The Biggest Opportunities For Businesses Utilizing AI

Forbes

In a world where AI is evolving rapidly, how can AI experts and entrepreneurs harness the rapid evolution of the technology? This panel at Imagination In Action’s ‘Forging the Future of Business with AI’ Summit of Rana el Kaliouby, Danish Goyal, Lin Qiao, Will Coffee and Yair Adato looks at what entrepreneurs can make the technology work for them.  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.

Transcript

00:00 So welcome everybody. My name is Rana Elkayoubi. I actually spun out my company Affectiva out

00:06 of the Affective Computing Lab right here back in 2009. Sold the company a couple of

00:12 years ago and I've been spending a lot of time investing in AI startups and I just feel

00:17 very lucky to be in this place in this moment of time experiencing all of the incredible

00:24 innovation we're seeing. So our panel today is going to be about how do you cope with

00:28 the rapid speed at which AI is accelerating. How many people here feel that every day every

00:34 week there's new advances and it's hard to keep up. It is hard to keep up. So I'll give

00:41 an opportunity to each of my panelists to introduce themselves and then we'll dive right

00:45 in. Sure. Thank you Rana. Manish Goyal. I'm an operating partner at Berkshire Partners.

00:51 Berkshire is a private equity fund based in Boston. We invest in mid-sized companies across

00:57 sectors consumer health care industrials business services and tech. So a broad swath of the

01:03 economy. I lead our work on AI and data science. And what that means is in that role I get

01:11 a chance to work with all 30 plus of our companies and I'm working with them directly on some

01:17 of the challenges that you mentioned that they're facing in terms of how to actually

01:22 introduce this into the operations and how to train and equip their workforces. Excited

01:27 to be here. Amazing. Just quick follow up. Do you have like a blueprint that you take

01:32 into these companies to help them become AI driven? So the way Berkshire works it's more

01:40 doing it in partnerships. We don't have a blueprint but what we do have is we run an

01:45 AI working group which brings in a lot of experts. And then we found the most value

01:51 is actually solving the last mile problem which is actually tailoring each of these

01:57 use cases for the companies. So we go in we work with them on the highest business problems

02:02 and that's where we found the biggest user. We can talk more about that. Great. Lynn.

02:06 Hi everyone. I'm Lynn. I'm CEO co-founder of Fireworks AI. Before Fireworks I've been

02:12 at Meta for seven years mainly building out Meta's AI platform supporting all like billions

02:19 of users across all product lines. And we started Fireworks because it took us more

02:25 than five years to bring Meta to be AI first. We really feel like the whole entire industry

02:31 is shifting to be AI first at the same time. And we want to reduce the time to leverage

02:38 AI from five years to five months to even five weeks. So that's our mission. And right

02:44 now we are laser focused on general AI and we provide the fastest and lowest TCO platform

02:51 to enable all application builders and products building on top of GNI. So I love your origin

02:58 story. You were part of the core team that built Meta's PyTorch which a lot of people

03:04 here have probably used you know if you're in the AI space. So what was the what was

03:09 the kind of the motivation for leaving Meta and starting your own thing. Right. So I was

03:15 head of PyTorch when I started. PyTorch was a five people team and grew that into 300

03:21 people and also took us five years to fully production PyTorch supporting all Meta's surface

03:26 areas of AI. That's quite a journey. At the same time I was through PyTorch open source

03:33 engagement. I was talking with many companies as partnership and realized oh you know no

03:41 matter how big or small those companies are there's a huge desire on one hand to be a

03:46 first and do the transition really fast and be in this way leveraged or even ahead of

03:52 wave. But on the other hand it's significant lack of expertise lack of infrastructure lack

03:57 of resource lack of know-hows that slow them down and that big contrast make me feel we

04:03 actually can make a huge impact and we can significantly reduce the time to market bring

04:09 the business impact and business value by alleviate operation aspect and kind of help

04:14 those companies transition faster. Now it comes to Gen AI and this velocity of change

04:20 is significantly improved. It's going crazy right now. As you just said as of now today

04:29 Meta just released their newest model Lama 3 and last week earlier this week we just

04:36 enabled a new mixture model. So the velocity of change is constant and we are here to help

04:44 all the like applications built on top of Gen AI to do not worry about that and just

04:49 leverage the platform we have and focus on your product building.

