Sami: Welcome, Sumant, and thank you for taking the time with us, tonight.
Sami: I know it's pretty late where you are.
Sumant: Thank you, Sammy. And it's a pleasure talking to you always.
Sumant: we get to talk a lot of interesting stuff. And, I I am sure it's going far.
Sami: Yeah. So one of the the things, you know, as an introduction, because if you look at your background, on LinkedIn, it's it's quite fascinating.
Sami: And you have a long history in machine learning and AI.
Sami: So what what triggered, you way back then to take this career path.
Sami: So I think it's quite interesting for for everybody.
Sumant: wow. so, So when I started my PhD, I would rather go back a little earlier.
Sumant: So I was working in the embedded system space, and I was quite happy with the OC programming and all those kind of things as well.
Sumant: but I have a deep interest in molecular languages, low resource languages.
Sumant: And I wanted to do something in that area.
Sumant: That's when I got to know an opportunity that, within, tripletiv Bangalore, which where they had completed my PhD, that they have a government, project where they're going to create a knowledge base of oracular languages, and, that would have a large number of AI challenges as well.
Sumant: And we, can actually join them to build some of these things.
Sumant: some of the solve some of the challenges. so I quit my job.
Sumant: Joined them as a a project coordinator. I took care of both content as well as a technology part.
Sumant: worked there for 6 years, but in in the same time, I also validated my MSE research.
Sumant: they found it to be good enough that I could they could can ask they asked me to convert that into a PhD.
Sumant: So I I went into a PhD and the, it flew.
Sumant: the scissors are I went to the flow kind of thing.
Sumant: so by the time I realized that I was in the AI, it was already 2nd year of my PhD.
Sumant: So, So it's not that I wanted to go in, but I ended up being in, because as I followed what I wanted to do.
Sumant: Yeah. then post my PhD.
Sumant: I also continued the same thing because I liked Vernacular languages.
Sumant: however, I have an inclination towards businesses as well.
Sumant: So the career moved in different ways and I'm I'm here where I am today. Yeah.
Sami: Very, very cool. how many languages actually there are in India?
Sami: Because, I remember I was talking at one point about this.
Sumant: See, I think the number of official languages used to be somewhere around 17, I think.
Sumant: But, as per the record, as I recollect, I might be wrong, but there are more than 5000 languages.
Sumant: Which are there around. So there might be one some of the dialects might be also considered as languages, and some of these languages are not officially spoken.
Sumant: They're usually spoken at home or in certain region, probably few cities and all because There are forest area.
Sumant: There are tribal groups. so read their, the the language barrier is quite high across, regions there.
Sumant: So people did not have a way to communicate between, these regions. So these languages developed.
Sumant: a lot of dialects also exist in the similar way.
Sami: Well, so a lot of work for NLP for natural language processing.
Sumant: Well, quite a bit of it. Quite a bit of it.
Sumant: At the same time, I have a view that this language unification is also going to take place.
Sumant: so it's not that all the 2030 languages, which are predominantly spoken are going to survive throughout.
Sumant: Probably some A10 would happen and how long it'll take. I don't know. And this is good or bad.
Sumant: Once again, I'm not there for a moral judgment. yeah, definitely, I love my language.
Sumant: Even I write my name in Canada in my LinkedIn.
Sumant: So, but, yeah, I I wish it survives, but it's a it's it's a tough competition between languages.
Sami: It's a tough competition in the in the corporate world as well. We we spoke earlier about OpenAI.
Sami: So has been a very fascinating, 4 days or so since last Friday.
Sami: One CEO being fired, another one hired. Yeah.
Sami: I guess he's gotta get fired too now that some old money is coming back.
Sami: So, the shortest tenure in the world.
Sumant: So so so it's it's it's like a technological advancement. Right?
Sumant: So the the same pace as the technology schedule advancements, the management advancements are also happening.
Sumant: So so I see it very similar. So, And, I don't know.
Sumant: So we are interesting in, we are living in interesting time.
Sumant: So there is this Chinese curse or I think saying which says that may you live in interesting times, and we are living in interesting times.
Sumant: So, yeah, so we'll have to observe what is happening and take the further actions.
Sami: And this is actually quite interesting because it ties back into the cognitive enterprise in a sense because, many of my friends, on LinkedIn, they they actually asked JWT, like, what should I do in this kind of scenario?
Sami: And, answers were quite interesting. So, what is your opinion and perspective on, cognitive enterprises?
Sumant: See, when we talk about cognitive enterprise, it's ultimately, technology and data driven enterprise.
Sumant: when we say data driven, the first mode part is about collecting data. Right?
Sumant: So, so you want to get started with cognitive enterprise.
Sumant: being cognitive enterprise, we need to start creating these frameworks where we start collecting data for all the activities that are happening across the automations.
Sumant: when you say all all important, all that matters, many times, we would not even know what matters and what doesn't matter while we start collecting the data.
Sumant: Probably, we should start collecting data, whatever is possible, everywhere.
Sumant: And that brings out a lot of insights and every decision has to be driven by these, this data which has been collected.
