AI has had a number of transformative effect on people and businesses, but how do you use AI safely in your small business? This panel of industry insiders at Imagination In Action’s ‘Forging the Future of Business with AI’ Summit breaks how entrepreneurs can be realistic about how they can use AI to grow their business and help customers.
Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1
Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:
https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript
Stay Connected
Forbes newsletters: https://newsletters.editorial.forbes.com
Forbes on Facebook: http://fb.com/forbes
Forbes Video on Twitter: http://www.twitter.com/forbes
Forbes Video on Instagram: http://instagram.com/forbes
More From Forbes: http://forbes.com
Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.
Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1
Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:
https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript
Stay Connected
Forbes newsletters: https://newsletters.editorial.forbes.com
Forbes on Facebook: http://fb.com/forbes
Forbes Video on Twitter: http://www.twitter.com/forbes
Forbes Video on Instagram: http://instagram.com/forbes
More From Forbes: http://forbes.com
Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.
Category
🤖
TechTranscript
00:00Let me start with my introduction and then we go down the line.
00:04I'm Azita Arvani, I'm the former CEO of Rocket and Symphony North America where we disrupted
00:11the telecom market by creating the very first mobile network that was cloud native and had
00:18full automation and also used open architecture.
00:22And that now we're starting to use a lot more of AI in our automation and we've talked to
00:28a lot of startups that did that as well.
00:30I'm also on the board of two public companies.
00:34One is using robotics in doing autonomous industrial cleaning machines and another one
00:43that does augmented reality that when you combine AI with AR you get to a lot of interesting
00:50use cases.
00:52On this panel I have these amazing CEOs and chief AI officers for these companies.
01:00So Shashank, why don't we start with you.
01:03You're CEO of Dextra.
01:04You've done this for almost 16 years now.
01:08So tell us about yourselves and Dextra.
01:11We are the youngest ERP company.
01:13I make ERPs.
01:15These ERPs are the building blocks of the blueprint for a small company and medium-sized
01:22companies sold to about 100,000 businesses in Asia, India, and now in the U.S.
01:30And I keep saying ERP is the most obvious use case of any intelligent system like AI
01:39for small or medium-sized businesses because large companies use giant systems like SAP
01:46and Oracle, but medium-sized companies which generate a lot of jobs, like 80% of all the
01:50jobs and 70% of GDP comes from medium-sized and small companies, and they don't have systems.
01:56And they take a lot of resources, money, people to get their systems right.
02:04So AI use case is perfect for those companies.
02:07I've been working with those companies for the last 15 years, and the last few years,
02:12AI has been our mainstay.
02:14Again, we'll qualify AI.
02:17Igor has a very good perspective on AI, and I'm a fan of his.
02:20We'll qualify what AI means, and there's a lot of snake oil in this vocabulary, so we'll
02:26qualify that.
02:27But what I'm trying to say is that if we can help small and medium-sized businesses, and
02:32I'm not saying we don't have large companies, they can help themselves, but if small companies
02:37can use automated systems, this can really help our economy because they have 80% of
02:43jobs, so this can help in the bottom of the pyramid, and that's what I do.
02:47Great.
02:48What do you think?
02:49Thank you so much.
02:50And Igor, I know you've gone back and forth between startups and big companies with IBM
02:55before, and I know your sister's first name, Alexa, so tell us about that.
03:02All right.
03:03So over two decades ago, I was leading the multimodal research team at IBM, where we
03:08discovered the baby version of Watson, and they didn't want to greenlight it.
03:11It was just way ahead of its time.
03:15So like Hartman from South Park, I said, screw you guys, I'm going home.
03:19And I picked off some of our top engineers and scientists, stood up our last company.
03:23If you've ever seen HBO's comedy, Silicon Valley, they lampooned a conference called
03:28TechCrunch Disrupt.
03:29While I went on stage at the very first one, I popped a razor flip phone out of my pocket,
03:34I spoke into it, and crickets.