04:53 Amazing. Will?

04:55 Hi guys. My name is Will Kofel. I am a longtime startup founder actually MIT alum. My first

05:02 startup was I was in the founding engineering team at Akamai Technologies when we spun out

05:07 of MIT in 1998. So some of some of you here are old enough to remember the dot com boom

05:11 the first time around. So it's great to be in another exciting revolution here. After

05:17 a long long stint of doing kind of every role in the startup ecosystem I'm currently leading

05:22 NVIDIA's Inception program which is our scaled startup support program. So we have almost

05:28 20000 startups that are in our program today including both of the startups on the stage

05:32 today mostly coincidentally actually. So a ton of great startups. So it's really amazing

05:38 to have this kind of front row seat to see the innovation happening around AI from this

05:44 place of kind of being able to support them. It's really awesome as a startup founder you

05:48 spend a lot of time focused on one particular company at a time. So I think we were always

05:52 jealous of the VCs who we felt had this portfolio approach to innovation you know and now I

05:56 get to get to see the scope of kind of 20000 startups innovating on AI. So super excited

06:01 to be here and always love this event. I have to say Affectiva my company was part of the

06:06 NVIDIA Inception program but I believe that was before you had joined NVIDIA and it was

06:11 just an amazing way to get access to an incredible ecosystem. You know as a startup you don't

06:16 necessarily have that access. So it was it was really powerful. So for any startups here

06:21 definitely consider the NVIDIA Inception program. Look that up and then also if you're a VC

06:26 here there's a lot of investors. We also have a great relationship through our VC alliance

06:30 with VCs both to sort of help them get access to startups to work with their portfolio companies

06:34 also so feel free to reach out to me if you've got either either role. I think my big question

06:39 for you is how do we get some NVIDIA AI chips. Yeah I have a bunch of H100s in my trunk.

06:45 So see me afterwards. That's awesome. I thought that this is for me. I promised Yair I would

06:51 give him my chips. Sorry. So sorry about the late or early. So my name is Yair I have a

07:00 PhD in computer science and this is a second startup that I participate. This one I'm the

07:10 founder and the CEO. A startup called Bria. We are doing responsible and open generative

07:16 AI. Last time that I was in Boston was 14 years back when I was when I did internship

07:24 at Harvard and my girls was one years old. Now they are 14, 15. So it's really nice to

07:31 come back to Boston. Great. Who are the customers of Bria? Who are the customers of Bria? We

07:38 took a different approach. We don't have application so it's not a B2C and it's not even for designer.

07:45 We develop the platform to enable other companies to use this technology and we give this technology

07:52 as an open source. The model themselves and the weights. So what we did, we developed

07:58 an attribution engine which every time that we generate an image we know how to trace

08:03 back in the training set which data influence the specific generation and we know how to

08:10 reward this data owner. Same like Spotify if you think about it. This is how the music

08:16 streaming started with Pirate Bay and then moved to Spotify which know how to reward

08:22 the data owner. So we did this attribution engine. With this attribution engine we have

08:27 agreement with almost 20 different image stock provider which gave us all of the catalog.

08:33 So every data that we are using is fully legal, fully licensed for generative AI. We train

08:40 it from zero and now we have a model which is perfect for enterprise to use it to build

08:48 their own innovation solution. So it can be WPP and publicists. It can be media companies

08:57 like Fox. It can be creative tools like if you think about companies like Canva and Figma.