Sumant: Right? So rather than trying to go by emotions.
Sumant: For example, what we saw with OpenAI, right, there are 95% of employees who are with the older CEO.
Sumant: And and, if we had known this information earlier, we would have been, I mean, the management would have been probably a little more cautious on actually calling things out and they would have taken some other decision.
Sumant: So making sure that we understand what is happening through data is the first important aspect.
Sumant: And second, which actually flows from here is Because we are getting data driven insights, the insights might change, and we should show enough agility to change ourselves.
Sumant: So that also brings in, so we should be able to change our systems.
Sumant: We should be able to change people in terms of the skill sets, not the people skill sets.
Sumant: So the cross training becomes very, very important.
Sumant: Should also be able to bring a diverse set of people in so that the ideas also percolate.
Sumant: So, various ideas percolate and the change becomes more sustainable.
Sumant: and with all this, we should be also careful about the biases within these insights.
Sumant: we have seen that there are partial informations that are collected within organizations.
Sumant: And because of that, there are incorrect insights. I would not even call them partial insights.
Sumant: Data can be partial, but insights will be incorrect.
Sumant: And, those incorrect insights If we use for decision making's making, we will also end up making planters.
Sumant: So whether we have a complete data to make a decision or not also becomes a major consideration when we take decisions through data driven systems.
Sami: Yeah. So in in a sense, it's, you have 2 parts.
Sami: You have kind of the generative AI, which we use in marketing and sales and communications and kind of gener asking general questions.
Sami: And then you have your decision science in a sense, right, that you use your own proprietary data and and you clean insights.
Sumant: Very true. And, these two streams are going to run, I think, parallelly for quite a long time because decision making.
Sumant: So these are called decision, decision systems or you can call it decision AI.
Sumant: These are, small, but very important pieces of, modules.
Sumant: this actually contribute significantly in terms of decision enablement.
Sumant: lot of tabular data streaming data that flows in. Goes through these systems to give decisions.
Sumant: And ultimately, obviously, there are as error prone as any LLM is a generative AI is.
Sumant: so that's where, in both cases, we have to have human in the loop in terms of critical decisions.
Sumant: so we'll also have to keep on monitoring the systems How do they go?
Sumant: How how are they behaving in recent times? How are they behaving for various changes in the data?
Sumant: And based on that, if we can rely on the system and we can if we can rely on the insights, And sometimes for critical info, insights, if we can rely on the external information as well.
Sumant: Along with the insights we are getting from the system, we should be taking the students.
Sumant: But, without these systems, see, I'm just talking about the negative side of it, but without these systems, there would be nothing, to take decisions on.
Sumant: There will be some outside sources which will help. There will be intuition driven decision making.
Sumant: There will so, there will be anecdotes driven decision making, which is even more dangerous than this.
Sami: Yeah. It's the, it's also known as the, the Hippo approach, which is the highest paid person's opinion.
Sami: which, which could have actually had an impact with this whole debacle with, OpenAI is probably 1 or 2 individuals got upset about something and decided that there is time for change And it had very little to do with, like you said, realizing that 95% of the people would be willing to quit over the weekend.
Sami: So massive, massive, implications and consequences, from not maybe thinking through a decision.
Sami: But so, So we always talk a lot about that.
Sami: And because of, you know, we like as an as an industry, and we like to talk about the consumer tools often, you know, whether it's a your iPhone or you know, AI enabled phone out from another vendor or something.
Sami: So we talk about JPG, and we talk about Bard and, Microsoft co pilot and, poet and all of these different LLMs and foundation models that exist.
Sami: how do you see those kind of, let's say, consumer facing tools, application within their organizations?
Sumant: Woah. Wow. Okay. So it's it's a
Sami: It's a loaded question. I know.
Sumant: Yeah. It's it's a very powerful tool, first of all.
Sumant: So and and and I'm just trying to balance it out. it's like a fire.
Sumant: so, fire actually helps us cook.
Sumant: Fire has fire runs steam engines, fire runs, the thermal power plants, the same fire actually kills people as well.
Sumant: So these these are very powerful tools today.
Sumant: And and I would say it's a paradigm shift in, the overall AI space.
Sumant: So if the first paradigm shift, I feel was the the the machine learning.
Sumant: 2nd was the deep learning space where, the feature engineering almost got, faded out and phased out.
Sumant: And then this is the third one where we are seeing a significant application in various spaces, which, of these elements.
Sumant: So, this user and tools generate lot of, meaningful information and some I would call it stupid information as well.
Sumant: why I'm calling it stupid is, it's still creative.
Sumant: It's still looks beautiful, but it's factually incorrect.
Sumant: So, but the lot, this meaningful information is being currently used by, large number of people.
Sumant: For example, I, myself, use chat deputy for my studies. A lot of times.
Sumant: I I sit and read things.
Sumant: I, so CPT4 allows you to upload a document and, actually, you can ask question on that and all.
Sumant: So you can get a lot of information, and the information which is relevant to the person.