03:37Nobody knew what the heck they were seeing.
03:40What they didn't know at the time is we were secretly working with Apple on Siri before
03:44the iPhone even existed.
03:46That's how early they were thinking about these natural user interfaces to their credit.
03:51And then five years later, we were acquired on the download by Amazon to birth what everybody
03:57knows as Alexa.
03:59The code name for it was Pryon, and that's why we ended up reusing it for this company.
04:04So yes, everybody in the audience can blame me.
04:07It's my fault you started having to pay sales tax on your Amazon.com orders, because they
04:12now had a Nexus in Kendall Square.
04:15Great.
04:16Thanks Igor.
04:17Next, we're going to Stefano.
04:18Stefano, you're the chief of AI R&D.
04:23How is that different from chief of AI without the R&D part?
04:28Well, it's research and development, so researching into it and doing it.
04:33So it's probably the same.
04:37Yeah, at Seeker, we make AI that can be trusted, it's explainable, and it's transparent.
04:46And we want to enable others to do so as well.
04:52And then, what does Seeker do?
04:54Yes, like I said, we have AI-driven products across different dimensions.
05:02We evaluate content of different modalities.
05:05Think of text, images, audio, and we score it, qualify it across different dimensions.
05:12Reliability, bias, toxicity, hate speech, profanity.
05:17And our users, customers, our advertisers, campaign analysts, the regular user that
05:29really want to discover what is behind content and how reliable and truthful it is.
05:36So we can call it like the karma of content, with the credit score that you assign to various
05:41pieces of content.
05:42You can think of it like that.
05:44Okay, very good.
05:45Thank you.
05:47You know, there's a lot of AI developments in the recent, I mean, every day you get up,
05:54there's something new going on, right?
05:57How has these advancements affected the landscape for the enterprise, you know, from the perspective
06:04of your customers?
06:05How are they looking at these AI advancements?
06:08So I'll take the liberty of defining not just enterprise, but also small business needs.
06:13Small and medium size.
06:14And I said 80% of our economy is around small and medium size, and that's where we play.
06:22And even in enterprise, because this is basically a superset of all the needs, AI has limitations.
06:29It can't do math.
06:30We all know that.
06:31It can't do math.
06:32So, you know, you write an algo where it tries to create a program to do math whenever you
06:38ask it to do, you know, a complex calculation, it'll basically, you know, turn on a program
06:42to do math in real time.
06:44That's what it does.
06:47If I told you that there was a chance that the answer that I give you in a business setting
06:53is going to be wrong, I'll be fired.
06:55So there's a 1% chance that my boss, I don't have a boss, but let's say I had a boss.
07:02You know, if I asked a question and somebody says, look, I say, what's my sales?
07:06And they say, it's $100 billion.
07:09And I know it's not $100 billion because, you know, then I would be alone sitting here
07:13alone with you guys.
07:15But it would be wrong.
07:16So they would be fired.
07:18And that's what AI does.
07:19There's a chance that it'll be wrong.
07:21The answer will be wrong.
07:22So you have to check every output.
07:24And there are ways to do that.
07:25You know, Arvind was here in the morning, his system was on RAG, which is retrieval
07:30augmented generation.
07:31It checks for answers, checks if your answer is correct or not.
07:35The point being AI, there's a chance it is going to be wrong in the answer it gives you,
07:41which means it doesn't work in a business setting.
07:44So then you need to quantify what it will do and what it can do.
07:47Now, what it can do is have a conversation with you.
07:50This is what we see in Chargivity.
07:52In the case of a business user or enterprise, 70% of, how many of you know that 70% of small
07:58businesses and medium sized businesses don't use software except QuickBooks?
08:02That's a big challenge.
08:03And you've heard of SAP and Oracle and ERP companies, but they haven't been able to penetrate
08:06even 5% of that market.
08:09It's because small businesses don't trust software.
08:12And look, what are we talking about, right?
