machine learning algorithms and models for the department of defense on nonlethal drones that was the platform that Google worked on specifically it was ah it was an interesting computer vision problem in the sense of. You know we were working with platforms that are like zeppelins or blimps right? they operate at really high altitudes and so Google was charged. Ah you know in a strange kind of way of technical scoping charged with detecting very. Small events and small items in kind of a classification context. Um from you know, ah satellite data or overhead data that was ah extremely far away right? So we were asked to detect you know. Hey. Can you show? ah men that might be armed right? They might be carrying guns or rpg launchers or something like that. Um and be able to detect it from from very high up in altitude which as you can see just as a technologist. It’s like the setup of the problem was was it was destined to kind of.

04:28.35 Amina Find us in the butt so to speak in my opinion. Um, but yeah, we were also asked to do things like be able to count the number of kids that came in and out of a school for example and this is you know a very common workflow um that all military services use prior to doing anything.

04:29.57 Anand Kumar (host) No, no, no.

04:49.25 Amina Whether it’s ah, an airstrike or whether it’s a ground raid if it’s you know and and in Afghanistan for example, schools have been used. You know very very prolifically. This is also public. Um by the taliban in the past as places to store weapons or places to put. A lot of precious things under because you know it was known that the United States wouldn’t strike a school full of children and so it was tasks like that so we had to be able to count humans and a human would be like half a pixel right? because these ah these shots that we were getting were were so far away. Um.

05:19.52 Anand Kumar (host) And.

05:25.75 Amina So the the kind of thing that project Maven was known for is bringing the conversation of Ai and ethics to the forefront which is really kind of how a larger company at Google got involved. Um and it really brought the conversation of what. What does ethics and technology mean together when it comes to large technology companies working on projects that might feel or seem to be controversial right? There are several of these kinds of projects that exist today across all vendors.

05:57.59 Anand Kumar (host) And.

05:59.40 Amina But you know very existential questions that were happening at Google like hey do we support the Us government do we work for the Us government. Do we work with entities like ice um or dhs or the border patrol um all those types of questions that were coming up that were were really kind of divining. Google in some kind of way from an ethical and and moral standpoint I think ultimately the decision that was made by Google was that yes you know as a Us -based company. We will work with not just the us Us partment of defense but a lot of other governments to be able to provide.

06:22.15 Anand Kumar (host) And.

06:38.41 Amina Technology and in now you know after project Maven an ethical way and that’s that’s how our Ai ethics principles were born um is a group of people got together at Google following the project Maven incident and formed the ai ethics charter that we now follow today.

06:44.12 Anand Kumar (host) I.

06:56.38 Amina Um, not just for Ai. But I think for for all the technologies collectively that that we implement across the space.

07:01.42 Anand Kumar (host) Interesting, very interesting and and fast forward from that to to to date with the generative ai space I mean I know you’re at sort of the forefront of it I’ve just started to get a hold of the elms and generative ai and I can I can easily say that this is a feel which is. Drastically changing almost everyd other day and you have been at the forefront of it. You have you’re the ai specialist in our organization. How does that translate I know you talked about some of the ethics conversations which was there at project Maven how does that translate to the Generative ai world.

07:22.50 Amina Yes.

07:41.32 Anand Kumar (host) And also how does it? How do you manage everything that’s going around from the genei space.

07:45.85 Amina Yeah, so I’ll tackle the first question first and the the first question’s actually super interesting. There are lots of different ways that I after being through so many Ai ethics just you know that. And felt like rotations in general between project Maven and other things that I’ve been involved with um, initially when this technology started surfacing when openai revealed chat gbt last November can’t believe that was a year ago um

08:15.79 Anand Kumar (host) So.

08:18.29 Amina There was a part of me that was very very angry about the way this technology was being introduced to the masses because the way it was introduced was it really wasn’t introduced at all. It was just kind of thrown out there and people just started consuming it and so you start to see a lot of like. Anywhere from Tiktok videos to Reddit threads of just the general masses in all over the world. Not just in the United States being like what is this is this an intelligent being is this a supercomputer Ai is it. You know that nobody knows any anymore or any better because.