Sumant: you can also generate, marketing content.
Sumant: You can generate, you can configure, if you are looking for a kind of a debugging tool and all those kind of things which come from this populate and all.
Sumant: But at the same time, one has to be very careful in terms of, data security, because if assume that you want to generate an email back to customer.
Sumant: And if you provided then sensitive information to chat GPT, it would use the same sensitive information to generate back the response.
Sumant: And if you could just blindly copy paste that the sensitive information will be sent to the customer.
Sumant: So there are issues there, kind of thing.
Sumant: So, so Samsung, I think there was the first incident which was seen Samsung's employer on the year back when was very, very popular.
Sumant: An employee actually pasted a significant piece of code, which was properatory to Samsung.
Sumant: whether they use it or not, but it's a breach of privacy.
Sumant: but I have been also seeing very interesting things. So it's not only about languages.
Sumant: You can also generate this you want the layer, diffusion models, right, latent diffusion models like, Dali and all came well before chat, GPT, and were popular enough.
Sumant: And I interestingly have a smaller number of parameters in them. So they're smaller models.
Sumant: But today, we are also seeing a a tabular LMS.
Sumant: So that that means that we might even start using chat, and, chat, JBT kind of systems for table decision making.
Sumant: So the decision systems, whatever we are talking about, might also be moved to these LLMs, but There are catches.
Sumant: One is the these are expensive. Your data is going out.
Sumant: And the retraining fine tuning becomes exorbitantly expensive.
Sumant: so many times, your tablet still would remain with you.
Sumant: So what it is in making, and a lot of these, vision and language based LLMs will be used for creative content generation and, text generation and, image generation kind of act.
Sami: Yeah. No. No. That's very true.
Sami: And it's kind of like the, you know, the the jobs of the newly minted MBA.
Sami: So we usually go to, investment banks to be quant their jobs probably will radically change, as these tools get better.
Sami: But there's a good point about that, the quality off the content generated by the LLMs.
Sami: I had the opportunity last week to talk to an attorney in London and and and so when I told you the story earlier, she was actually complaining that 2 separate law firms in London were talking about, a specific employment law, act.
Sami: And it looked like completely logical and, and correct on the outset.
Sami: And because it was 2 separate firms talking about that particular thing, she thought that was the real thing.
Sami: and and she said that she wasted 45 minutes, researching that in, in the legal tools from LexisNexis and Thompson Reuters and just to realize that they had used, Judge GPT or something similar to generate that, that content.
Sami: So it looked good, looked perfectly logical, but it was factually incorrect.
Sumant: yeah. So once again, so this is one major mistake, we can do.
Sumant: where, see, we have to treat all these LLMs as a junior associate who is actually doing the work for us.
Sumant: And the way we have in the, you know, human driven system, we have a the one person does the work.
Sumant: There is a review. There there is second level review, then only they get into some critical systems.
Sumant: So they, I mean, the I would say if somebody's website has this kind of content.
Sumant: It's a critical system because you are providing information which is is expected to be reliable.
Sumant: So If somebody treats it as an expert, then there is an issue.
Sumant: So it's the chat, GPT, or any other LLM is not an expert.
Sumant: It can expedite the process of generating certain content and make it look good.
Sumant: But there is a still a possibility of making errors.
Sumant: So having those repeat process, this is where I see, I don't completely agree that jobs will go away because of this and all.
Sumant: That I I mean, it would not affect the jobs, but rather it will end up creating more tools on top of it.
Sumant: There will be more people who will be cross skilled to take care of content or output of these models, will be I I would see a lot of new applications utilizing this and because of that new jobs coming out, just that we have to cross skill and get there.
Sumant: So, so, yeah, it's like a junior assistant for. So for us
Sami: I like to I like to think of it as some of the grunt work going away or becoming more effective.
Sami: Yes. it's like with the image creation Adobe has put a lot of effort over the years into the their Firefly AI engine.
Sami: Right. And obviously, some of the demos they do now on how you can manipulate images and videos very rapidly.
Sami: You could always do it with Photoshop and other tools, but now it's just so fast.
Sami: You can you can lose the sky on the back and replace it with sunshine and, and rainbows and whatever you want.
Sami: Yeah. You could always do it, but now it just takes 2 seconds to type in the prompt, and that's that's it.
Sumant: Very true. So the so when So some people might treat this as grant way work.
Sumant: Some people might treat this as their creative work.
Sumant: So if you want to if we want to call it creative, yes, there's this these engines have become creative, but creativity doesn't mean factual.
Sumant: So that is a that is the line we have to draw between these things.
Sumant: So they can get really, really creative. They can write beautiful stories which are non factual. Fictions.
Sumant: And, they might make it sound as if, it's true.
Sumant: For example, there was a question asked to bar by one of my senior colleagues of the previous organization.
Sumant: It says who is going to win World Cup Twenty 23 Cricket.
Sumant: And, it was asked before the final match, and it said, okay.
Sumant: The work of final is already over and it has 1 and this this is so it went on telling certain things and I got that screenshot a day before and I was saying, okay.
