How will generative AI impact the factory floor? In this conversation, Bill sits down with Jason Nassar, who is responsible for Dell Technologies’ global strategy for manufacturing edge products and solutions as a Senior Consultant Product Manager. The two dive into the future of generative AI on the factory floor, musing about the role of robots and small language models.
How will generative AI impact the factory floor? In this conversation, Bill sits down with Jason Nassar, who is responsible for Dell Technologies’ global strategy for manufacturing edge products and solutions as a Senior Consultant Product Manager. The two dive into the future of generative AI on the factory floor, musing about the role of robots and small language models.
---------
“I think as the technology improves itself and it's going to happen faster than we think, there will likely be robots and even humanoid robots that are going to be performing tasks that in the past people performed.”
“At the end of the day, these LLMs or large language models, they're best put in the cloud right now, large data centers actually processing this. But this is quickly going to move directly to the edge. And that means that this technology is going to be utilized on the factory floor.”
--------
Timestamps:
(02:00) How did Jason get started in tech?
(02:40) IT and OT in manufacturing
(06:45) Evolution of customer needs in manufacturing
(20:54) What is possible with traditional AI?
(14:54) The potential of generative AI in manufacturing
(19:00) Hallucinations and human error
(20:50) The future of work and education
(33:32) Robots on the factory floor
--------
Over the Edge is brought to you by Dell Technologies to unlock the potential of your infrastructure with edge solutions. From hardware and software to data and operations, across your entire multi-cloud environment, we’re here to help you simplify your edge so you can generate more value. Learn more by visiting dell.com/edge for more information or click on the link in the show notes.
--------
Over the Edge is hosted by Bill Pfeifer, and was created by Matt Trifiro and Ian Faison. Executive producers are Matt Trifiro, Ian Faison, Jon Libbey and Kyle Rusca. The show producer is Erin Stenhouse. The audio engineer is Brian Thomas. Additional production support from Elisabeth Plutko.
--------
Follow Bill on LinkedIn
Connect with Jason Nassar on LinkedIn
Read Jason’s latest blog on CIO
Learn more about https://edgeresources.dell.com/?industry=manufacturing
Producer: [00:00:00] Hello and welcome to Over the Edge. This episode features an interview between Bill Pfeiffer and Jason Nassar, who's responsible for Dell Technologies global strategy for manufacturing edge products and solutions as a senior consultant product manager. The two dive into the future of generative AI on the factory floor, musing about the role of robots and small language models.
But before we get into it, here's a brief word from our sponsors.
Narrator: Over the Edge is brought to you by Dell Technologies. To unlock the potential of your infrastructure with edge solutions, from hardware and software to data and operations across your entire multi-cloud environment. We're here to help you simplify your edge so that you can generate more value.
Learn more by visiting dell.com/edge for more information, or click on the link in the show notes.
Producer: And now please enjoy this interview between Bill Piper and Jason Nassar, senior consultant product manager at Dell [00:01:00] Technologies.
Bill Pfeifer: Jason, thanks so much for joining us today. I know you've got a long history of technology and integrating that into the manufacturing process.
And this should be a really fun conversation. Thanks for joining us.
Jason Nassar: Yeah, Bill.
Bill Pfeifer: Thank
Jason Nassar: you for having me. It's
Bill Pfeifer: been
Jason Nassar: a
Bill Pfeifer: pleasure. So let's start with a little bit of history. How did you get started in technology? How did you get here?
Jason Nassar: Yeah, it's been a long journey. I actually started hands on. I joined the military when I was young and I ended up being a technician for aviation equipment for five years in the Marine Corps.
I ultimately ended up Becoming a journeyman. And it was there where I realized that I really loved the idea of engineering and design and technology, electronic software. And so I ended up going to school at the university of Michigan, where I got my undergraduate in electrical engineering and my graduate in energy systems engineering.
And from there, you know, I moved on to the auto industry for two years. I worked for General Motors where I designed hybrids and electric vehicles, really exciting [00:02:00] job straight out of college. And then from there, I moved into energy. I worked for Siemens. I also worked for General Electric where I designed.