08:37.46 Anand Kumar (host) And.

08:54.85 Amina General Ai literacy in the global population is very concentrated amongst a few people who care about it or typically that’s how it’s been in the past but now what is happening and and as you and I may and may or may not know generative Ai is based on something called transformers.

08:59.17 Anand Kumar (host) And.

09:14.49 Amina And transformers job all this mathematical algorithm is supposed to do is predict the next word or words in a sequence or in a sentence or in a pattern right? So this can also be applied across pictures. So technologies like midjourney and doli. Are are just predicting what the next most likely pixel should be in a picture and so it’s really not selfaware nor self intelligent in any way. It’s quite clunky actually at the end of the day in terms of actually making llms and generative Ai useful for.

09:34.84 Anand Kumar (host) Um.

09:40.90 Anand Kumar (host) Um.

09:52.55 Amina You know large enterprise applications but the masses didn’t understand that right. They didn’t know that all this model really was was a really good prediction machine and so a lot of chaos was happening there and I think that kind of escalated the ethics conversation around generative Ai. But also kind of distorted it in a lot of men in a lot of a lot of different ways right? We have people who are now exclaiming that you know generative Ai is passing the bar exam. Well, of course if you have it memorize all of the content that’s in a bar exam. It’s going to pass it because that’s what it’s designed to do. Machine learning is just mathematical algorithms training over and memorizing patterns that are seen in data. Um, yeah, right? And that’s that’s kind of where we’re at um so that was the first interesting ethical conversation that I think.

10:36.81 Anand Kumar (host) So sort of like a super powerful Auto to complete if you may right.

10:50.45 Amina You know how this technology was ushered into the space I think could have done a little could have been done a little more responsibly but at the same time it might have not made the splash that it did um if it had been done in an explainable way. So that’s you know, ah left for debate and.

10:58.55 Anand Kumar (host) You.

11:07.30 Amina Of course the other ethical conversations around generative Ai that I carry over from my days at you know project Maven is are ll lens and generative ai actually the right solution for a customer’s problem. Right? So since generative ai has become this new sparkly magical dust it feels like customers are coming to me every day and kind of ditching anything that looks like traditional machine learning which like I would not call neural networks traditional at all. There’s still a lot that we don’t know about them. Um. And and wanting to just slap generative ai onto a use case or solution to make sure at the end of the day. It’s a case of fomo right? Nobody wants to feel like they’re missing out. Um, and that causes a lot of ethical issues too. The most recent one I ran into for example was.

11:50.19 Anand Kumar (host) So.

12:00.40 Amina Well is it actually ethical and appropriate use to apply large language models and something like chat Gpt or bard against math problems because neither like large language models in the transformer core transformer architecture is not designed. To do math that is not what it’s actually designed to do as mathematical computations even though the system itself is bay based on matrix multiplication which is ironic. They’re actually not very good at doing math consistently and so.

12:26.98 Anand Kumar (host) So.

12:34.18 Amina I’ve run into this in public sector right? because I’ve got all kinds of customers that either want to do something education related or they want llms to do all of their finance and accounting analytics further to whole departments. So there are some more kind of I guess does the glove fit the hand.

12:50.18 Anand Kumar (host) M.

12:52.29 Amina Ethical questions that I think should be the focus of ethical conversations around generative Ai as opposed to like is the technology inherently evil because it’s actually it’s not it has no intention. It just is a really good prediction machine. Um.

13:05.78 Anand Kumar (host) Um.

13:09.40 Anand Kumar (host) And.

13:10.13 Amina So yeah, and the second question that you mentioned how how do I keep up with it. Um I’ll be completely honest with large language models. This has been kind of the hardest phase for me to to answer that question. So here’s what I generally tell people and engineers. Especially.