Sumant: So such a true hallucination. Right? It's a and A
Sami: true prediction, if you wish.
Sumant: No. But we did not go on winning Right? So we lost.
Sumant: So, yeah, so it's but the point is the way it narrated it.
Sumant: Sounded as if it has happened already.
Sumant: Yeah. So that's where we'll have to be very careful.
Sami: Yeah. That ties, with organizations, and enterprises, particularly, because, especially in litigious, societies, like in the United States, the accountability and liability on on that becomes a little bit complicated.
Sami: even though you can say on your website is that, you know, we're not giving you advice on the website.
Sami: This is kind of marketing fluff.
Sami: somebody could potentially take that the wrong way and potentially sue you or something.
Sami: So, yeah, Like you said, you have to be very careful reviewing what comes out of these models.
Sumant: Yes. But at the same time, I have seen a lot of educational content coming out really well.
Sumant: if we actually upload certain documents and ask specific questions on those documents, I've seen, these models doing well on them.
Sumant: So so context becomes a major player in this. Right?
Sumant: So how unambiguous we are in terms of querying as a user? That is one part.
Sumant: Then another part is how much the LLM knows about the space.
Sumant: So this is where the fine tuning of LLMs for domains becomes very, very important.
Sumant: So if we want to run to, run LLM based systems or particular domain, which is, not covered by the data training data.
Sumant: Rather we should fine tune it enough that these hallucinations do not happen.
Sami: Absolutely. Absolutely. So on the more on the maybe decision AI side then as well.
Sami: It's like it's one thing that we generate marketing materials out of the or emails or something like that.
Sami: Or chat bots for customer service.
Sami: But, you've done a lot of work with, with machine learning algorithms as well.
Sami: I'm looking at very large datasets for helping retailers particularly do certain things better.
Sami: So could you talk a little bit about about that and how different it is to the gen ai?
Sumant: Sure. So, once again, whenever there is a new trend, people forget about the old stuff.
Sumant: But that doesn't make the whole stuff obsolete. the whole stuff is still very exciting.
Sumant: The conventional machine learning did not get killed by deep learning, and similarly, LLMs will not kill either deep learning or machine learning.
Sumant: So they have their own space.
Sumant: So, for example, if we want to build a quick custom decision engine, whether my customer will churn or not, based on the data I have.
Sumant: and I would want to maintain the property or the ownership of the data without sharing it to someone.
Sumant: Still use this custom AI.
Sumant: So we have been working on, whether a customer whether an item is stocked out or not, even though the book says it is available.
Sumant: Right? So this is one of the major problem our customer space today in retail space where they go to the rack.
Sumant: They check the the item is not there.
Sumant: But, they go to the, pause and check if the book says it's, there are 28 terms of particular chocolate.
Sumant: And you don't know where they have gone, but can we even predict that? Okay.
Sumant: It might have happened that the we have lost or the they've been stolen or broken, and the current valid stock is 0.
Sumant: So if we can so we are working on this kind of problems where customers' revenue gets impacted immediately.
Sumant: And uh-uh these there are many use cases like this.
Sumant: So in these use cases are going to remain there.
Sumant: there is going to be significant, application of machine learning and deep learning, in this kind of use cases as well.
Sami: Yeah. And it's it is interesting also because, like, like, I remember, going to Home Depot in in in New Jersey.
Sami: Luckily, I had, couple of Home Depot Sandoz nearby where I lived, but looking online is I was looking for something, and they would have 5 units in stock supposedly online.
Sami: I drive to the store just to realize that there aren't any.
Sami: And later, I learned that it has a term of ghost or inventory. Yep. and it's super annoying.
Sami: And it's also super annoying, not only for the consumer, but also for the staff So, of course, I would go and ask some guys, like, hey.
Sami: It says you have 5. Where where is it? You know?
Sami: They're like, well, maybe it's up there on the shelf, but I don't wanna take the forklift or it's backroom or and then later, I kind of, gave up on trusting any of that inventory data as a consumer.
Sami: So I would always if they have 15 or more, I would start trusting that they actually might have some.
Sumant: Yeah. This is the problem. Right? So So the trust deficit, which gets created because of this.
Sumant: This has a major impact on the business. So so I might stop trusting their website.
Sumant: So So the website was the one which used to drive me to their store, but now I don't have an access to their store online.
Sumant: So I might even stop buying from them. So it actually is a major challenge for retailers.
Sumant: And, if so that's where we are trying to solve this problem and not trying to solve.
Sumant: We have a product on this. we already have it.
Sumant: And, I'm very sure that this is going to help a large number of retailers to not have any negative impact because of ghost inventory.
Sami: Yep. And I think this is, you know, like you said, is that the the training can happen frequently.
Sami: So it's important as the consumer behavior changes For instance, if you are doing that trying to predict, renewals or churn or some behavior of a consumer, for instance, the company, and you do certain activities, those activities will have an impact in, in the behavior, which means that the model needs to kind of retrain itself in the in the most recent data
Sami: in order to be accurate.
