Introduction

Okay, so what we're going to talk about today is basically like AI versus the industry. So is the goal to solve industry problems with AI or with the AI solutions we have today, try to find problems to solve?

So I was in San Francisco like the last two months and the latter was kind of the pattern I was seeing. Like lots of very technical people were trying to like wedge their way into industries.

I think both... I guess both methods work, but today I'm gonna touch a little bit more about like having industry experience and then using kind of common sense to look at new technology and think about like, does this actually make sense? Is it objectively better or is it just kind of a fad?

1And I really think that people in the industry who like deal with the problems every single day are like the ones that actually understand it the most. So it's like this confidence that like tech is very accessible and like, as long as you can figure out kind of like, how to incorporate it into your business or into like your industry, like there's a lot that can be done.

Personal Experience and Background

So a bit about me. So right now I just started a company about three months ago. So in the early stages of being a founder, if anybody's in the same boat, would love to connect.

Previously, I was VP of growth and product. And before that, I was in sort of the medical health care business. And we did zero to 100 in two years.

Diverse Industry Exposure

So I have a lot of industry experience, I'd say, just like dabbling across a broad industry. So it includes like fintech, public equities. My background is actually in business, government contracts, medical supplies, health care, logistics, supply chain. And I've kind of seen the pattern that like there's a lot of transferable I guess use cases for technology in all these industries but people in the industries don't actually know that because they're too busy like working in the industry rather than just like following what the newest technology is so my I guess interest in AI has been

Initial Interest in AI

Since 2018, I've been in women in AI. At first I was interested in like GAN art.

So instead of AI art today, there was this phase where the process of creating art through like models was actually art itself. Like it was really abstract and really interesting. It was like the process was more unique than actually the output, which is what AI art actually is today.

Focus on Industry Applications of AI

So I'm interested also in industry applications of LLM, which is what I'm kind of talking about today. And then unsupervised learning for R&D is really cool.

So kind of my background and my relation with technology is I don't have a technical background, but both my parents were computer engineers, software engineers for their entire career. So I grew up not being afraid of technology. It was just always accessible for me since I was a kid. And technology was more of a tool.

How do you use it to practically support and improve an industry? How do you use it to... um like innovate on existing like businesses um so something that happened a lot when i was a kid would be people would ask like what your parents do and it would constantly change so i would like literally never know i had like no answer for that and i think that says a lot about just being in tech like things are always changing there's always new stuff going and like a true technologist is just somebody that like follows like what's going on and like uses common sense to like apply it to like problems

So I think that's kind of the same thing that's going on with AI and LLM right now.

The Evolution of AI Hype

AI has been really hyped up for the past decade through AlphaGo and that Netflix documentary came out. Everyone was really into reinforcement learning. That was the next big thing, right?

And then after that was machine learning. Everybody was like, data is the most important thing. having in-house models is super important because you had to train your own data and that's your competitive advantage. Otherwise, like you are going to be left behind.

So, so many companies invested like tons and tons of money into like an in-house ML team, like experts, their own data. And the thing is, technology is not always better. so it's not always better or it's not always that much better it's not always cheaper or relevant or applicable to like an industry that you're trying to disrupt so

I think machine learning is a good example because so many people in 2018, 2019 tried to incorporate machine learning into a lot of their systems, like underwriting or risk tolerance, like assessments, like data analysis. And to be honest, if you looked at it with the common sense, some of these models were not actually better than you know, just like regular multivariable regression models or like literally an Excel spreadsheet with somebody like dragging the formula over, right? So I think it's just being honest with yourself, like, is this technology actually better or did I just spend like $10 million because somebody told me to do so?

And I've just seen this many times in different industries, like, I've seen this in a supply chain where people use it to forecast, use it to forecast demand. And the thing is, if you don't have enough data or you don't have enough orders, it actually doesn't really make a difference, not that much better, versus you just understanding your SKUs really well and forecasting it based on seasons or something. You don't need that many complex variables for certain tasks.

Current State of AI

where are we today with AI?

And I think LLMs, I think why everyone's really excited is that we've seen chat GPT come out and like LLMs are objectively better for certain things. Like objectively, like really, really good at certain like tasks, text related like document related analysis.

So I think knowing that then you can think about okay like what can I use LLMs for or what shouldn't I use LLMs for.

Okay one second.

Utility of Large Language Models (LLMs)

1So this is kind of like a high level just like I took a couple examples of like what LLM products that are objectively useful.

And to backtrack a little bit, I think the reason LLMs are objectively useful right now is like 90% of the work is done. Like OpenAI, you know, X, Google, they've all like invested hundreds of millions of dollars and like computing power, hardware, like really, really smart people, like all the machine learning experts, they've all gathered to like do 90% of their work. And basically they train language models on the entire internet plus some and their models are next level like physically.

So it's like it's really, really good. Like compare this to like I don't know, a medium-sized business that wanted to invest in machine learning, and they had an in-house team. Also, I hope I'm not offending anyone, but had an in-house team and not enough data, and then they wanted to forecast some variable, and it didn't really work, and then they're kind of fooling themselves, right?

So LLMs are actually really good. And I think that's why we've reached mass adoption. A normal person can go on ChatGPT and be like, oh crap, that's really, really good.

It's actually... Remember when Siri came out? It was good, but it was not mass adoption. It's not that good. So I think that's just a signal that when it comes to text, it's really impressive.

I think if you use the same lens for, I think, image generation... I think many people can say it's not that good yet.

I probably wouldn't use DALL-E for every single visual thing I would do because it's just not there.

Practical Applications of LLMs

So with LLMs, I think a couple of good use cases is not just chatbots. Those are really popular and they are good because if you upload information into like I don't know, a UI or, like, ChatGPT, and you ask it questions, you can interact with it very quickly in real time, and that's good.