Started off designing gas turbine power plants controls, but ultimately became a product manager leading the technology in regards to that. And then, you know, there's a need for people with this operation technology expertise in the IT space. So, I spent a couple of years at HPE, where I helped develop their gateways and servers, and I led that product strategy there, and then Dell brought me in to lead, you know, the manufacturing strategy.
Bill Pfeifer: Yeah, bringing together that IT and OT. That's, they're two very different mindsets and they're really coming together as we digitize and transform. And that's easy to miss because, you know, each side has their priorities and their view of the world. And it's, it's such a deeply held view and so personalized.
that it's easy to miss that other people don't have [00:03:00] the same set of priorities and the same set of worldviews as you. So yeah, having, having someone on the team to call that out and be aware of both sides of the conversation is, is hugely, hugely valuable. So I know you spend A lot of your time talking to customers about how to digitize, modernize, transform manufacturing operations.
Can you talk some about how much change you've been seeing over the past couple of years?
Jason Nassar: Yeah, I mean, there are exceptions, but compared to the tech industry, change is extremely slow when it comes to the manufacturing space. And a lot of it. Makes sense. I mean, the real decision makers, when it comes to putting technology on the factory floor, they have this goal at the end of the day of producing as much product with the highest quality as possible.
And so they've really viewed putting IT on that factory floor as a hazard in many ways, because of the fact that, you know, IT equipment is susceptible to [00:04:00] viruses, if you're doing updates of that software, if it impacts the production on that factory floor, it would slow it down. And so, rightfully so, these folks, there was a lot of conflict between IT and OT, and that's what you would see a couple of years ago.
The biggest change that I've seen lately is that, These same folks with these OT or operation technology personas, they're really opening their mind. They recognize what's happening when it comes to modernization and digitization of what's out there in the real world. Everybody sees it on their smartphones, their computers, their daily work.
And they recognize, Hey, you know what? We have to bring this IT technology on the factory floor. And the only way to really get this job accomplished with low latency. And optimal efficiency is to try to bring the software applications local to where the operations are. They also have a lot of challenges with when they're implementing things, how are they going to scale it?
And so I've seen more trust in IT lately because IT, in general, is designed to [00:05:00] scale to businesses. And now we're talking about scaling to the edge, which is local to where that data is actually being created.
Bill Pfeifer: Okay. Well, and. I would imagine IT has been coming toward the OT side, as well as OT coming toward the IT side, right?
Younger versions of IT were all the move fast and break stuff kind of thing, and change, modernize as fast as you can, and some downtime is acceptable. And to OT, downtime is not generally acceptable. You know, it's Tens of thousands, hundreds of thousands, millions of dollars a minute or hour of outage, depending on how you're measuring and what you're measuring, you know, like 911 systems, telephone systems can't go down ever.
Hospitals can't go down ever. Bad things happen. So yeah, I can see why pulling too much IT in would be seen as a risk, but then I think IT has become more of the utility type service. And it's getting closer toward those lots of nines reliability as opposed to, I mean, when I was younger, [00:06:00] I was a little bit of a cowboy in terms of tech, like, it's fine.
It may go down for a while, but we'll bring it back. And they're like, no, you can't do that. So yeah, following my career path through that, I've become a little more reliable in terms of my planning. So I would imagine the whole IT industry has, has kept up with that as well.
Jason Nassar: Yeah. I mean, ultimately everything is being deployed right now in IT.
It's, it's, you're a few feudal at this point. You can't rely only on the OT companies to make these bigger decisions because the entire infrastructure of most businesses relies on
Bill Pfeifer: IT. Yeah. Pretty crazy. So it's interesting to hear how products have evolved and how, you know, solutions have evolved and things like that, but customer conversations evolve, I think, faster.
Then all of that stuff in terms of, you know, what customers are concerned with struggling with trying to figure out now is what's going to be [00:07:00] relatively commonplace probably in a year or three. Do you have a sense of what customers are struggling with trying to figure out like what's the leading edge of those conversations that you're having now?
Jason Nassar: Yeah, ultimately customers, they're looking for simpler ways to deploy their solutions at the end of the day. They want something that actually doesn't require IT to be there, but has all the benefits of IT. You know, they don't want to be slowed down. They want things to become natural. They want it to feel more like the OT tools that they're so familiar with.