Sumant: Yeah. So continuous learning is a part of any, machine learning production system.
Sumant: Especially if it is running on, real new data. Right?
Sumant: So the user behavior keeps on changing, unless And even even if it is a machine, it's so we have seen it's not just in users.
Sumant: If there there are machines whose behaviors change because of wear and tear or the places where they work and a lot of other things.
Sumant: So so weather condition. So Continuous learning involves monitoring the data patterns, monitoring the outcome of the model.
Sumant: And based on that, there are ways to set up for retraining, and we retrain to make sure that, better model is selected.
Sumant: Which can actually give better output for the new who do paid off data.
Sumant: So this keeps on happening in a production system and that's how we keep the model running it?
Sami: Yeah. So that's it's really interesting when you mentioned kind of the machines, it it because we often think that His AI or machine learning, it's mostly good for retail or consumer behavior recommendations like Amazon.
Sami: But in fact, it has its place on predicting maintenance, for instance, there'll be airplane engines. Right?
Sami: They they send sensor data constantly to Rolls Royce and other manufacturers, and to optimize the, the maintenance cycles and make that things are running constantly.
Sumant: Exactly. Exactly. So, anywhere there are real systems, physical systems, whether they're biological or mechanical, And if they are generating data, I think machine learning has a play.
Sumant: So we so be it shipping industry, be it airline industry, be it any transport.
Sumant: We have a significant.
Sumant: So a lot of this routing of airplanes and all takes care of some trivial kind of AI and all in some cases.
Sumant: ship there are some ship softwares which actually try to utilize this.
Sumant: there are the traffic control mechanisms within cities that significantly deploy.
Sumant: So machines also generate relevant data, which can be utilized in machine learning.
Sami: Yeah. And it's, it's actually interesting because, our New York office, is in World Trade Center, and they have the smart elevators where you have to essentially tell which floor you are headed as as you are, you're getting into the elevator.
Sami: So the elevator bank can optimize and use machine learning algorithms on making sure that the right elevator comes and grabs you and takes you to that right floor.
Sami: Yeah. Which is quite kind of a basic use case in a sense, but, but it has impact in in real life or real people.
Sumant: Yeah. Yeah. It might be a simple, rule based AI as well, but a AI exists.
Sumant: So, and and that's what it is.
Sumant: So ultimately, it might also have a complex machine learning algorithm.
Sumant: I won't deny it because it might So some, elevators might use higher amount of energy when they go up or there might be various factors which are there.
Sumant: And if they exist, then a machine learning algorithm can also be utilized, but, many times, they use use it using the rule based AI as well.
Sumant: And you rule based AI has been a prominent, problem solver from long time. Yeah.
Sumant: I mean, it it has its own limitations, but it still exists.
Sami: Yep. That's, that's really interesting. Yeah.
Sami: One other, interesting use cases I I heard, some time ago was, there's a company called, Framery that does these, like, phone booths, like soundproof, meet small meeting rooms and phone booths for companies.
Sami: but they also require because there's some, air conditioning happening and and depending on the ambient temperature outside versus how many people are inside, how much, like they're using.
Sami: They also have some sort of an AI, hard coded AI, in the logic that essentially manages the, the electricity for the booth and, and AC.
Sumant: Yeah. Yeah. I I'm sure.
Sumant: And, a lot of these principles also can be applied for a larger systems as well.
Sumant: So today, the power grids, the smart power grids and, smart grids, what we are talking about have been using AI, to supply electricity.
Sumant: They also give signals back. They give provide forecast of next 24 hours and all those kind of things are being done by some some I mean, I'm not saying all all smart grids to it, sounds smart grids to it.
Sumant: And, what it also means is that, there is a knock operational optimization that is happening.
Sumant: So optimization of operations taking place and which intern reduces cost increases revenue, makes it more productive.
Sumant: And also, especially in terms of electricity, if there are areas where There is a scarcity of electricity.
Sumant: This can be diverted. So, so it's also a social cause. it's not just, money.
Sami: Yep. And similarly, you know, like, retailers could, could forecast their staffing requirements depending on when the Shipments are coming from distribution centers, or depending on the weather.
Sami: Yeah. They can run promotions. I think, yeah, it's, it kind of feels like whether it's machine learning or well, AI, like, I think you said it sometime.
Sami: There is no AI. There is only machine learning and algorithms.
Sami: But, AI seems to be everywhere in our language nowadays.
Sumant: Yeah. So today, top popularity, is with machine learning.
Sumant: I would retract my statement on that, because I know that a lot of small, small systems use regular AI as well, which is not even a machine learning.
Sumant: So there are systems which run on graph based AI. So there are graphical connects.
Sumant: And based on that, they actually create optimal algorithms to identify shortest path and all those kind of things, which get used in knowledge graphs and all.
Sumant: But The neurosymbolic AI, which involves this symbolism as well as neural networks. Are also there.
Sumant: So they have evolved where people want to utilize them if the systems are more complex.
Sumant: so Yes.