08:13If AI cannot be trusted, they're not going to trust, you know, whatever you sell them.
08:20So they don't trust QuickBooks.
08:21They don't even trust the data that comes out of QuickBooks, because they feel that
08:26the numbers that they see in those reports doesn't represent their business.
08:30And anyone who runs a small business knows that.
08:32They have some, you know, they have a view of the business that is not seen in the numbers.
08:36And AI can't solve that.
08:38It can have a conversation.
08:40It can do onboarding.
08:41It can do a lot of other things.
08:42And I have a Twitter account, which I spend a lot of time writing about what it can do.
08:46But these are the limitations.
08:48And I, you know, I'll leave it for the rest.
08:49But I think we have to, you know, we have to have this, a reason of expectations on
08:55AI, as far as it concerns about enterprise and small businesses.
08:59It can't do everything.
09:00It can do a very small amount of things.
09:02Actually, it's interesting.
09:04I did a search on one of the companies I'm on the board of, and it said that their revenues
09:09was one, not a search engine.
09:13It said the revenues are 1.6 billion.
09:15And I knew that's a little higher.
09:17It was actually 1.2 billion that I checked.
09:19So you're right.
09:20It's 1.2 billion.
09:21Yeah.
09:22But it was saying it's 1.6 billion, which is, you know, it was a nice thing.
09:26So I said, oh, I don't even know the revenues of the company I'm on the board of.
09:30That's strange.
09:31That's why I checked.
09:32Anyways, the point being that you obviously have to check these numbers as they come up.
09:37Igor, what do you think, how should the enterprises think about, you know, they hear all these
09:42things, they get excited, and they come to you and say, hey, give me the latest gen AI.
09:48I want to, you know, take that and brush it all over my enterprise.
09:52Yeah, it's actually worse than that.
09:55Every pandemic, I remember giving a talk at the chief digital officers forum.
09:59And they said, hey, what's one thing that you're going to now tell us?
10:03It's very important that we should know.
10:05I'm like, here's what's going to happen.
10:07You're going to waste my time for one year.
10:10Because it's going to be a build versus buy.
10:12You're all euphoric.
10:13The picks and shovel crowd are knocking on your doorsteps.
10:16And your IT teams are going to try to figure this out.
10:18And then I have to channel my inner Japanese soul to figure out how you don't lose face.
10:25You know, I can't let them use face, so I have to do all this theater for the better
10:29part of a year before you realize there's no way you can build what we've been building
10:34for 20, 30, 40 years.
10:36So that's the first thing that we typically encounter with these enterprises, is the stages
10:40of grief until they finally realize it.
10:43The other thing is, I mean, I adore that he's starting on SMBs and bringing in capabilities
10:48that would otherwise be unyielding to them, like the SAPs and the rest of them that they
10:53would never hope to touch.
10:54I did what my COO calls SaaS backwards.
10:59When he encountered me, he's one of the original Oracle execs that Benioff brought in to help
11:02him go to markets from year two through IPO, and he kind of was laughing his head off.
11:08He's like, you're going after the biggest, baddest organizations on planet Earth, the
11:12largest government agencies and things like that.
11:14Why are you doing that instead of going SMB?
11:16And I'm like, here's why.
11:18If you want your kid to be an Olympian, you hang out with Olympians, and you get trained
11:22by them.
11:23And if we go and we can service and install things in nuclear reactor sites and government
11:29agencies and things of that sort, then I know everything is down market, and the team gets
11:35that experience in a lot more aggressive way.
11:38So over a half decade ago, we said, hey, there's an intersection of AI and knowledge management
11:44that doesn't exist.
11:45Look, we're all going to these conferences, and all the big tech folks are showing up.
11:50Hey, look at all this stuff that I now have for enterprise and AI and stuff like that.
11:542017 is when we started.
11:56Who was talking about enterprise AI then?
11:59Nobody.
12:00Who was talking about having conversational interfaces on these systems?
12:04Nobody.
12:05You know why?