They also want to track the status of outcomes, so, you know, factories right now, they're always looking for, you know, yield optimization, they're looking to, you know, optimize the production of, of their product, they're looking for overall equipment effectiveness, which is efficiencies that are happening on the factory floor, and But in the past, even though all these, these outcomes have been delivered, there really hasn't been a method to track this.
So customers are looking for, okay, fine. [00:08:00] I have all these ways to visualize what's going on on the factory floor when I collect this data. But how do I actually track that issue as it started all the way to completion and realize that I've saved time and saved money by implementing these solutions on that factory floor?
Bill Pfeifer: You said overall equipment effectiveness, OE, which I know from conversations with you is a huge metric across manufacturing and it's one that everybody's going for. Can you talk a little bit about how you calculate that, what that means? Because I know it was really new to me and it's a, it's a very comprehensive way of.
Sort of measuring success, measuring efficiency. And I bet a lot of folks listening would be interested in, in knowing that once they do.
Jason Nassar: Yeah, really it's, it's a calculation of the uptime of equipment and the quality of the product that's being produced ultimately within time. And it also can be based off of shift and you're really just trying to get to [00:09:00] that overall calculation of 100%.
I mean, if you have 100 percent overall equipment effectiveness, then you're in this ideal state. Where you're producing product at the maximum amount that you possibly can for that factory. But the actual truth of the matter is if a factory is producing at 80%, it's actually in some of the highest standards that you'll probably see out there.
But there's always that goal to try to get to that 100%, to have that 100 percent downtime, to be able to be producing a product as efficiently and as fast as possible.
Bill Pfeifer: Right. So it's a measure, it's a kind of a mix of uptime and good products produced and a couple other things, but sort of all coming together into a fusion of sort of is the factory or is the line or is the shift operating at peak efficiency.
However we define that and then gets decremented from there.
Jason Nassar: And it's a simple calculation at the end of the day, but every single customer might add or remove certain aspects of that calculation based on their specific [00:10:00] operation, right? So a company that's producing toothpaste tubes is not the same thing as producing an automotive part like a brake.
Makes sense.
Bill Pfeifer: So, as part of The transformation, the modernization that you've been working with customers. I know AI is, is a huge part of, of everything that everybody's talking about right now. And traditionally manufacturing has been doing some really interesting adoption of that using computer vision for defect detection on parts and digital twinning to You know, mimic complex machinery and figure out what's going on before something goes wrong, things like that.
I'm sure you're seeing a lot of AI projects, simple ones, more advanced ones, you know, big ones, small ones. What's kind of the state of the of the art, the state of the possible in terms of manufacturing with traditional AI.
Jason Nassar: Yeah. So when you're [00:11:00] talking about traditional AI, that is being implemented and there's a lot of legacy software that's doing that, you know, old basic software that's not even containerized.
And what you're seeing is usually Applications like computer vision that would detect defects that are on a part that's coming through and you can make decisions, barcode readers is another use case that you see quite a bit, or safety, you could, traditional AI is being utilized to see if people are wearing the appropriate PPE and safety equipment on the factory floor or if they're in unsafe zones, and all of this is fantastic, but the problem really is, is that These end customers struggle with the AI expertise.
You know, companies have to hire data scientists to help with this or have the companies that produce the software come in and actually create the models with them. And when there's a change that happens on that factory floor, there's not always someone that is knowledgeable enough to [00:12:00] implement the changes to these models on that factory floor.
And so that's kind of a challenge from a personnel perspective with traditional AI.
Bill Pfeifer: So how much do you see that having to be customized and how much, how much can you buy reasonably stock, right? I've, I've been wondering for a while, how much customers are, are training AI from scratch versus doing just refinement, transfer learning kind of stuff.
And how much, again, the art of the possible, what's common out there in practice. I haven't worked on AI implementation projects. I don't know, but you know, how much can you buy versus how much do you have to build?
Jason Nassar: Well, when it comes to traditional AI, ultimately, if a factory is doing something for the very first time on their product for the first time, they're starting close to scratch at the end of
Bill Pfeifer: the day.