09:03 They all need this model but there is no economic for them to create it from scratch even though

09:10 the chips are in the truck. It's quite expensive. Make sense? Yeah I love the emphasis on the

09:16 data. We've had a lot of conversations today about the importance of data in this AI equation

09:22 and I love that you are making sure that it's like the AI is explainable, the model is explainable

09:27 and it's a very responsible approach to building these models. So this is why we call it responsible

09:32 and open generative AI platform. It's not only the data ownership. It's also public

09:38 figures, privacy, trademarks, fairness, bias. If you start in using the data that build

09:45 for commercial use and you train the model with data that build for commercial use then

09:50 you get also a platform that you can use with less risks. So I want to ask you each about

09:55 the opportunities, the biggest opportunities you see as we kind of embrace being an AI

10:01 driven organization and I love that this panel is so diverse. We've got large enterprises,

10:05 we've got startups, we've got kind of big tech. So what are the opportunities you see?

10:12 Yeah so the biggest opportunities we've seen in companies is those are doing it really

10:16 well are people are thinking about both productivity use cases as well as revenue generation. So

10:23 the productivity use cases have been talked about all day, right? Conversational AI, call

10:27 centers, chatbots, knowledge servicing, etc. Where I think it gets really interesting is

10:33 when you're looking at your core business processes and thinking about how do you improve

10:37 them for a customer benefit. So just to give you one example which I particularly like

10:42 is say you are in the distribution industry and typically when a customer calls you and

10:48 says something is broken, the rate at which you can predict what part they want is roughly

10:54 50%. Now if you can move that number that the tech goes out with from 50% to 70% it's

11:01 a game changer. It includes customer satisfaction, it has a significant impact on cost, you can

11:07 stock a lot more and what's good about this is two other things. One is there is actually

11:13 low risk of hallucination because all I'm caring about is improving my batting average.

11:18 There's very low risk of hallucination and the second thing is there is something really

11:22 different about the data. A lot of the data which was sitting in customer service manuals

11:27 and old PDFs and call logs was not being used. So what we are trying to encourage all of

11:32 our companies to do is to think about the one or two use cases in their business model

11:38 where this really makes a difference and then working hard creatively to solve that issue.

11:44 And then I would imagine you could probably use a platform like Firebird.

11:48 Exactly. Which is great segue.

11:50 How would you help Amish's company?

11:53 I had a takeaway to follow both of you to understand how we can use.

11:58 Definitely. So I have a question to the audience. How many of you think GNI technology is eventually

12:05 going to touch your life? You will personally use it in one way or the other. Okay, we don't

12:11 have a very thoracic audience here. But throughout this event, I've been in a lot of lightning

12:18 talks, startup ideas, a lot of discussion about building all kinds of assistants. I

12:26 personally, our company has been supporting legal assistant products, medical assistant

12:32 products, coding assistant, all kinds of coding assistant products for productivity. Today

12:36 you don't need to write PowerPoint, you can generate PowerPoint just by saying what you

12:40 want. Or you can -- you don't need to type for web application building. You can say

12:47 what kind of web application you want to build. It will generate for you. So a lot of those

12:52 is coming up. And I do think this GNI is going to create a tidal wave of disruptive applications

12:59 and products. But all these products are going to be used by all of you eventually. Whether

13:06 you're a consumer or developers. It's going to benefit from that. So when you use those

13:13 products, we don't want to wait. We want to see results immediately and keep thinking

13:19 and processing and keep using the product. So latency is extremely important because it

13:23 has to come back to you quickly. And more so, this is a year that we no longer talk

13:30 about human as the receiving side of GNI. We talk a lot about agents. So there will

13:36 be a lot of agents playing different roles. They will self-organize, self-coordinate with

13:41 each other. So for one task, it will have multiple agents talking with each other to

13:47 finish that task. And then it becomes -- latency becomes even more important. We're not talking

13:52 about one second, half a second latency. We're talking about millisecond latency. So then

13:57 quickly those agents can -- through GNI technology can finish a task to improve productivity.

14:04 So all comes to, hey, those models -- at the same time, we know the models are very big.