12:07Because every time I said, hey, this is what it's going to be like, everybody was saying,
12:11no, the interfaces are going to be more Tableau-like, more Looker-like, all selectors and things
12:16of that sort.
12:17That's how enterprises wanted to use that stuff.
12:20And I'm like, that's peculiar, because all of us know how to talk to each other before
12:24we learn how to read and write as well.
12:26So I'd rather focus on putting the human at the center of these experiences, and then
12:31everything will work itself out in the wash.
12:34And then what happens with ChatGPT 24 months ago, that gets revealed, 100 million people
12:39start using that, and then everybody starts parroting the same thing, and I get to roll
12:43my eyes.
12:44Stephanos, can you talk about how enterprises are seeing the advancement of AI, but in particular,
12:53maybe touch upon the fact that since you come from the R&D side of this, the technology
13:01has democratized across, whether you're large or small, you can use AI for your customer
13:09service, whether you're a $1 million company or a $1 billion company.
13:13How do you respond to that, and what we heard from Shashank and Igor?
13:17Yeah, so I guess I think democratization is one way of seeing it.
13:23There's open source, there's open source AI, but there's also, I think the vast majority
13:27of companies are not using open source AI, they're using closed source AI.
13:35The big challenge with really open source AI is computing costs.
13:44In order to make AI, in order to build AI and maintain AI, you need computers, you need
13:50power, and that's expensive, right?
13:59There's going to be a correction, I think, with where things are going, and there's going
14:04to be some kind of change in the status quo of the hardware providers with more competitive
14:14AI accelerators and chips that are coming out.
14:19But yeah, generally, I think democratization is an interesting way of putting it, because
14:26there's only still a very few that can operate open source due to the prohibitive costs.
14:35Great, thank you.
14:38Igor, which specific strategies, since we have a lot of people here that really want
14:43to take the ideas and put it into action, so what specific strategies, what lessons
14:49learned do you have from your work at Prion that has worked for you in the enterprise
14:56to grow your revenues?
14:59Follow the pain.
15:00You only go to the doctor when something hurts, right?
15:03Ooh, I have an owie, right?
15:04And then you go get it handled.
15:06And I'll give you a specific example of a piece of pain, all right?
15:10So here's a big energy company, and for years they've been pouring millions of dollars to
15:16construct an AI for their outage and maintenance services.
15:19Now why do they care about that?
15:21Because they predicted if they had such a thing, they could reduce the downtime of nuclear
15:25power plants by half if they had such a thing, meaning they don't have to spin up fossil
15:30fuel burning plants.
15:31It makes the grid more resilient, especially in deep summer and deep winter.
15:36Now there's a dark side to what I just said.
15:40If you ever watched any documentaries on Three Mile Island, when Congress did an investigation,
15:44we were only within 30 minutes of a Chernobyl-like event that would have irradiated the Eastern
15:49seaboard.
15:51And here's what they found.
15:53Six major issues, and I'm oversimplifying.
15:55One was a design flaw.
15:56One was a faulty valve.
15:58Four out of six issues were knowledge management issues where engineers and technicians were
16:02not rapidly getting the pieces of information that they had.
16:07So for years they've been trying to figure that out.
16:11We had them in production for the spring refit cycle of the reactor plants in less
16:15than five business days.
16:18So those are some of the outcomes.
16:20So frankly speaking, you're going to have to build the rapport and relationships and
16:25trusted to become a trusted advisor to these organizations and essentially follow that
16:30pain.
16:31I know it seems very simple, but it's amazing how many of us think that we show up and we
16:37know their use cases in terms of, hey, this must be your sales enablement pain.
16:42Hey, this must be your technical pain and things of that sort.
16:46And then we're not actually sitting there keeping our mouth shut and actually listening
16:51to them and following that pain ourselves.
16:54How do you get to the point where they trust you?