Jason Nassar: I mean, these models have some intelligence. You can point out what the defects look like, and you can start to train them, obviously, from that perspective. Uh, but it [00:13:00] requires quite a bit of work just to create the models, to tune the models, get them exactly where you need them to be so that they're improving your efficiency and your quality consistently.
Now, With that being said, once you've implemented it in one factory or in one workflow, anything that's very similar, you could port that over and then you can save that time on operation number two, three, four, and five. But that really is the challenge when it comes to improving.
Bill Pfeifer: I mentally kind of took a step back and listened to our conversation from a different angle.
And on the one hand, we were starting out talking about traditional AI and lots of the, you know, older software packages do it and things like that. Yeah. A couple years ago, AI was like this arcane science that you had to have highly educated, specialized mathematicians and, you know, specialized hardware.
And they were like, Oh, traditional AI and older stuff and whatever. But then it's still getting it adopted, getting it trained, getting, finding the people who can do it is still difficult. And so it's that [00:14:00] IT conversation again, right? Like, this is the brand new technology. Oh, but now something's newer. So this is the old technology, but it's still not in common usage yet.
And it's hard to adopt and do right. And not enough people know it, but it's already not the new thing, which is just, it's kind of dizzying the way change is running right now, especially in terms of things like manufacturing. It's not, it's not generally, you know, it's not like a hyperscale cloud data center where you just rip out thousands of racks of stuff and put in the new stuff.
And that financially makes sense somehow because there's so much money that, you know, it's, it's a much leaner operation and it's not super tech heavy. So that's, boy, catching up. They have so much catching up. They're just starting to catch up. And now there's Gen AI. Whoa, a whole new conversation. So it feels like Gen AI is still too big, too heavy to really run.
In a manufacturing environment, it's still like a cloud [00:15:00] data center centralized kind of conversation, but it's probably going to get really small, very, very fast and very specialized and, you know, get, get really atomic so that we can run it anywhere, everywhere. I haven't, I've been struggling to come up with why would we need Gen AI at the edge?
What would we do with Gen AI in a factory, in a retail environment, on your phone, you know, other than. Voice to text kind of stuff, getting better at that natural language kind of queries, but that's not, you know, I'm looking for defects on my part. I don't want it to make stuff up. I don't care what kind of wine I should have with dinner, I want to know if my part is good.
Do you see Gen AI coming into factories? And what do you see it doing?
Jason Nassar: I absolutely see it coming to factories. I mean, Gen AI is moving at breakneck speeds. We're all seeing this right now, you know, with the introduction of open AI, BARD. The fact of the matter is, is [00:16:00] that we've quickly gone from something that was just an idea, and it seemed almost as if it wasn't achievable to models that everybody can use every day as if they're an expert of something that they've never touched before.
And so you're right. At the end of the day, these LLMs are large learning models. You know, they're. Best put right now in the, in the cloud, large data centers actually processing this, but this is quickly going to move directly to the edge. And that means that this technology is going to be utilized on the factory floor.
And the use cases for me are extremely obvious, right? So right now, when you look at a factory floor, one of the biggest challenges that you have is you have all this equipment that. is from, from these different OT automation companies, the Siemens, the Rockwells, the Emersons of the world that are dispersed all throughout the factory floor.
And they're talking over different protocols. So you can talk over Modbus, OPC UA, MQTT, all OT [00:17:00] based, and they have different allocations to each of these messages and codes. It takes. A engineer that understands that equipment to decode all of that, to create these databases, to relate that data so that you can actually come up with a calculation that does predictive maintenance or overall equipment effectiveness.
What we're going to see in the not so far future is customers uploading their schematics to a Gen AI model and taking pictures of what's going on on the factory floor. And with that information alone and the data that's coming through, dashboards for optimization are going to automatically be created.
Reports are going to be generated. Optimization plans are going to be created. And we're not very far from that right now.
Bill Pfeifer: So one of the big things that always comes up when we, when we talk about LLMs is hallucinations. Which I think is a little bit funny, right? Oh no, the AI made up something. I make stuff up all the time.
Oh no, the AI was wrong. I'm wrong all the time. Don't tell my wife that. [00:18:00] And you know, that's, that seems like we, we wanted AI to behave like a person and then it did. And people were upset about that. They were like, no, not that, not that type of behavior. That's like a person. But as you start to get into dynamically building your own dashboards and decoding these protocols, that's the sort of thing that people used to sit down and do over a good bit of time and really parse through what are the nuances of it and what are the downstream ramifications.