13:13.53 Anand Kumar (host) And.

13:28.86 Amina 1 if you think you’re ever going to get ahead of the wave and you work in the technology space. You might not be in the right space. Um, because as you and I both know as engineers um technology changes all the time and if there’s any expectation for 1 single person to have um the the bandwidth. Be able to absorb what’s happening across the entirety of the technology space even a highly specialized space like computational linguistics. My example, it’s not going to happen. So don’t expect it right? But in the meantime what you can do to try to keep up with things is I love my newsletters.

13:52.71 Anand Kumar (host) Oh.

14:07.43 Amina Um I use newsletters a lot both internal like Google and external newsletters to make sure that other people especially researchers or people who specialize in areas that I care about like Ai and machine learning um can filter out what’s important for me.

14:18.33 Anand Kumar (host) And.

14:23.45 Amina It can be a really really long dark deep rabbit hole if you go out there and just start reading research papers on different methodologies and you don’t have the ability to filter them especially in generative Ai right now there’s and the plane is being built as we fly it.

14:40.11 Anand Kumar (host) So.

14:42.46 Amina The people who are building the systems that we are using now like bar like ah like chat Gpt any of them or anything offered on the Enterprise side. Everyone is is actively putting out research and then the machine learning engineers that are building. The product are consuming those research papers the same week. Right? And then trying to go back and evaluate product and see if if these research methodologies fit and so ah, it’s ah it’s an interesting time in technology and everyone should not feel panicked because everyone feels like this of like research like you said is coming out every day.

15:14.40 Anand Kumar (host) Yeah.

15:20.15 Amina And all of us are getting shook every morning when we log in like oh there’s a new retrieval system or now there’s a new theory behind how um to ground ah hallucinations For example, so it’s it’s.

15:22.85 Anand Kumar (host) Um.

15:31.29 Anand Kumar (host) A.

15:34.81 Amina I love my my newsletters and I also put my my logic hat on when it comes to all the new products and research that’s coming out nowadays on generative Ai just to kind of look at all of them with a grain of salt and ah, a healthy dose of skepticism in terms of what works and what doesn’t.

15:48.71 Anand Kumar (host) You.

15:52.68 Amina And what I should pay attention to and what I shouldn’t So hopefully that’s helpful.

15:54.91 Anand Kumar (host) So that is actually very reassuring because I had this inforster syndrome saying that you know whenever I read something about Gennii then there’s a new article which comes out I’m like oh my God I have to catch up to that. So it’s good to know like especially because we represent some of the organizations that created this technology.

16:05.40 Amina Yeah.

16:14.37 Anand Kumar (host) And we go have this conversation with the customer that expected us hey you’re Google you should be knowing this I know we have an army of resources to help but at some level we sort of get like you know an impulse sy I should know this and let me go back and read all the weekends and extended time. So but it’s good to know that you know there’s no way to keep up. Distant this space. Yeah.

16:36.64 Amina Yeah I mean it’s it’s you can try. Um, obviously with all those research papers definitely use some kind of Llm um to summarize early pal right? Yeah, um, but no I mean don’t don’t expect yourself to say even remotely up to date because that’s going to cause a really.

16:41.86 Anand Kumar (host) Yeah, exactly That’s what I was hoping like.

16:54.67 Amina Horrible frankly like cycle mental health-wise. So I I know I’ve spoken about this a little bit and a lot of other people at you know Google and other companies that I have relationships with I mean people that are working in the llm um field right now are are really tired and they’re also really excited.

16:58.85 Anand Kumar (host) Um.

17:13.95 Amina Um, but I think a lot of them are really tired because this past year has just been so fast and if you try to keep up with all of it. You have to realize even hype cycles are marathons. They’re not sprints right? So you you kind of have to pace yourself even when everyone else seems to be losing it around you.

17:14.42 Anand Kumar (host) Um.

17:17.68 Anand Kumar (host) Right.