16:56Like from the moment you go to an enterprise, how do you make them trust you enough to give
17:02you enough people for you to see how that pain point then goes down the line and creates
17:09this massive cost that you could potentially reduce?
17:15Because I tell them, I understand your serious environment that's highly regulated and there's
17:19certain controls that are necessary.
17:21So my team may not even be ready with all the things that you need.
17:26And so you're not allowed to become a client of ours.
17:29And so we start this journey years ago with them to start building a rapport and relationship.
17:33And I start saying, though, hey, start, I don't want you to be a client, but let's have
17:38our R&D team starting to work together.
17:41Think of us as a trusted advisor on the outside.
17:44Start bringing us into the environment where we can start working together.
17:48That insurgent style relationship building actually works because then we become deeply
17:54embedded in the organization.
17:56Look, whenever you're servicing these entities, you're not Johnny on the spot.
18:00You're not always there.
18:02So somehow when you leave and your competitors show up, they have to become Teflon coated
18:08and they know that you've put in that sacrifice.
18:10You've worked with them, you solve their problem and your product management teams keep absorbing
18:15these requirements because at a certain point in time, the two entities are going to meet
18:20and say, yes, you now have, let's say in DoD, the impact level five and six that you need
18:25for classified information.
18:28You now have some of the HIPAA stuff or PCI stuff that you actually need to operate in
18:32their environments, but you don't just show up on the first day that you have those certifications.
18:38I probably was there years ago.
18:40Yeah.
18:41Right.
18:42I was probably there years ago.
18:43That makes a lot of sense.
18:44Stefanos, how do you see, you know, what kind of lessons have you learned at Seeker?
18:48What works, what doesn't work working with enterprises?
18:52What works and doesn't work?
18:53Yeah, I think that one big lesson learned is doing the cost benefit analysis of, you
19:01know, what model you're going to use for which application, really understanding your application
19:07and doing the cost benefit analysis of where you're going to apply this model, what hardware
19:12you're going to use can go a long way.
19:18Large language models are not a panacea, right?
19:21They're not going to solve every problem that you may have on your business or your
19:25use case, and they're also very expensive to operate, and, you know, even if, like I
19:32said earlier, like when you, even if you don't operate your own kind of large language models
19:38and use a closed source one, you have the issue of trust and control.
19:46You can control it.
19:47You can trust it.
19:48You can change it.
19:49You can understand it.
19:50So, I think, really, lessons learned is understand the problem, do the cost benefit analysis,
19:58choose the right solution for the problem, and focus on your ROI.
20:04Okay.
20:05Stefanos, while we're at, you've got the microphone, we could also talk about what is the biggest
20:10challenge that you see with companies deploying Gen AI?
20:15It kind of flows from what you were just talking about.
20:19If they decide to do Gen AI, what are the problems that they would encounter?
20:23Yeah, challenge, again, it goes back to cost.
20:27It's very expensive to build it, make it, build it, deploy it.
20:34That's why you have to be smart and always do the cost benefit analysis of what is the
20:41right solution for your problem.
20:44And again, the other challenge is trust and control.
20:48Yes, closed source is a little bit less expensive, but you can trust it and you can control it.
20:57Okay.
20:58Thank you.
20:59Shashank, anything in terms of the biggest challenge that you see with them deploying
21:06Gen AI solutions?
21:08Yeah, so the one thing that we have seen work is AI agents, and I want to speak about AI
21:15agents.
21:16Let's say you are a small company and you had few employees and you want to model a
21:21business where you, if you had more employees, what would you do with them?
21:26And every small and medium sized business has this hard bottleneck where they can't
21:32hire more people, so they don't think beyond that.
21:34But with agents, let's say you can deploy agents and we have agents which run on a very
21:39tight guardrail of ERP.
21:41So you have an agent that can do accounting, you have an agent that can do purchase, agent
21:45that can buy, manufacture, and so on.
21:47So you can then simulate your business.
21:49So let's say you simulate your business on Gen AI and you say, look, I have 50 employees
21:55and how does it look like?