Sumant: The they exist everywhere and, cognitive enterprise definitely is not an option today.
Sumant: nobody can say, okay. I will think about it.
Sumant: what they can think about today is, where do I start?
Sumant: And how fast I can get there and what are the decisions I want to take quickly?
Sumant: So based on that, we can, I mean, they can start collecting data?
Sumant: They can start building the systems which can provide intelligent insights, and also, start the cultural shift where agility becomes the nature of the organization.
Sumant: Where people are ready to upskill, cross skilled, and, quickly take up any relevant thing that is to be done by them without much effort.
Sami: Yep. Yeah. Absolutely. Absolutely.
Sami: And some organizations already have big AI teams, so to speak, like Diageo got owns a lot of, alcohol brands or Decathlon, which is a big sports, producer, sports equipment producer, but Not all the companies are that advanced.
Sami: So would you recommend that everybody would kind of start looking at their data and start experimenting and looking at the use cases, would they work with somebody, like, maybe, like, cloud 9 to figure out the best use cases, or do you recommend that they start figuring these things out themselves?
Sumant: See, figuring these things out themselves is an effort.
Sumant: especially if one is, one is new to this. Right? So that's one part.
Sumant: And, second thing is even if they have a large number of data scientists, I'm sure larger number is of problems.
Sumant: So I don't think they'll be able to address every problem they want to solve.
Sumant: And we have seen this earlier as well.
Sumant: Like, we have worked with very large organizations where they very strong data science team.
Sumant: Still, there are very good problems. One second, I'm using this very, very multiple times to just emphasize.
Sumant: There were very good problems which they could not even touch 4 years.
Sumant: So people like CloudNine are essential for them to, have these problems prototype and see whether they are feasible to solve or not and then take them to the production as well.
Sumant: So we can definitely do that.
Sumant: And, because we have a strong air team, we will also be able to handhold them in terms of technology.
Sumant: Right? So we, we understand what kind of methodology should be employed.
Sumant: We use something called co innovation driven delivery to deliver this stuff.
Sumant: So We work with customer on this and help them solve their important problems.
Sami: And and typically, you need the data. Like I said earlier,
Sami: Some with accurate data and ideally lots of it. Right? It's kinda hard.
Sumant: Yeah. So so, usually, we start with problem.
Sumant: And if the the problem demands some data that is not available, we will do a kind of a reverse engineer because ultimately, problem is important because we identify, value through problems.
Sumant: So If somebody says that, okay, my, my trucks are running almost a week late, every time they start throughout the country.
Sumant: So something like this. Right? So there might be a truck which actually is having cold storage and runs throughout and So overall supply chain delay is a week for me.
Sumant: That is a real problem. So the because of that, they might be losing, say, 10 k dollars every trip.
Sumant: And if it is a high value good, so, that is a real problem for the customer.
Sumant: Now We go. We verify their systems.
Sumant: we also sit with business to understand whether that's really the problem they are facing or there are other problems.
Sumant: there might be probably a loading issue or something or there might be a manufacturing issue because of which the trucks are waiting and, and it's not a truck issue.
Sumant: Right? So we get deeper there.
Sumant: And then we identify what is the real problem, and then we get into data.
Sumant: What is required for it? So we get to the data store, understand whether we have the relevant data or not.
Sumant: if it is not available, we start to we do 2 things. 1 is, if it is some amount of decent data available, we start with the prototyping.
Sumant: At the same time, we also perily built, the data collection engine, which can start collecting the data required for solving this problem.
Sumant: At the same time, this prototype is evaluated.
Sumant: If it has even a bit of value generation for our addressing this problem. We start with that introduction.
Sumant: And as in when the new data comes in, the model changes, there will be higher efficacy And eventually, they'll be able to get to a stage where, there'll be significant reduction in whatever problems they are facing.
Sumant: but once again, it's a it's a very interesting thing.
Sumant: I have seen people generating very good insights, but operationally not using them.
Sumant: So it's important that we generate insights good insights.
Sumant: At the same time, we also operationally effectively use them so that, the overall benefit gets realized.
Sami: Yeah. So it sounds like there's a there's kind of like, few buckets in in in a ways that a technical side, we have the data, we have the models, then we have the productionalization, if that is a word.
Sami: But then we also have the responsibility for people to do something in many cases.
Sami: You know, like you mentioned, the, the the retail ghost inventory is that we can give the insides of stockouts, but somebody still needs to, you know, ship the the goods into the store and put them on the shelf.
Sumant: Yes? Yes?
Sami: I hope that nobody steals it until it can be sold.
Sumant: Exactly. And, I mean, technology definitely, and technology can go so far and and, yeah, when somebody might even hypothesize and say, okay.
Sumant: I'll deploy robots to do it. Yes. That's all fine.
Sumant: Still, you will have larger gaps where human beings have to pay play critical roles.
Sami: Yep. Absolutely. And sometimes automation, frankly, also has a cost Yeah.
Sami: yeah, it's, I don't know if if you've seen it in the news, but, but several of the supermarket chains around the world are stopping the self checkout.