14:12 We know that open AI's mission is AGI. AGI means building AI systems that's smarter than

14:18 human. Those models have trillions of parameters, very big. But it's kind of at the same time,

14:24 we know all these applications, we talk about whether you are on the receiving end of consuming

14:30 them or it's agents. It pushes to very low latency. So these two actually don't go hand

14:37 in hand. And what we believe in is we're going to invest in much smaller models, make it

14:43 on par quality with large models for solving specific problems. At the same time, the smaller

14:49 models, we hyperoptimize for those smaller models because all these models are models.

14:54 We know exactly how to optimize them to bring in the lowest latency to those products and

15:01 applications. At the same time, we will also bring the lowest TCO because, hey, those models

15:08 run on top of GPUs. We know GPUs are a very expensive resource. And kind of we make it

15:15 affordable to new application builders and product builders. So that's kind of very exciting

15:20 to me, that kind of direction.

15:22 And what Fireworks is doing is you're lowering the barrier to entry for any company, including

15:29 startups that want to leverage generative AI in their product portfolio. So Will, out

15:33 of all the companies, how many companies do you have in the Infection Program?

15:36 I think we're just over 19,000.

15:38 Okay. So in that kind of universe of startups that you're supporting, where would you say

15:45 the bulk of the opportunities are in the AI stack? Is it like, where is it?

15:50 Yeah. So I guess I'll take a bit of a cop out, but I think it's the natural extension

15:55 of this velocity. So if we really are in a place where we all agree there's huge amounts

15:59 of change, I mean, you never know. Maybe wherever we are on the hockey stick, you could debate.

16:03 But we're somewhere on that curve right now. And I think that that leads to the unique

16:08 opportunity here, which is actually across the ecosystem. What we see is that there's

16:13 opportunity at every single place in that stack. And what I would say is the more focus

16:18 that people can have right now on advancing the ball on one particular thing, whether

16:22 it's regulatory, whether it's platform specific, I mean, both of these companies have taken

16:26 like platform approaches here. There's massive opportunity in the pickaxes and shovels for

16:30 this revolution. But there's also massive opportunity for domain experts to understand

16:35 the implications of AI in their domain. So we see incredible legal experts, incredible

16:39 health care experts and biotech, I mean, bio IT world just wrapped here in Boston. And

16:44 we had somebody, we had a bunch of people from NVIDIA there. There's massive interest

16:47 in AI, not just for drug discovery, but for so much else in that world. So I think to

16:51 understand if you're playing in a domain specific applied AI area or in sort of a platform area

16:57 or in the much needed areas around AI bias and AI regulation and AI safety and AI open

17:02 source and all these things, I think the reality is that we're seeing opportunity across all

17:07 sets. And I guess the last thing I'd say on that is because that's the reality today,

17:12 I would encourage companies to be partnering as much as possible, like own the part that

17:18 you're going to move the ball for it on and then partner the heck out of everything else

17:21 because just that opportunity, like a rising tide is going to lift all boats. So the more

17:27 partnership we can do, I think the better.

17:29 Gary, you are in the kind of the AI space safety and responsible AI space. What do you

17:35 think about all of the upcoming AI regulations? Is that good news? And how does it affect

17:40 your business?

17:41 I don't think that I'm in the responsible AI space. I think that AI should come with

17:47 responsible and also with security and also with privacy. No one says, well, I may develop

17:53 an application and my application has privacy because it's clear. So the way we suggest

17:58 to think about it is that the same when you develop something, when you have go to market,

18:03 when you have design review and you think about elements like privacy and security and

18:08 scale, you need also to think about responsible AI, which basically means to take accountability

18:15 of what you do. If you don't take the accountability, I hope that the government will come and create

18:21 regulation that helps you to aim how the accountability should look like.

18:26 I will give two perspectives here. First, 10 years ago when we started with the privacy

18:31 regulation, everyone thought that this is crazy. I thought that it will not succeed.