21:56How would it look like if I grow at this pace for the next three years?
22:00And if I, you know, invest this much from my business from, you know, taking a loan
22:05or raising a financing, AI agents can do a very good job at simulating your business.
22:12And I think that's what we have seen, A, first of all, it's very, very less harmful than,
22:18you know, operating your business.
22:20It's just simulating it, but B, it gives you ideas that would not necessarily come from
22:27your operations.
22:28And there's something more that I'm writing a paper on generative agents, which are sort
22:35of, you know, there's something called GAN, adversarial agents, adversarial network, sorry.
22:41So we're writing up, I'm writing a paper on adversarial agents.
22:44So let's say you have two agents, you have, you know, two employees and their goal is
22:48to, you know, achieve some, they have individual goals, but they're adversarial goals.
22:54And you're just simulating that, those goals.
22:56You have a manager, you have an employee, you have a manager and two employees and five
22:59employees and you're then seeing process improvements.
23:03You're asking questions that you would not necessarily see answers to in your regular
23:08operations.
23:09For example, if you have three or four offices and you buy stuff from, you know, overseas,
23:13you sell in the US and you have a, you know, you want to see what's your labor cost going
23:18to look like for the next 12 months, you can actually run a very good simulation on agents.
23:23And that's what we have seen small, even medium sized companies with 50 employees or 100 employees
23:28use and then get more ideas on process improvements.
23:33The second thing is that the process boundaries are going to dissolve.
23:37So an accountant can also do purchase, a marketing person, why can't a marketing person do, you
23:43know, better operations?
23:45So in traditional, you know, companies, medium size, small enterprise, it doesn't matter.
23:50Traditional companies like manufacturing, industrial companies, they have been stuck,
23:55you know, with the business process for like 10s of years, even hundreds of years in some
23:59cases, like industrial age, right?
24:01They've been doing the same thing over and over again for hundreds of years.
24:04And the reason they can't change that is because the risk of changing that in real world is
24:08very high.
24:09I mean, you can't raise capital as easy as, you know, a West Coast entrepreneur scan.
24:15I have raised a lot of capital.
24:16So I know, you know, how a small company struggles with raising capital.
24:20So if you can't raise capital, you don't have use cases.
24:22You can't experiment, you can't take a risk.
24:23If you can't take a risk, then you don't have success that can then define the next, you
24:28know, generation of companies or the next generation of business processes.
24:32So Gen AI can help you simulate a lot of use cases in, you know, environments which were
24:38hitherto untouched.
24:40And that's the most exciting thing.
24:41So you can, you will probably have, I mean, people will talk about, you know, jobs going
24:47away and, you know, people getting lesser jobs.
24:49But I don't, I don't believe in that.
24:51I think AI is a good tool for humans to use just like a calculator is.
24:55As I said, it has limitations, but it can be useful.
24:58And I see agents.
25:00I see, you know, very helpful co-pilots that can enhance a small business.
25:07And then again, I keep saying that enterprise, large companies can use AI or not use AI.
25:13They are still going to be growing at the same rate.
25:16But if small companies can grow even 5% faster than they are growing at, our economy will
25:21go 5% faster.
25:22It's a very straight mathematical calculation.
25:24Thanks Shashank.
25:26Igor, if you look at how rapidly the AI landscape is moving as a, as a startup company, a very
25:36successful startup company, but still a startup company, how do you make sure that you have
25:41an enduring business value, that your business model and your competitive advantage, you
25:50know, stays relevant and that, that you're not getting competed out by, by the new startups
25:59that are coming?
26:01Only two out of, there's 10,000 reasons a startup could fail.
26:06Only two out of 10,000 are probably some competitive threat.
26:10The rest are self-inflicted wounds.
26:12You're not taking care of your clients.
26:13You have the wrong pricing, the wrong packaging, your code is buggy and things of that sort.
26:18There's a lot under your control.
26:20You really do have to find equilibrium between R&D and go to markets.
