• 6 months ago
A bevy of companies have released revolutionary AI and LLMs that have already changed people's lives. What could come next and who are the entrepreneurs working on that technology? This panel from Imagination in Action's 'Forging the Future of Business with AI' Summit convenes entrepreneurs working at the forefront of AI to talk about what's next and what they hope for.

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
00:00Hi, everybody.
00:06With that introduction.
00:07But hello, everybody, and it's really exciting to see all of the people in the room and all
00:12the energy here and to return to the stage.
00:16I've been part of the Media Lab ecosystem.
00:21I was part of the Media Lab ecosystem on and off for a good 20 years, this building and
00:25the building over.
00:27And the energy that you feel in this room is very much the energy with all the other
00:31sort of technological breakthroughs we've seen over the past couple of years.
00:37And I think one thing that's really interesting is that if you go back to the first event
00:42I went to in one of these two buildings was in the 90s.
00:45And at that point, a lot of people in the audience would identify the vertical that
00:48they worked in as the internet.
00:49Right?
00:50Do you remember?
00:51How many people remember?
00:52Raise your hand if you remember when internet was a vertical from a dropdown menu.
00:57That's awesome.
00:58And I think AI is really in that place right now where the breakthroughs that we've seen
01:01in AI are pushing it forward from being a category to something that is table stakes
01:07and literally everything we're doing.
01:09Everyone on this fantastic panel is working on using artificial intelligence or helping
01:14companies use these breakthrough AI techniques to do something really awesome.
01:19And so I'm going to turn it over to the panel right now to give a quick introduction of
01:23themselves and what they're working on.
01:25AJ, do you want to start?
01:27Hey, everyone.
01:28I'm AJ Keller.
01:29I'm the CEO and co-founder of Neurocity.
01:32We make wearables for mental health.
01:34I am a computer engineer from Clemson University originally.
01:38So my first time up here at MIT.
01:41But I live in San Francisco now.
01:43Really trying to leverage a lot of the adversity that I've gone through with my own mental
01:48health issues into creating solutions for the next generation of kids and the current
01:54generation of people who have contraindications with pharmaceuticals.
01:59Can we leverage AI and wearables to really take a chunk out of big pharma's pocket or
02:04just work complementary to them?
02:06But that's what I'm doing here, spreading the word, and just excited to be here.
02:11Thank you.
02:12Alan Chhabra.
02:13I grew up down the street here in Boston.
02:15I'm a Course 2 and Course 16 alumni of MIT, and it's an honor to be back.
02:22I work for a company called MongoDB.
02:24I'm on our executive team.
02:25I've been there for nine years.
02:27If you don't know what MongoDB is, we believe we're the most popular data platform in the
02:31world.
02:32We have 50,000 paying customers, over 300 million downloads of our product, and hundreds
02:37of thousands of applications are built on MongoDB.
02:41And our customers are demanding that we help them take those applications and make them
02:45Gen AI ready.
02:47Our platform combines operational meta and a vector database together, and what I do
02:53for Mongo is I bring partners together, like the companies on this panel, and I'd love
02:58to share how that's helping drive innovation.
03:02Thank you.
03:03Brendan.
03:04I grew up in Menlo Park, went to school for a couple of years, and dropped out, and founded
03:08Mercore, which uses AI to automate talent assessment, similar to how a human would manually
03:14review a resume or join a Zoom room and conduct an interview.
03:17We use AI to perfectly emulate those processes with a talent pool of over 300,000 people,
03:23where we work with a lot of the top AI companies in Silicon Valley to help them when they want
03:28to hire thousands of people in a given year, to query over and understand the exact background
03:33of an individual that they should add to their team.
03:38I'm Raj Agarwal, I work at a company, AWS, a cloud computing platform.
03:44And my role there is to build the generative AI capabilities for all of our go-to-market
03:50teams, and through that end up being the first design partner and customer for all of the
03:57internal generative AI capabilities that we're building and putting out to the market.
04:02Prior to this, I had built and sold two companies here in Boston.
04:06That's probably enough about me.
04:09So the title of this panel is Next Breakthroughs in AI, and I think one of the really good
04:15questions is, we've seen so many interesting technologies, hardware, software applications
04:20over the course of this day.
04:22Which breakthrough do you think your particular applications and your customers that you work
04:28with would most benefit from that could happen in the next year, and which breakthrough do
04:33you think you're going to get?