17:27.10 Anand Kumar (host) Um.

17:33.80 Amina Um, that’s that’s the only way to stay healthy in the field and and sustain right? your work. So.

17:36.64 Anand Kumar (host) Um, green yep totally agree on that I want to go back to what you were saying earlier amina with respect to ah you know how you felt like the way. The technology was the released in normal but 20202022 for chat gpty might have been better like I just wanted to give you my perspective like I was on the other end like I I did not know much about machine learnings or neural networks at that point of time but I was just a avid consumer and I felt this is amazing like this technology.

17:56.79 Amina Yes.

18:09.97 Anand Kumar (host) And that led me to start reading and getting more curious about how this technology works. But I I know you said this might not have been the best of but sort of sort of genuinely curious from your side if you had to rewind the clock and go back to normal 2022 and you had the controls to you know. Make this release in a more controlled way like what would be the ideal scenario that they should have come out and.

18:33.47 Amina Oh gosh I don’t know if I mean the the classic story that I use is you know we came out with the transformers paper which is the attentional is all you need paper back in 2017 right so it’s not new.

18:45.92 Anand Kumar (host) And.

18:50.89 Amina And the research has been steady in buildings since then from the computational linguistics community. Um, there was a googler who um, you know you may or may not remember was ah actually fired from Google. Um.

18:54.52 Anand Kumar (host) I.

19:09.94 Amina Back in 2021 and he was fired because he um mentioned something about you know, hey there’s something that Google’s working on. It’s a sentient being. It’s a sentient chat bot. It’s able to express human emotions and feelings. That’s where we were in 2021 with this technology.

19:19.32 Anand Kumar (host) Oh yeah.

19:29.28 Amina Right? So I imagine open a I and the folks there were in a similar place. Um I don’t know who made the judgment call to finally unleash chat gpt onto the world If if I had been in charge you know which is above my pay grade and I shouldn’t be in charge.

19:34.32 Anand Kumar (host) Um.

19:47.74 Amina But I wasn’t right if I was in charge I think I would have done things in a very similar way to attract Market attention right? So like the attention that was that was attracted by this new tool being released was was really phenomenal right? So the marketing piece was great.

19:48.00 Anand Kumar (host) I Just ah, hypothetically look.

19:58.11 Anand Kumar (host) A right exactly.

20:05.62 Amina But was not great is chat gp ended up getting a bad rap because there were things like the Samsung week and people putting in personal information and you know chat gpt auto training on all of the data that was put into the system. So from a product perspective and a marketing perspective.

20:12.14 Anand Kumar (host) M.

20:19.16 Anand Kumar (host) And.

20:24.47 Amina And think I would have done things very similarly, but what I would have changed were things like 1 educating people and what the system actually is so maybe including something in the ui or the user experience that educates people very briefly about what it is. They’re about to interact with the kind of.

20:26.14 Anand Kumar (host) Yes.

20:43.10 Anand Kumar (host) Um.

20:43.45 Amina You know analogy that I used of you know or actually you brought it up of of these technologies being just really powerful Auto completes something to that effect needed to be communicated to the end users so that they knew that there was no sentience behind the technology that they were using. Um.

20:59.66 Anand Kumar (host) And.

21:02.57 Amina I think that would be a really good disclaimer to put on all Ai enabled products at least for now right? until the general one the general human population becomes used to it just like you know. Industrial revolution was disruptive back then this is going to be disruptive now 10 years from now we’re going to look back on this conversation and go why are we so worried. Um, at least that’s the hope right? and the other thing that I would have done um would be what we we all have done and Google did from the get go. Which is one not train on whatever the users put into this unknown system that they know nothing about so don’t retrain on that data or if you do retrain on that data make it very clear that you do right? So put a disclaimer or something to that effect. But says hey whatever you put in here like it’s going to be part of like anyone else could access this information. Um, so I think you know aside from from the launch and how it happened obviously you’re an you’re an engineer so you get this like we could continue perfecting the technology for another 105 years if we wanted to.