26:25This generative stuff, I mean, we were using a lot of these things, but didn't know the
26:29words, right?
26:30We were doing cloud before cloud existed, using neural networks before people knew these
26:34things existed.
26:36In some ways, I wouldn't oversell it because I can tell you in the environments that we
26:42find ourselves in, you have compliance and legal officers that when Gen AI shows up in
26:47their environments, they shank them.
26:49They literally shank them.
26:50They don't want any part of them because do you want hallucinations in a nuclear power
26:53plant, in a hospital, in an air base and things of that sort?
26:57So the things that actually operate the world as we know it, they don't want any part of
27:02that.
27:03They want determinism in some ways.
27:05Some of you are like, well, let's figure out how to guardrail these things.
27:11They're developed with stolen content.
27:14Let's call a spade a spade, right?
27:16Now the regulatory environment and the legal environment has to figure that out between
27:21the publishers and the AI industry.
27:25But trying to guardrail these things with whatever fancy agents you want to use to try
27:31to do that is like trying to keep Donald Trump on a teleprompter.
27:35You may be able to get away with it for about 30 minutes or so while he delivers his speech
27:39in the Oval Office, but later that night he's on social media and letting you know what
27:43he really thinks about what he previously said and it could be even contradictory.
27:47Okay, Igor, let me push back on that a little bit, right?
27:50So as a consumer, right, even as a consumer, I see Chat GPT-4 is doing great, so I'm paying
28:00for that.
28:01And then I see Anthropic has got Cloud 3, that's awesome, I'm getting that.
28:05And I've got Perplexity, and that's great, I'm getting that.
28:09But at some point, these models are like, I'm not going to pay for all these models,
28:14right?
28:15But as a startup that I'm putting my future on one or two, how do I know?
28:22Because it seems like it's a race, right?
28:25Chat GPT is good, and then Cloud 3 comes in, and then GPT-5 is going to come in.
28:31And then, you know what I mean?
28:33It's moving way too fast.
28:35Yeah, as a child of the 80s, I remember watching war games and what's the lessons learned there?
28:41What's the last scene?
28:42The only way to win is to not play, okay?
28:45So there's two ways to make a hamburger.
28:47You can take a cow and turn it into a hamburger, or you can be inspired by a cow that could
28:52make a hamburger, but you go over there and you get some potatoes and some carrots and
28:56some turnips and turn it into a hamburger, right?
29:01So an LLM could just model the language.
29:03It doesn't have to actually produce the answer, and you're not going to be able to use any
29:08answers that an LLM emits in any sort of highly regulated industry and serious construct.
29:15So that's how you can still remain relevant.
29:18And look, there's two ways for startups to be born.
29:21Let's not think that there's only one way, because when I started the last company, everybody
29:26was looking at this famous company in UK called Spinvox that had, I remember Guy Kawasaki being
29:33at the TechCrunch.
29:34He's like, Igor, I use their product.
29:36It's fantastical.
29:37You have one and a half million in funding.
29:39They have almost 300 million in funding from Goldman Sachs.
29:43They cratered.
29:44BBC did an investigation and found that they were a mechanical Turk.
29:47They were lying about doing speech recognition.
29:50When this company started, everybody was saying, whoa, why are we going to get this
29:55thing funded?
29:56Luckily, we count, you know, Steve Case and his team as some of our biggest earlier backers
30:01as well, and I appreciate him working with us.
30:05But everybody was looking at LLM and AI, right?
30:07The famed team up in Montreal, 150 million in funding.
30:11They were going to corner the market on enterprise AI.
30:14They got cratered.
30:15Then Reid Hoffman shows up with his inflection thing with a billion plus, and everybody's
30:19like, oh my gosh, that's going to, you know, take the wind out of your sails.
30:23There's two ways to make a company, ladies and gentlemen.
30:26You can grow it in the rainforest of the West Coast where there's plenty of resources, food
30:30and water to tend to the creation of these organisms, or you can grow in a desert.