04:35In a dream world, what breakthrough would you get, and what breakthrough do you think
04:38is actually going to happen?
04:39Actually, let's start at the other side.
04:42So I'll start with Raj.
04:43Sure.
04:44Maybe on the granular or the maybe nearer term side of things, a lot of people are experimenting
04:51with RAG, and then they're trying to bring that together with just the LLM, right?
04:57And right now the results that a lot of people are getting are a little bit, for enterprise
05:02use cases, are a little substandard.
05:06They're not quite cutting it.
05:07They're not giving the right quality of answers.
05:12Still hallucinations is part of it.
05:14So I think a near-term breakthrough would be to figure out how to really bring that
05:18data that you're retrieving connected to the model so that you've got the best of your
05:24internal company data and the best of what a chat GPT and other LLMs can do in one single
05:31package.
05:32Yeah.
05:33Super interesting.
05:34I am really interested in sort of the breakthroughs that could happen in model self-improvement,
05:39where right now there's a workflow that researchers at Frontier Labs go through where they identify
05:44a weakness in a given model, and then they curate a data set that they can train that
05:50model on to help improve its abilities in a given area.
05:54And this process is run by researchers, done relatively manually, but soon the models are
06:00going to be able to identify their own weaknesses, to curate the kinds of questions, the kinds
06:05of prompts that they need answers to, to learn from and solve those weaknesses, and then
06:10distribute those tasks to the humans anywhere in the world that can solve them.
06:14And I think once we get there, it'll be one of the most profoundly impactful things on
06:19model self-improvement and just sort of the foundational technology that all of this is
06:24built on.
06:25My hope is that that's for the better.
06:28My fear is, you know, some risks associated with where that leads us.
06:32That's great.
06:33I would say two breakthroughs.
06:35First, we've already seen that developers are taking over the world.
06:42Developers around the world are writing software, helping companies innovate, stay ahead of
06:46the competition, startups, enterprises.
06:49I think the advancements in code generation and co-pilot tooling with AI allows anyone,
06:55wherever they are in the world, whatever age, to learn how to code better and faster.
07:01So the more people who can code using co-pilot and code generation is going to just cause
07:06so much more innovation and great production of applications to help customers.
07:11I think that's probably what I'm most excited about.
07:15The more developers who can do it better, faster, it's going to be better for the world.
07:19I think second, six months ago, especially in the enterprise customer segment, which
07:24MongoDB I spent a lot of time with, there was a fear of AI, trust, is my data secure?
07:32Is my compliance, governance?
07:34And I think vendors like AWS and Mongo, we've learned how to help them with these types
07:41of problems so that they become less nervous and trust AI more.
07:47And that's going to let them take that leap of faith to go innovate their applications
07:51and their estate.
07:56The breakthrough that I want to happen is a large foundational model for time series
08:00data that essentially takes brain data in from multiple sources, whether that's electrical
08:07activity, imaging data, focused ultrasound, and essentially is able to create a more cohesive
08:16embedding and essentially be able to create a whole new layer of applications for monitoring
08:24brain activity that are based on this embeddings layer.
08:27If we can remove the complexity of having to acquire copious amounts of data that is
08:34required for making any sense of brain activity and augment that with a foundational model,
08:39we really stand at this amazing breakthrough of really meaningful software applications
08:46being built surrounding AI and surrounding the brain.
08:52What I think will happen is we're going to release a first foundational model.
08:55We're going to do it open source.
08:58And I really want to see pickup of the academics and industry really help make a better collaborative
09:05model so that we can all have a better application suite for our users.
09:10I think underlying all of this is sort of a bunch of architectural questions that we're
09:15still really hashing out with regards to how are we going to operationalize using generative
09:21AI and large AI systems.
09:25And there's almost two schools of thought, right?
09:26There's been a lot of momentum towards building data lakes and the like, towards a digital
09:33transformation, towards putting all of data in one place and then building an insights
09:38layer on top of it.
09:40But if we look at RAG and we look at some of the distributed open source projects that
09:44have come out, we see also a multi-cloud, multi-silo almost approach where we're using
09:51where the data is more distributed and the insights are more centralized.
09:55Alan and Raj, can you talk a little bit about what you see in the customers with regards
10:00to sort of these two different ways of approaching data architecture and where you think it's
10:03going?