Sami: while it was great for them for a period of time because it reduced the amount of staff they required at the checkout lanes But now they're realizing it that the shoplifting has at least doubled compared to, you know, assisted checkout compared to checkout.
Sumant: Very true. So so, I think the kind of emotional intelligence that is brought in by human beings being in the system cannot be, at this point of time, be brought in by a agents.
Sumant: And Lot of these business processes require emotional intelligence too, not just the insights and the decisions.
Sumant: So, It's it's a it's like the, once again, OpenAI's decision to save. Okay.
Sumant: the CEO may not be very good for them in terms of their business, but There are other things which get impacted if you don't consider them.
Sumant: there are going to be, catastrophic impacts.
Sami: Yeah. So it is actually, like you said earlier, is that, you know, we often talk about how many people are going to lose their jobs.
Sami: But in fact, it sounds more like the jobs will change and potential into a more meaningful way.
Sumant: Yeah. Yeah. So so I believe there are there are umpteen number of areas where even technology has not even reached.
Sumant: So for I'm a I'm a positive guy.
Sumant: I look at it this way that, okay, these spaces will now be touched.
Sumant: And we might be able to solve a lot of problems, be it health care, be it space, with your political problems, can we have some, AI driven suggestions in that space as well.
Sumant: Right? foreign affairs. Lot of these kind of things have actually, I mean, they use some technology internally, but they still are, not so open to use this and energy sector.
Sumant: Right? So energy sector is, I think, is another area. So large number of areas rural development.
Sumant: Quite a few areas where technology has not yet penetrated. So I would not panic out.
Sumant: I would rather look at it as an opportunity where new spaces will open up.
Sumant: So we and they will open up for those people who actually have that hunger to crosskill, upscale themselves, and go and solve the problems.
Sumant: So problem solving becomes a major requirement rather than just technological skills. So can I understand the problem?
Sumant: Can I solve the problem rather than giving a piece of code?
Sumant: That is going to be a major thing.
Sami: Now that is that is very true.
Sami: And and when you when you mentioned about the changing of the jobs, who's thinking about also traditionally very hard jobs like farming or mining.
Sami: Both of those things are are slowly with the technology and AI and other things, we are starting to become more kind of controllers somebody sits in an office with a joystick and and drives a machine, 2 kilometers underground.
Sumant: Exactly. So if you if you have heard about this, there are 40 people who are stuck under a tunnel in India now.
Sumant: And, and government is doing everything possible to bring them out.
Sumant: But had there been systems which could alert them that, okay, there seems to be there is a possibility of an accident just move out.
Sumant: Can it predict certain accidents? And if that is the case, so it can save lives.
Sumant: Yeah. So, yeah, so There are areas where technology has not at all penetrated them.
Sumant: It has penetrated it to 1.2% of whatever is the potential.
Sami: Yeah. And hopefully, with the technology becoming becoming cheaper, we'll have more sensors that are affordable
Sami: That even in industries where they traditionally weren't, accessible now become, you know, meaningful support.
Sumant: Very true. Very true.
Sumant: I think that's that's a the good mix of the sensors and the data generation from them that is the data engineering and then ultimately the insight engine on top of it.
Sumant: That is the AI. It it is it is definitely a good combination, in spaces which are difficult to manage that.
Sumant: Definitely, there are challenges as well.
Sumant: For example, the sensors get dusty and there's there are issues in terms of, the reliability and all, but I still believe they would be much better than not having it.
Sami: Absolutely. Absolutely. So we're we're coming to our time, but I always like to ask a question about what's your favorite AI tool that you're using in your work
Sumant: Very good. Okay. So
Sami: Tough question. I know.
Sumant: Yeah. So, I got I got inspired by some things which I I see when when, Bert came in and when codecs came in, right, I think they fascinated me a lot.
Sumant: that was a that was a leap drop. Right?
Sumant: I I could see that technology is solving problems, which we used to fantasize earlier, right away.
Sumant: Can we actually even work in this space? So I loved those tools as the tools.
Sumant: So so if you look at what are my first cloud that that I also love the way when it might look very traditionalist, but Google search works and the Google Mapworks.
Sumant: we take them for granted, but The relevance they bring in is so high.
Sumant: lot of us don't even realize how how would be life. Without this.
Sumant: So but in today's word, I I once again vote for, copeilot because I still believe consumer services has to utilize significant human intervention before before them exposing the virtual assistant to the users, so that they provide very good user experience.
Sumant: But developer experience is not that big a challenge.
Sumant: Developer always is hungry to utilize content which can solve their problem.
Sumant: And if there are tools which can provide them automated solutions without them typing match, without them browsing match, they'll be loving it.
Sumant: 2nd day is also lot of the business, say, techy guys.
Sumant: Technocrats are also becoming, coders nowadays.
Sumant: They actually like to quote small small POCs and all those kind of things.
Sumant: I think that's going to be a trend. So, Copilot once again enables them significantly.
Sumant: Yeah. So so it enables developers. It enables business guys.