21:50.65 Anand Kumar (host) Um.

21:53.70 Anand Kumar (host) I.

22:09.70 Amina Um, but I think on a consumer level. The technology is ready on the Enterprise level I think we have a little bit of a ways to go to make it enterprise ready. But as a fun consumer tool. It certainly was you know released and and caught attention that way. And think I would have just taken a few extra steps like I mentioned to kind of make sure that there were the right disclaimers on the product to educate end users a little bit more about how to use the tool in a way that wasn’t going to hurt them at the end of the day.

22:27.92 Anand Kumar (host) So.

22:38.60 Anand Kumar (host) Makes sense and essentially you’ saying have more guardrails but the go to market it certain strategy is fine but just move the guardrails perspective and I think it is very similar for me like whenever I got access to this tool back in November it fed like magic right? But the more I read about this. And the more I understand like the transformers paper and how it works behind the scenes once you unveil the curtain. It. It just feels oh. It’s just very fancy auto complete so some of the magic component is like died down but it’s still like amazing how this technology works. Read.

23:17.29 Amina Yeah, agreed it. It definitely still continues to amaze people every day in terms of what transformers are actually capable of doing and I think we’re discovering new capabilities that that architecture introduces um like it like we’ve agreed. On almost a daily basis and what that does is it unlocks some really good use cases for a lot of enterprises that want to use um large language models and generative Ai But what it also does is it creates a very unstable and shaky ground.

23:43.78 Anand Kumar (host) A.

23:50.81 Amina Businesses who do want to integrate that technology because it’s developing and and moving so fast. They really have to stay on top of it if they want to do anything truly unique or truly um, groundbreaking with with any of these generative capabilities that are coming to the table.

24:08.45 Anand Kumar (host) Makes sense and I know you had also mentioned that you know we want to pick and choose use cases not just rub Ai on it for everything but just have it for specific use cases which is you know, sort of tailor made for that particular purpose. Ah what would be like.

24:21.86 Amina Yeah, yeah.

24:25.77 Anand Kumar (host) Some of the interesting projects that you’ve worked on in this space in the generative Ai space something that you know we can disclose publicly any interesting projects that you have worked on.

24:35.51 Amina Yeah, um, so in proper fashion I’m going to go ahead and highlight I guess proper machine learning engineer fashion because and the reason why I say this is historically when I’ve scoped in machine learning projects for customers. Because machine learning has been kind of the bleeding edge of the field of Ai for a while now. What I generally like to do with high risk technologies is 2 things 1 I’d like to advise to start integrating those high risk technologies in areas of the business that. Are not high risk for the business as in customer facing or you know potentially um in an area that might cause significant harm to the business overall um or at an individual level if you’re an app developer. You know, developing an app that that uses llms for maybe things in the background. Um that that are mundane tasks that need to be automated right? Those are the use cases right now that really excite me because it’s frankly where generative ai is the most reliable.

25:34.59 Anand Kumar (host) And.

25:43.86 Anand Kumar (host) And.

25:45.64 Amina At this point so some of the use cases that I’ve seen that have been pretty magical on my corner of the world public sector wise have been things like let’s use generative Ai to go out and crawl government websites with public data and make that data available for our employees to chat with. And be able to extract insights from this is especially useful for basically any government agency you can think of under the sun. But also all of the companies that are commercial that serve any part of the government um area right? So let as an example that i. I’ve interacted with I deal with a lot of education companies and education technology companies. A lot of those companies serve public school districts and so they need to be able to extract information from government websites on a reliable and automated basis in a low friction way to be able to do their jobs. Right? So one use case is hey let’s just go out and systematically index that information and then have an actual quality either chat or search and retrieve conversation over that information. Um, another use case that I really like is using llm um capabilities to. Co-create code. Um, the reason why I like that is obviously code and the code base in general has been a really big part of the training data set or um, a lot of these large language models and because they’ve been trained on so much code.