10:04I mean, getting your data is the most valuable thing that you have and putting it into action
10:10is the whole game, right?
10:12But the problem in most companies is that data is very difficult to put together.
10:18It's messy, it's different structures, different taxonomies.
10:23And so, you know, even pre-generative AI world, right, like that's the big challenge is how
10:28do you clean this data, how do you make it usable, how do you pipe it together?
10:33And so that continues to be one of the reasons why you don't see as many things in production,
10:39right?
10:40So you've got a prototype, you've got a little sample data set, you create something, CEO's
10:44excited, right?
10:45But then to go from there to in production, there's this huge amount of messiness in the
10:50data.
10:51Now, if you're a multi-cloud or your combination of on-prem and cloud, it's even messier.
11:00So look, if you are in one cloud, maybe arguably there's some advantages there.
11:05You have same access permissions and you don't have to do as much data piping.
11:10But ultimately, you still need to get that data into the prompt, let's just say for simple
11:14use cases, right?
11:15You still need to get it into the prompt and it still needs to be structured.
11:18So what I look forward to, maybe this is the last question, is breakthroughs.
11:22Can generative AI help you automatically clean that data and structure it and make it more
11:27usable?
11:28So I think we'll get there.
11:29But today, companies are struggling with that and it's probably a big opportunity for the
11:35Accentures of the world to come in and earn some more fees and help them make that usable.
11:41I would echo what Rod said and I think there are a lot of companies, startups in the room,
11:46people building out go-to-market as a startup in the AI world.
11:51It's important to know if you are targeting enterprise customers, they have a challenge
11:56with data, especially in AI makes it even more difficult.
12:01A line of business for an enterprise is going to want to figure out how to use AI tomorrow.
12:06How do I make my apps better, smarter?
12:09How do I deploy rag architectures to make me leaner and faster?
12:12A line of business enterprise wants that.
12:15On the flip side, ops and sec ops, they're under a lot of pressure.
12:20They want to keep the data on premise.
12:22They want to put it in the cloud.
12:23Do they want it multi-cloud?
12:24How are they going to deal with the compliance and governance and data sovereignty?
12:29If you are a startup or even a more mature company vendor selling to the enterprise AI,
12:35you need to understand that that struggle isn't actually made easier on AI by AI.
12:42It actually makes it more complex.
12:43You need to be able to communicate with them that first you understand that pain and I'd
12:47recommend you go in with messaging of how you can bring those two groups together.
12:53If you can bring them together, 100% innovation will happen.
12:58You mentioned that a lot of people in the room are startups or thinking of starting
13:02companies or somewhere in that journey.
13:05Breakthroughs for startups can kind of be a double-edged sword.
13:08With the rapid pace of innovation, it can be really difficult to figure out what to
13:12work on and what's a safe area to be building technology, be building a moat.
13:18Brendan and AJ, can you talk a little bit about how that decision has worked for you
13:22guys and what you think about moving forward and what advice you would give?
13:25Yeah, absolutely.
13:26I think obviously the models are going to get really good really fast.
13:29The key thing to focus on is what kind of data sets or code bases you should curate
13:34to prepare for that.
13:35I think right now, people aren't really thinking about their post-training data set in particular
13:41in the right way where people are going to start to realize that similar how you would
13:44iterate a code base across 10 years and that's where the core enterprise value of a SaaS
13:49application might lie.
13:50They're going to start curating their data sets for post-training, investing huge amounts
13:55of money into getting well-structured data that has the right tokens in it to give the
14:01model whatever abilities they need.
14:04We're already seeing billions and billions of dollars flowing into this from both the
14:08Frontier Labs as well as application layer companies.
14:11All right, come on, come on.
14:25Yeah.
14:26Yeah.
14:27Yeah.
14:28Yeah.
14:29Yeah.
14:30Yeah.
14:31Yeah.
14:32Yeah.
14:33Yeah.
14:34Yeah.
14:35Yeah.
14:36Yeah.
14:37Yeah.
14:38Yeah.
14:39Yeah.
14:40Yeah.
14:41Yeah.
14:42Yeah.
14:43Yeah.
14:44Yeah.
14:45Yeah.
14:46Yeah.
14:47Yeah.
14:48Yeah.
14:49Yeah.
14:50Yeah.
14:51Yeah.
14:52Yeah.
14:53Yeah.

Recommended