And when you automate that and it just sort of shows up and you go, wow, that's beautiful. But is it right? How do you handle errors within an automated system like that when you can't really afford to be wrong or very wrong? How do you put safeguards in place?
Jason Nassar: Yeah, so, I mean, you're asking all the right questions, and I don't know that all of the answers are completely figured out.
Hallucinations are [00:19:00] introducing a defect into this entire mix that is definitely going to make, especially the OT persona, extremely concerned. First thing Is if it's something that's safety related, you absolutely need to have a really good set of second eyes on what's going on at the end of the day, there, there can be no tolerance for that.
But one thing that I would like to bring up is that in general, human error is out there. So the fact of the matter is, is that even when the work is done manually, you know, people, we introduce defects ourselves into what's going on because we're human beings. And. You know, when you're working on a project for a week, two weeks, even with peer reviews, there are still are defects that are introduced into things.
And I look at hallucinations almost the exact same way. The difference is, is that you're going to come up with an outcome way quicker. Something that might take you, I would say, a couple of days to actually work out. You'll have the answer. Within maybe a couple of hours, and then ultimately you can [00:20:00] spend the rest of that time looking through it to find out if there are any hallucinations or any false aspects of this optimization that's being implemented.
And that's really going to save people time over and over and over again. And ultimately, hopefully, this whole issue with hallucination is solved fairly quickly.
Bill Pfeifer: So that's an interesting perspective, I guess. That makes a lot of sense, right? Instead of spending so much time developing, you spend a lot more time testing and validating and sanity checking.
Jason Nassar: That's correct. I mean, as long as there is a closed loop feedback with the people that have the expertise and know what they're looking for, people are going to save time ultimately at the end of the day and hopefully come up with a better outcome quicker.
Bill Pfeifer: And then those people have to understand not how the AI model was built, but how it operates, how it's thinking, what sorts of gaps might be in there.
I mean, that, that was kind of leading toward the next question, which is, as you start to automatically build this stuff, how do you maintain the level of [00:21:00] expertise and understanding of what's going on behind the scenes for the next generation? Right. Like I would typically expect that the person who would be doing that testing and sanity checking and building those dashboards to begin with would be the guy who's been running the factory, you know, running that machine for 20 years.
And he understands exactly like, Oh, that nut does this thing, the knob over there. Don't, don't twist that too far, whatever. And, you know, then the next generation is going to come in and things just happen. And how do they really get the understanding of the nuts and bolts of what's going on?
Jason Nassar: Yeah, I mean, you're bringing up a very difficult challenge, right?
As generations move on, is this expertise completely lost? And I would say that that would be a big risk of generative AI. But ultimately, The truth of the matter is, is you need that expertise still there. You need that expertise to create the prompts that are actually going to create the outcomes in the first [00:22:00] place.
And then you need the expertise there to also review what the output is from these generative AI models, to make sure that it makes sense, to make sure there are no hallucinations that are actually messing up the data, ultimately at the end of the day, and to do a thorough review, I see a shift. Moving away from the operator that's actually implementing all the software and code more towards the generative AI model that's actually going to be implementing it.
And then there'll be reviewers that review what's going on based on their subject matter expertise.
Bill Pfeifer: And I guess it's a different skillset that you need anyway, because the equipment's going to be so much more automated that you don't need to know so much about exactly what it's doing and how it's doing it inside the machine.
Cause it either works or it doesn't. And if it doesn't work, then the people who built it need to look at it as opposed to the more manual processes that we've hit in the past. And you said something about prompt review, which is [00:23:00] another thing that's been making me chuckle, right? Every time there's a major conversation about Gen AI, someone says, and prompt engineering will be the next big job.
I still remember when I put typing on my resume and how many words per minute I could type. And at this point we don't put that on because. Everyone types. That's how you enter stuff into the computer. It's just assumed, you know, like nobody helps me buy my pants. They just assume I know how to do that.