27:16.96 Anand Kumar (host) I.

27:21.88 Amina And coding is essentially kind of ah a hyper predictive um exercise in of itself even as a coder um llms are really good at generating code. So now if you think about where you can apply that in a lot of Back-office Tasks. You could have llms writing Sql queries for you programmatically. You could have llms um, deploying infrastructure for you programmatically in the form of terraform scripts and kind of automatically generating those you have llms creating websites now these are kind of low friction or or low hanging fruit use cases that are also low risk.

27:58.11 Anand Kumar (host) And.

28:00.14 Amina Are really good areas to start integrating generative Ai without the potential for like the entirety of the business to collapse because you surfaced a chat bot that now turned into you know someone rude or didn’t exactly answer the question that a user needed.

28:07.93 Anand Kumar (host) So.

28:17.37 Amina Or it it had you know a couple of bad interactions with the user right? So those are the use cases that I’m especially interested in now are those kind of boring back-office task use cases that can be automated with a high amount of roi to the business.

28:17.78 Anand Kumar (host) Um.

28:34.31 Amina That might not look sexy but actually does save a lot of time in man hours.

28:39.29 Anand Kumar (host) Agreed yeah I think I’ve had more luck with those kind of use cases. Well so I totally agree on that. Um I knew initially you had said you know llms are not meant to do mathematical calculations. And these are some things that you know the listeners and the general public might not be aware like what are some of the use cases that jenii I know you mentioned some of them which is not good for but what’s 1 thing that people Misunderstand the most about how Gen Ai works and.

29:07.84 Amina Oh gosh I think the thing that people most often Misunderstand and again this is kind of due to the way and the novelty of the way that chat gbt was unveiled to the public is that I can ask this model. Absolutely anything in plain english and I will get back exactly the answer that I want right? that instant gratification mindset keeps coming back to haunt us right? when it comes to unleashing new technologies when you look at the area of prompt engineering.

29:26.62 Anand Kumar (host) So.

29:45.32 Amina And how that area is so robust right now. One can only conclude that large language models as anything beyond like a creative companion or a novelty tool in the consumer space really does need this new type of engineer that understands. Um, language semantics aspects of language combined with someone who understands kind of general programming structure I wouldn’t even say needs to know how to code um to be able to actually get the desired results systematically and programmatically from generative Ai.

30:15.17 Anand Kumar (host) And.

30:25.11 Amina Right? So when you start looking at things like prompt engineering techniques that are popular right now like react like Rag like chain of thought prompting like tree of thought prompting this whole new field has now blossomed and I think that is a good indicator of 2 things 1 that generative ai is not for everyone. It’s not quite at that stage yet. Although I think we’ll get there very soon and that too. Um that generative Ai right now needs a kind of more sophisticated more ah concentrated look in order for it to be applied in the correct. Way. It can’t just be an end user with a question and you know it’s just me and my bot just sitting there talking to each other. So I think that’s probably the biggest misconception about generative Ai right now that’s it’s kind of this ready-made agi fully comprehensive system that can answer all of your questions when in.

31:18.10 Anand Kumar (host) And.

31:23.18 Amina Fact there’s a whole area of engineering that is developed over how to ask the right question of llms right? This is where search was ten years ago when we had people doing googled working right? people who knew how to use that search bar really well frankly, using basic components of how programming languages go together.

31:33.64 Anand Kumar (host) Um.

31:40.79 Anand Kumar (host) So.

31:42.73 Amina And symbols and how to use ah symbols in your query. So I think that’s the biggest probably misconception as as it relates to gen Ai right now.