18:35 And after two years, oh wow, it's actually a good thing, right? We are all in favor of

18:39 that today. There was some market education, but we all agree that it's important. Same

18:45 will happen with the AI regulation. The other side of it is that we need to be careful that

18:54 big companies will not use that to prevent other companies not to be part of the game

19:02 because there are great opportunities for platform and for startup, but also for existing

19:08 companies. And you see today companies like Publicis and WPP, which are the advertising

19:15 agencies, Red Bull, which is, and they are all hiring AI experts, computer vision, machine

19:24 learning groups to start develop that. And they need the source code. So we need to balance

19:30 the regulation that it will help society and will not create monopoly or do monopoly to

19:35 other companies. Absolutely. All right. So what is one piece of advice that you would

19:42 give our audience on how to stay up to speed with AI innovations? I would say just get

19:50 started, right? All of us are learning. All of us are behind the curve. Just get started,

19:56 have the confidence and try different things and you'll be fine. I always say my son is

20:02 15 and he is my, he's my AI mentor. He's always trying these new tools and breaking them. I'm

20:09 sure at some point he'll get into trouble, but I love the mindset. I'm sure he's teaching

20:14 you quite a bit. Yeah. And he gets all his information from TikTok. So anyway, how about

20:21 you? What's your advice for people to stay up to speed? Yeah. I would say there's a little

20:25 bit of fatigue of how, how, how many updates there are every day. Part of the fatigue is,

20:33 Hey, there's a new model coming up. What do I do? Do I try it or do I wait and see, you

20:40 know, how people react and so on. It has been a constant confusion and people feel like

20:45 tired of catching up with, I think just this week there are two new models, like two big

20:50 models got dropped. So that's where we come in as Fireworks AI. We actually provide, offer

20:58 more than a hundred models and we actually monitor them constantly and put the most state

21:04 of art as our featured model. And you can easily access them through our serverless

21:10 integration and super easy API, which is also open AI compatible. So I'm pretty sure a lot

21:16 of developers are very familiar with open AI's API and you can seamlessly migrate to

21:21 use ours and use smaller models so your TPO goes better and faster. So we hope we actually

21:27 alleviate that side of the concern and make your job much easier. And you just focus on

21:32 your next level of application product development and be innovative.

21:38 Will, how about you? Yeah, I mean, I guess as a personal piece of advice, I would, I

21:43 would remind us all, myself included, to spend more time hands-on with this stuff. It's

21:50 really easy for us to sit around and read the blog posts and read the newsletters and

21:54 all that. But like I would challenge people in this room, if you're not in there playing,

21:58 building, acting like a consumer, acting like a developer, then you're probably not

22:02 digesting it at the level that it's really, that you're really going to understand where

22:05 those opportunities are. So I think one piece of advice is, you know, get, get as hands-on

22:09 as possible. You know, and then if you're building businesses out there, you know, the

22:12 other piece of advice is just, just clean up your data. You won't be upset if you have

22:15 clean data. Absolutely. I'm going to, I'm going to, to take the geek answer and to reference

22:22 to a book and suggest don't panic. Don't panic? I love that. It's, it's, it's not

22:29 new. After Alex met and deep learning, we said exactly the same thing. And I got exactly

22:35 the same email from AWS that there is not enough GPU at the region in 2014. And we all

22:43 thought that it's all going to replace everyone, but the cars still have drivers and AI help

22:48 people. Don't panic. I love that. I love that as our ending. So get hands-on, but don't

22:57 panic. Thank you. Thank you very much.

23:00 [Applause]

23:01 [End of Audio]

23:02 1

23:03 Page PAGE of NUMPAGES www.verbalink.com

23:04 Page PAGE of NUMPAGES www.verbalink.com

23:05 Page PAGE of NUMPAGES www.verbalink.com

23:06 Page PAGE of NUMPAGES www.verbalink.com

23:07 Page PAGE of NUMPAGES www.verbalink.com

23:08 Page PAGE of NUMPAGES www.verbalink.com

23:14 Page PAGE of NUMPAGES www.verbalink.com

23:19 Page PAGE of NUMPAGES www.verbalink.com

23:24 Page PAGE of NUMPAGES www.verbalink.com

Category

Transcript

Recommended