Nobody asks me if I can type. They just assume I know how to do that. I think prompt engineering is very quickly going to get to not, do you know how to do the prompt engineering to ask the AI, but do you know how to ask the right questions? And that's going to be really domain specific and job specific and, you know, company specific.
What questions do you need to ask? That's going to be a tough skill to build,
Jason Nassar: right? So you need to be able to ask the right questions. And ultimately you need to be able to develop the correct persona at the end of the day, because a lot of [00:24:00] these models, when you're prompting them, you. You want to tell them the persona of who you are or the persona that the AI model needs to take on.
What is your job function? And then you ask the question so that it understands the way that it needs to answer it. Because someone that's a poet is not going to give you the same answer as a doctor. And. That's why with generative AI, it's so important to understand your prompts and how to write them up correctly.
Bill Pfeifer: Yeah, that feels closer toward the prompt engineering side of, you know, how do you ask the question for that generative AI? I'd be surprised if we end up like forum manufacturing. Facility using an LLM, a large language model, as opposed to a small language model, something that's focused on a particular task.
And then you'll have multiple, multiple of these running inside the factory. And so how you ask it is probably Maybe, maybe, I don't know, less important than are [00:25:00] you asking the right question and can you see the right question? And that's,
Jason Nassar: yeah, if you have a dedicated model to that specific persona, you're spot on.
I think at that point, that's where your questions can be a lot more specific and you don't need to get directly into personas. And that would be more feasible for being located directly at the edge, close to where the operations are. I agree with that.
Bill Pfeifer: And there's always this conversation about creation versus destruction of jobs, of roles, of skill sets.
And, you know, the VCR was going to destroy the film industry. It didn't. And streaming music was going to destroy the music industry. It didn't. And Gen AI is going to take all of our jobs. I don't see that happening anymore than it did in the past. There are going to be more jobs created. They're just going to be different.
And so, It, I mean, that's, that's where I was going with ask the right questions, right? The Gen AI is going to give the answer to the question that's asked. Did you ask the right question? Did you ask the question that's going to lead to the best [00:26:00] outcome for the business based on the priorities with, you know, all the factors in mind.
And I'm still trying to get my head around, like, I don't, I don't know that I would. Be the right person to come up with those questions for any sort of thing. Like where, what, what skills would I need to develop for that to be ready to start in the workforce in 10 years? I don't have any idea. And that's that.
Wow. I mean, you know, we're going to need. Fewer people doing the, the manual processes move this thing over to there and more people like have the AI schedule these things appropriately. But how do I prompt it to do that so that we're not missing anything? And I don't know, that's, I don't
Jason Nassar: know. There's definitely going to be a science behind it and a profession behind it.
There would have to be when you think about it ultimately at the end of the day. How can you get to the right answers? You know, you can't just wing it. There's going [00:27:00] to definitely be some training involved. And I think every major corporation needs to get on board with it and figure out what that actually is going to look like in the future.
I mean, generative AI is already programming robots. I mean, I don't know if you've seen the latest video of the open AI robot, figure one. It's another company that partnered with OpenAI. It's amazing to watch this robot in only six months. It's gotten to the point where it can speak and move at the exact same time.
Wow. The observer was asking this robot if he could give him something to eat and in front of that robot, there were a bunch of plates and cups and he hands him the apple and then he says, can you please? Put away this trash and tell me at the exact same time why you gave him the apple and the robots moving around as if it was a human being throwing away the trash saying, I gave you the apple because it was the only thing that was edible in front of me.
This was not programmed. This is a result of [00:28:00] generative AI and this robot can see what's going on around it and react to people speaking to it all at the same time.
Bill Pfeifer: That's pretty incredible. And I mean, it just, it goes back to things are changing super fast. And so we have to make sure we're getting ready for it for ourselves, but also we're getting the next generation ready for it.
I have kids, you have kids, you know, how do we keep our kids on the right path so that when they enter the workforce, they're in a good position as opposed to. Like they're already redundant by the time they graduate college. Oops, I missed. Yeah. I mean, just keeping ready is going to be. A fascinating challenge.
Do you have any thoughts on that? How are, I mean, if you don't mind my taking an abrupt left turn here in the conversation, you know, you've got kids. How, what are you doing to help make sure they're ready for all these [00:29:00] changes that are coming?