30:37And economics is a study of the allocation of scarce resources.
30:43I've always bet on the desert creatures.
30:45The ones that are rise of the rest that grew up in other markets know how to be more resilient
30:50creatures and adapt to these market conditions.
30:54And I don't need, you know, billions of dollars of Jensen's parts in order to operate.
30:59Okay.
31:00Thanks, Igor.
31:01Okay.
31:02So since we only have a very few minutes left, very succinctly, Stephanos, what are you looking
31:09forward to see for AI in general, not just Gen AI, in the enterprise, and now I'm going
31:16into 2025?
31:18Yeah, what I think is going to be amazing to see in 2025 is basically businesses and
31:26enterprises being enabled to build their own AI solutions, build their own models that
31:34on their own data that they can trust, they can control, they can explain, they can change.
31:40And you know, I think that there is an added competitive cost.
31:45I think there are going to be some amazing opportunities to do that.
31:50And looking forward to that.
31:52Great.
31:53Thank you.
31:54Shashank, what's your?
31:55Same discussion.
31:56Build versus buy.
31:57For the next 12 months, build versus buy, half or close to 70% of teams will build and
32:03then fail, and then they'll go and buy.
32:06And this is, you know, tail as well as time cloud, you know, everybody wanted their own
32:10cloud and then eventually sell around three or four brands.
32:12So build versus buy is the biggest, biggest conversation right now in enterprise.
32:16And you know, your recommendation of, you know, you'll go around and then learn your
32:19lessons will probably hold true.
32:21So that's one thing.
32:22But I personally believe for small businesses, and even for enterprise agents, that can be
32:27guardrailed.
32:28There's a lot of reporting solutions.
32:30I've not seen any promise because, as I said, you know, if it's 1% faulty, it's 100% faulty.
32:37But we'll see more progress in reporting.
32:39Great.
32:40Thank you.
32:41Igor, what do you see coming up in the rest of this year and in 2025?
32:44All right.
32:45Here it is.
32:46I'll unpack it.
32:47There's regulation coming, isn't there?
32:50And so let me share something with you.
32:52Thanks to you first and then US.
32:55A good country lawyer knows the law.
33:00A great country lawyer knows the judge.
33:03All right.
33:04So some of us are going to be doing what?
33:07Waiting for these things to show up.
33:09The EU AI Act, what's happening in California, the executive order.
33:13And some of us are going to be part of the negotiations and the constructs and then figuring
33:19out how these things mutate.
33:21Again, to have proper, you know, responsible AI being birthed, ensuring, as Steve made
33:28in his previous comments, that we can democratize access to AI so that there's opportunities
33:33for companies of all sizes, small, medium, and large.
33:37And so that's the thing that over the course of the next 12 months we should be paying
33:41attention to.
33:42Right.
33:43And also what I've seen from enterprises is that they have been very ambitious in the
33:47last 18 months.
33:48You know, they start off, we were having a conversation with a good friend of mine that's
33:52here with, like, 500 AI projects and they have to filter it down to a reasonable number
33:59that they can actually put the investments that it deserves in order for it to go from,
34:04you know, proof of concept to the actual deployment.
34:07Thank you very much for your attention and your time.
34:10Really enjoyed it.
34:11Thank you to my amazing panel.
34:12And have a wonderful rest of the day.
34:22Bye.
34:23Bye.
34:24Bye.
34:25Bye.
34:26Bye.
34:27Bye.
34:28Bye.
34:29Bye.
34:30Bye.
34:31Bye.
34:32Bye.
34:33Bye.
34:34Bye.
34:35Bye.
34:36Bye.
34:37Bye.
34:38Bye.
34:39Bye.
34:40Bye.
34:41Bye.
34:42Bye.
34:43Bye.
34:44Bye.
34:45Bye.
34:46Bye.
34:47Bye.
34:48Bye.
34:49Bye.
34:50Bye.