31:52.18 Anand Kumar (host) Got it yup makes and then I’m also seeing like the importance of prompt engineering like these certain roles which are dedicated for just prompt engineer roles. So which is great that the industry is you know? Also, the awareness is spreadading across and. How to use gennii It’s not really like a ready-made solution for every use case for for consumer playing around that’s fine. But for enterprise customers. There’s like a methodical way to access it with with that said I know you mentioned like as of now this is why this technology is but you also gave the analogy of search how it was. Back in the days when we had to most of the you know the autopro you have to like really be good at some of the search terms and the search queries but it has evolved a lot from then to now how do you think from your perspective Gen Ai would be in the next six months five years and

32:34.91 Amina Yeah.

32:47.26 Anand Kumar (host) Like sort of 10 years down the line and.

32:50.20 Amina Um, well like I don’t have a crystal ball but this is probably the question that keeps me up at night the most um and I’m not you know I’m not the top expert in the field to be able to be making these predictions but some of the ones that I have that I think are a little bit more tactical in nature.

32:51.28 Anand Kumar (host) I.

33:08.80 Amina Some of them are a little bit more strategic I think while prompt engineering is important and it’s a big field right now I think our models are going to eventually get good enough that we won’t need. It’s going to be like search right? We won’t need that high level of um. Prompt engineering to be able to get what we need out of these llm. So I think they’ll become sophisticated enough that they will be able to execute without any need for a specialized human at the other end of them. So that’s the first prediction that I make I think that’ll probably come in the next. Probably six months to a couple years as the models as we figure out as engineers. What the best approach is for making these models understand and comprehend queries better and prompts better. Um.

33:45.76 Anand Kumar (host) M.

33:53.42 Anand Kumar (host) Um.

34:00.26 Amina And figure out more about the architecture of how transformers worked in order to optimize them right? Not to optimize human behavior and interacting with them but the other way around um the other prediction that I think I would make is that large language models.

34:06.10 Anand Kumar (host) No.

34:16.22 Amina And Generative Ai as a whole instead of looking at them as just a singular capability will become kind of a um, permeating platform into every capability that we have whether it’s and every capability that’s technologically delivered. Right? So it could go as far as writing emails for you. It could code for you. It could manage your infrastructure in the background and provisioning of infrastructure all the way to being chatbots all the way to managing your hr handbook right? So like these are tasks. That generative Ai will eventually be able to do kind of end to end in more of a pipeline format and viewed as another tool to add to your toolkit from a platform perspective when you’re evaluating tools platforms cloud providers. You name it. You’re gonna want to see and and I hate to too my own horn. But at Google we’ve launched duet ai into everything that we do right across the board whether it’s workspace whether it’s working in the cloud console whether it’s working in coab um llms are everywhere now and so it’s. Become more of a platform um as opposed to a um, singular capability and I think that’s going to continue as we see growth in like third -party extensions and third -party integrations and rappers and this whole like very ah ah, rapidly growing ecosystem.

35:42.22 Amina Around how and what you can plug llms into to increase productivity.

35:49.71 Anand Kumar (host) It interesting so like a personalized agent for everyone you know, help out in day-to-day after writing emails or you know composing docs etc. Interesting. Um, if if somebody had to start off today in.

35:57.20 Amina Right.

36:05.28 Anand Kumar (host) The you know the machine learning ai space and the generative Ai Space. What would be the advice that you would give to them I Meana like how should they think about like from getting started and I’ll give you my experience like I actually started ah when I started reading about these Llm cit did some of the white papers. But then now I’m also going back to some of the foundationals wherein I’m reading this book called you look like a thing and I love you that book is Amazing. It talks about the fundamentals of machine learning basics of Ai etc I felt like I should have started off with that Book. What is your recommendation from you know. Anybody starting off in this space. What would you recommend as a step bystep approach to get to Me. You are oops.

36:46.37 Amina Yeah, that’s a.

36:48.65 Anand Kumar (host) Right? ready to go.

36:50.15 Anand Kumar (host) And.

36:25.89 Anand Kumar (host) And.

36:54.15 Anand Kumar (host) Um, nice.

36:44.87 Anand Kumar (host) Um.

37:03.55 Anand Kumar (host) Me me. Me right.