Jason Nassar: So my, my children are, are young. Right. My oldest is seven and we already started them, you know, trying to learn like software and logic and things like that.
We've taken them to these bootcamps and these courses, but the interesting thing about it is if you listen to Jensen from NVIDIA, you know, he says that, you know, in the past he would have told people, you know, To go to school to become software engineers. And that's not what he's saying anymore, because AI is going to be creating the software that we see in the future.
And I'm paraphrasing, right? That's not his exact quotes, but it's something along the lines of that. And so, Bill, I don't have the answer for that either, to be honest with you. That's how fast this stuff is moving. When you hear, you know, the CEO of Nvidia say something like that, that's mind blowing, you know, how fast we are going to move and how quickly it's going to happen, because.
Everybody is going to be a programmer. Anybody can create a software application because it's all going to be about creating those prompts at the end of the day.
Bill Pfeifer: [00:30:00] Yeah. Kind of everybody and nobody is going to be a programmer.
Jason Nassar: If anything, I'm going to make sure my kids understand AI. They're using AI daily and they're getting more and more proficient at it, so it comes second to nature.
Bill Pfeifer: I was hearing not too long ago about, I think it was a college that the professors were freaking out because people were using chat GPT to write their papers and they're not writing their own and they're cheating. And, you know, how do we stop this? And I'm thinking. But when they hit the workforce, they're not going to start with a blank sheet of paper.
So, you know, they have to, the schools already have to get ahead of this, because if you're trying to get people ready for the workforce, it's not be afraid of chat GPT because you can't touch it. It's how do you work it in so that you've still validated everything and you're citing the right sources and what you're saying is true and you've drawn conclusions, you know, like what.
That piece of you fits into [00:31:00] that, and it's already, it's already landing.
Jason Nassar: Well, here's the thing too, like, I still think there are many people, because it's still early, that don't really recognize how powerful it is right now. And how much more powerful it's going to be. I mean, one example, you know, my wife, she works for NASA.
She's a flight controller for the International Space Station. She trains astronauts. She has a totally cool job. I didn't know that. And, and she, and she's, she's never really dug into AI that much. And I, I have a subscription to chat GPT. And I said, you don't realize how amazing this is. I said, ask it something about the space station that only you would know, or somebody that's worked for NASA for a very long time.
Yeah. She puts in the prompts and asks the questions, not even seconds later, it spits everything out in great detail and she's reading it and she's like, Oh my God, it is like, it gave a better answer than she could have given. So you worry about hallucinations. They're in there, but the amount of truth that comes out of this and how [00:32:00] intelligent these models are right now.
is mind blowing. I mean, these models are passing the bar. They're passing oral exams from a medical perspective with flying colors and the 90th percentile plus. And that's where we're at right now. You know, six months from now, we're going to be way beyond that.
Bill Pfeifer: Boy, that just, I remember. Seeing something about the first Moonlander and the amount of processing power that was in that.
And at the time I was looking at it and thinking about it and realizing my Apple watch has a bigger processor than the, than the Moonlander, than the first Moonlander did. And now we're talking about chat GPT. Whoa. Okay. So what is that going to do next? It's Man, well, and then we start talking about just in time manufacturing and personalized retail, you know, walk into a retail facility and they'll customize a thing because you want it.
And so that has to [00:33:00] be integrated with the manufacturing process, which then has to be flexible enough to change on the fly. And that's just tons of processing and trying to figure out how to, how to get all that stuff interlinked, but maybe we don't, Have to. I don't know. That's, that's going to be interesting.
Jason Nassar: Well, one of the use cases from retail that fascinates me is, you know, if you're a customer and you're walking around a department store and you pick up a shampoo bottle or anything else, for example, and you put it back on the shelf, you know, AI can recognize your face. And then when you're going through the checkout counter, it'll give you a coupon for that exact same item.
Then you brought up, but that's, that's traditionally I generative AI, not only is it going to do something like that, it's also going to make decisions on what else could I do? I being the AI, what else can the AI do to get this customer to come back even quicker? And then you get to, you know, the robotics discussion that we were just having, you know.