Sumant: So if once the business guys start doing this small work, developers also can focus on larger challenges.
Sumant: And provide quicker solutions. So what I believe is Copilot is one of the very important tools for software development and utilizing it will help in faster development.
Sumant: Quicker outcome. And, hence, we'll be able to solve larger number of problems, eventually.
Sami: Very true. And it's also the, the other co pilot for the kind of for folks like me, you know, the kind of more business users is the, you know, the Bing chat, which is also known as the co pilot that easily connects you to the chatty BT and Dolly.
Sami: I was I was showing my wife literally today on how she can prompt Dolly a lot easier through, the browser than any other way and get some pretty neat results.
Sumant: Yeah. Yeah. And the same time, I have some other favorites.
Sumant: So DeepMind has been doing this alpha fold, where they're talking about this protein folding and hence enabling the drug discovery, right, the faster drug discovery.
Sumant: And I I'm sure that's one space I will be watching out, as a personal interest, where, there are so many because of globalization, there are there are going to be health issues, and those health issues are going to spread faster as well.
Sumant: So how do we use technology to address the challenges in terms of health space is going to be one of the major things.
Sumant: It's a survival question for us.
Sumant: And, I would be very keen to look at the space of health care, like alpha board kind of spaces.
Sami: You're right because there's a vast amount of, medical research that has been done in articles written in, in, in, in a number of different countries.
Sami: Essentially, all the countries in the world for the last 100 plus years, which you talk about Elsevier or Walter's Cluru or one of these companies, And when whenever new things come up, rarely the new things are completely new, something like that has has appeared maybe in the past.
Sami: Just being able to find faster, some shortcuts and, and something that worked before, how could we essentially leverage something like that for the future as well?
Sumant: yeah. So starting from just a pure semantic search, to a level of drug discovery, I think there are a large number of applications.
Sumant: Can you even so, I used to work, as a visiting scholar with Doctor.
Sumant: Amit Chet earlier. And, his his team used to work on one area where can you use social media to identify whether there are health issues anywhere?
Sumant: So so data is available everywhere.
Sumant: Just we need to understand how do we identify the right channels and how do you utilize it and how to make it more reliable.
Sami: And, again, people need to be involved in the curate so that we can avoid the bias.
Sumant: Yep. Exactly. Exactly. So whereas as you come to us, I was just we were talking it about it yesterday.
Sumant: Right? So So there was a question whether can you solve bias and hallucination, when LLMs and I was like, you can solve hallucination probably because it's a technical problem.
Sumant: You may not be able to solve it 100%, but you might be able to solve it to a very large extent.
Sumant: It's a technical problem. But bias is something which is a social problem. Right? So, people generate content.
Sumant: And, so people are involved in generating data.
Sumant: And unless that behavior changes, it's very difficult to change the data patterns.
Sumant: we can put to scrutinize the data, filter it, but still the subjectivity comes in.
Sumant: So can we address the virus problem completely? I don't know. It's it's a it's a huge space.
Sumant: and because it is also subjective, the definition also starts becoming gray in terms of by us.
Sumant: so it's it's definitely, requires more a sociological address rather than technological address at this point of time.
Sumant: And once that is we see something in sociological space, probably technology can help
Sami: Yep. Yeah. I like to think of the biases. One person's bias is the other person's fact.
Sami: You know, it's, it is, it is, like you said, it is, it is a bigger problem to solve than purely tech technical.
Sumant: Yeah. Yeah. So in, in some places, for example, crimes and or probably 99% of the population agrees on what is bias and what is not.
Sumant: But there are a lot of areas which are not that black and white and where bias plays a subjective role.
Sami: Yeah. Opinion. A lot of things are opinions driven as well when we do that in the content, particularly.
Sami: Math is simple. 2+2 is 4. But Exactly.
Sumant: So exactly you you see the the today's war between Gaza and Israel. Right?
Sumant: There are opinions on both sides, and it's very difficult to say which side is what.
Sumant: And, everybody wants to defend their opinion.
Sumant: So, yeah, might I might take an opinion on certain space, but that's my opinion. That's all.
Sami: Yep. Absolutely. And facts are hard to find.
Sami: Thank you. Thank you so much so much for your your time today.
Sami: And, and I I I wish we we would have a lot more time.
Sami: I know this is a fantastic topic to discuss, but, we will have another opportunity, hopefully, very soon.
Sumant: Thanks, Damian. And as usual, I said, That's just it's always fun to talk to you.
Sumant: great, great discussions, thought provoking. I had good learning as well.
Sumant: So, yeah, we'll talk to you again sometime, and we connect on this kind of sessions frequently.
Sami: Absolutely. Fantastic. Thanks so much. Thank you. Thank you. Have a great day, everybody. Thanks.
Sumant: Have a great day, everybody. Have a good night as well. Thank you.
Sami: Happy Thanksgiving for the Americans.
Sumant: Happy Thanksgiving.
Sami: Good. Thanks. Take care. Bye bye.
Sumant: Bye bye.