37:06.24 Anand Kumar (host) And I like that at the end of the day like you know you said it’s to some extent. It’s the community which is sort of shaping where this technology is going and having going to redit is a great way to like be in touch with the community to see how they’re playing around with the the technology that’s.

37:25.12 Anand Kumar (host) Great idea. Thank you? Um I know you you know you’re like the jei specialist and you have like a really busy job at work but outside of your day job I know you’re an author if you’ve written 2 books or is it 2 books or have you written more books.

37:44.23 Anand Kumar (host) To work. Okay, awesome. So How do you?? How do you?? Ah so can you give us a little bit of you know, background on what your role is outside of your day job and how do you get time to. You know, ah or how do you manage the work in life.

38:03.16 Anand Kumar (host) You.

38:04.79 Anand Kumar (host) And.

36:32.55 Anand Kumar (host) Um.

36:54.78 Anand Kumar (host) Okay.

38:10.31 Anand Kumar (host) And.

38:12.73 Anand Kumar (host) Me.

37:51.96 Anand Kumar (host) I.

38:16.52 Anand Kumar (host) Um.

38:19.78 Anand Kumar (host) Yeah, very to I think I agree to that like the priorityization is sort of key and it’s helped me as well in the past at least ah when it comes to personally myself I feel like the phone is my biggest enemy. And anytime we start using it all my priorities that I had just goes for a tos and now it’s the agenda of you know the companies that designed those apps which makes it super Addictive. So yeah, 100% agree like the prioritization is super important. Um, yeah, So thank you for sharing that I meana.

38:52.82 Anand Kumar (host) When’s your next book coming up and what is it about.

38:53.72 Anand Kumar (host) Move nice.

38:29.88 Anand Kumar (host) Um.

38:49.46 Anand Kumar (host) The.

39:01.30 Anand Kumar (host) Um.

39:00.70 Anand Kumar (host) You see.

39:03.55 Anand Kumar (host) Interesting. Ah, how do you? This is again sort of a follow-up questions I know writing fiction and even writing a book for that matter ah requires some amount of creative thinking right? like you’ll have to ah.

39:16.38 Anand Kumar (host) At Least the way I have seen it when we zone out and we’re not really doing anything. That’s when creative ideas come like how do you get or what point do you get the creative idea like do you do something specific take a walk etc or is it just um. But regular routine routine like what’s the secret to you know Churn out New Creative ideas from your site.

39:36.28 Anand Kumar (host) Are.

39:43.25 Anand Kumar (host) E.

39:58.12 Anand Kumar (host) Um I I.

39:57.30 Anand Kumar (host) I.

39:58.65 Anand Kumar (host) And.

39:56.65 Anand Kumar (host) Um.

39:59.95 Anand Kumar (host) And and when what do you do when you I’m just going ask like when you get a writers block like what is what are some things that you do you just like like take a break from it and then you come back like what’s the we can.

40:13.81 Anand Kumar (host) I.

39:34.77 Anand Kumar (host) And.

39:44.21 Anand Kumar (host) Um.

40:03.92 Anand Kumar (host) You need.

40:15.25 Anand Kumar (host) To name m.

40:20.53 Anand Kumar (host) Yep, that makes a lot of sense I was recently reading something about diffused mode thinking and focus mode thinking like for the most part when we are focused on a certain object and then when you have the writers block you engage in a diffuse most thinking activity which is.

40:35.63 Anand Kumar (host) Sort of like you said you know something with the engaging the hands physical environment rather than a digital or even just you know sitting in a bar listening to other people speak all of those. So yeah, thank you for sharing that awesome. Ah yeah. Once again, thank you so much Amina for taking the time is there anything else that you would want again I’ll go and put the url for the books in the show notes but anything else that you would want our listeners to know before we sign up. Thank.

41:03.50 Anand Kumar (host) Thank you so much. This is a pleasure talking to you I mean I really had a good time. Thank you.



Leave a comment