Tesla, of course, they have their optimist robot, generative AI. They're definitely planning on putting those robots all throughout [00:34:00] their factory floors as, as time moves on, ultimately trying to speed up the processes because, you know, robots don't get tired, you know, that's going to happen, I believe all throughout manufacturing.
I think as the technology improves itself and it's going to happen faster than we think, there will likely be, you know, robots and even humanoid robots that are going to be performing tasks that in the past people performed.
Bill Pfeifer: So the piece that I took away from that is maybe not the piece that you were thinking that I was going to take away from that.
That's, that's, but I was working on streaming analytics for a while and then thinking more about AI and starting to, starting to build assets around that and such, but that's always based on data that's been created, which means it's always a view of the past. And so traditional AI. That doesn't easily extrapolate forward.
And I wonder if that's what we've just gotten to with Gen AI, right? It's more, you know, human conversational kinds of things, and you [00:35:00] can dynamically build dashboards and things like that. So it's more creative, more generative, cool. But I wonder if what that's going to do is give us a better view into the future.
The example that you gave, I picked up the shampoo and I put it down because I have no hair for those of you who can see me on video, but you know, then I check out and it spits out a coupon. That's cool, but I didn't buy the shampoo, right? It's a view of the past. And so it didn't affect my checkout bill today.
And so what could happen in the store to get me to buy something else? Maybe some head polish. I don't know. Um, yeah, but that might be more of a generative AI thing as opposed to like, there's no data for that. Okay. So what would he buy? Right now in this moment. Exactly.
Jason Nassar: No, you, I mean, and it's going to be like this across all verticals, all industries.
I mean, look at the video processing capabilities that are coming from Gen AI right now. Once again, open AI, they [00:36:00] have a new demonstration out for their video processing called Sora, leaps and bounds above what they were showing a year ago. The quality of the physics is on point, very few defects, off of two sentence prompts.
You have puppies that are playing in the snow with each other, and the snow is falling off of them, and the hair is flowing in the wind, and the dynamics are just incredible. And this is shown video after video, they have an ocean and the water's moving and crashing into a building. This is all auto generated, and it took companies like Pixar.
Over a decade to figure out how to get the hair to move in Monsters, Inc. You know, I watched a documentary on this and these generative AI models right off the bat already understand physics and they don't need any human intervention. That is just mind blowing. So what is that going to do for the future of development of everything?
It's going to accelerate it. I mean, we're going to learn things that [00:37:00] we never knew and it's all going to be because of generative AI.
Bill Pfeifer: But that's at least in the near term, the foreseeable future, that's going to be focused on the mechanics of how you do stuff. So I guess the, the question of how we get our kids ready for, for what's coming and how we get ourselves ready for what's coming is more the creative side of it, the questions, what should we do as opposed to what can we do?
Jason Nassar: Right.
Bill Pfeifer: And, and why?
Jason Nassar: So, even, even in my own house, for example, my pool pump went out. And so I have ChatGPT on my phone, and I go out there and I take a picture of my exact pump. It has some barcode numbers and things on that. And I let ChatGPT know what was going on. And it very specifically explained to me exactly how to troubleshoot that pump from a picture.
Okay. Wow. That's cool. And, and, you know, I love it. That type of intelligence in a [00:38:00] smaller model, of course, right there on the factory floor, providing that to a technician, it's just going to save a ton of time and a ton of money for a corporation. The technology is already there. It just needs to be implemented correctly.
Bill Pfeifer: That's pretty incredible. Okay. Yeah. I was, I was getting a lot into the whole, into the why and what next and stuff like that. I guess, I guess I was just having a philosophical day. So thanks for playing along. Yeah. So Jason, this was a super fun conversation. How can people find you online, keep up with your latest work and tap into your latest thinking?
Jason Nassar: You LinkedIn. Jason Nassar is my name or at Jason. Nassar. That's my ex account.
Bill Pfeifer: Fantastic. All right. Thank you so much for the time and the perspective. This was a great conversation. I appreciate it. Thank you very much. Appreciate it.
Narrator: That does it for this episode of Over the Edge. If you're enjoying the show, please leave a rating and a review and tell a friend.
Over the Edge is made possible through the generous [00:39:00] sponsorship of our partners at Dell Technologies. Simplify your edge so you can generate more value. Learn more by visiting dell. com slash edge.