But other, like, less known applications, I think, are, like, smart search. So traditionally, keyword search was, like, I think keyword search is still the most popular method of search, but right now with LLM, what it can do is it can infer what you're trying to search.

So I'll show you a couple examples later, but it's like, if you search up warehousing, for example, normally when you get your search results, it'll find the word warehouse and all the other examples. But what you really mean is like manufacturing and supply chain and potentially warehouse management software. Like there's a whole breadth of, I guess like adjacent words that you actually,

are looking for but like normal like non-natural language search just won't help you find another one that's seen more common maybe is like document extraction and summary so you upload a document and then it can summarize things for you or you can tell it to summarize specific requirements or like specific things you're looking for and it'll find it content generation obviously so That's what ChatGPT is. You like type something, it'll generate content.

And then like style formatting. So kind of like the Zapier demo, it's like, I want the output to be in an email format with very formal language, like to this one person, like who has this personality and it will format your output to like that. So these are things that like LLMs are really good at.

It's like anything related to like words or documents or like lots of words and lots of documents. It's been able to really like replicate the equivalent of like a really smart like analyst. You know, you ask somebody and they have like more thought than just a very black and white view on language. Like they know what you're trying to get at.

So I think this goes back to the frame of mind of common sense with tech. Does it make sense? And I think with LLMs, it does, which is why there's so much hype around it.

The Final Push in AI Product Development

But I think... the part that people in the industry are missing out on is like 90% of the work is like done. So the other 10%, oh yeah. So the last 10% is like putting it together for like a use case, right? So how do you add these components of like product or potentially like agents or things to like create an actual product or a potential use case for your industry?

Case Study: AI in RFPs

For example, when I was in San Francisco, lots of people were adding these blocks and trying to build an AI-powered RFP business. And that's fair because that is actually a really good application of LLMs because government contracts are super tedious to read through. Well, first of all, they're hard to find because you have to find the exact word in order to find it. Then you have to parse through hundreds of pages in order to find exactly if you're applicable if like you match their criteria and if you can apply and then you have to respond to like pages and pages of like to submit.

And then language translation localization is also something LLMs are really good at because it can translate in real time. So if, I'll show you an example, but if the government RFP is in French, right now, without any of this, you have to go to Google Translate or you have to find somebody who understands the language in order to parse through things. AI can actually do it all automatically.

So yeah, you have people who are like, on the technical side, trying to break into these industries. And then I think you also have people on the industry side because like I've done RFPs before and it's been really painful and I kind of know like immediately I can see like, oh, this is where Owens would actually benefit like the process. And it's like I don't know if, I'm sure like all of you guys have like really unique backgrounds, like industries and you can kind of like go back and be like, okay, like what's something I've done that like would actually be like easily like improved with like LLMs today.

Social Listening with AI

Another use case is like, like social listening. So right now, if you, If you guys have ever, for sales, try to find leads on Reddit or LinkedIn or Twitter, it's pretty painful, because you search the word and sometimes you find it, sometimes you don't. Many times you manually search and you just hope that somebody's talking about something related to what you sell or what you know. So with AI, it's cool because

if I am a warehouse provider and somebody is searching fulfillment, AI can basically make that inference for you. And obviously you stack on other things that are not cutting edge, but it's like, okay, then API feed. So the second I find a relevant post because of smart search, I can feed it into Slack or email or something like that.

Oh, so this is like, I just took a screenshot. It's kind of small, but this would be, so we actually kind of built this for somebody we know that is heavily in the RFP Canada space and struggles with consolidating like five different channels and like searching. So what's cool about this search is I'm searching medical supplies between January 15th, 2024 and February 20th, 2024.

And the first result is actually in French. So I think that's like, like I don't, I don't know if I've ever used a search that does that, but it's like, it infers what you mean and then it translates it and then it like, um, surfaces your result. So like, I think this tool is just like, Hey, you'll never miss a relevant opportunity. You need to see it first.

And then, um, sort of how you would take this business to the next step is you add the content generation part. Like, okay, now that I know, um, now that I know, um, Or now that I found the right ones, like, can you summarize the RFP, like, the document for me? And then can you respond with my, I don't know, with my details, right? Like, personalization is really big with LMSG. Like, you have your profile, like, all the things that you can, all the criteria you can fulfill. And then they'll say, like, yes, you're good for this, like, RFP. Like, lots of people are, like, kind of building on that.

um and this is like the social listening tool right so these are posts that are coming through linkedin because we're just searching like people looking for a 3pl provider and like some of these posts will never say the words like 3pl provider but it'll show you relevant things that a human would be like oh yeah this is obviously relevant so it's not perfect as in like l1s aren't perfect um and obviously like you're not going to build a perfect system but like I said, 90% of the work is done. You kind of just have to add your like industry knowledge and like your use cases and you can get like some really powerful, I wouldn't even say like workflows, it's really like powerful like businesses or like in-house tools that you can use that are like standalone products that just like are really exciting.

Conclusion

So yeah, to like, I guess wrap up, The question I guess for you guys is who would you bet on?

Who Will Lead the AI Revolution?

Is it industry experts applying LLM as they need with a lens of will this actually make sense? Is it cheap? Is the ROI there? Will it actually improve what we're doing?

Or is it AI founders that know how to apply GPT prompts and know how to code and all these things and trying to find an industry to improve? Like my opinion is that obviously both, but I think when it comes to like, cause there's lots of like very technical people that are doing work in like the models. And this is like back in the machine learning days, they would be like in-house building models and like cleaning data and extracting and building like an in-house model. Now they're all like working on like one big giant model or like multiple giant models.

But I think the disruption in the industry, a lot of the opportunity actually comes from the people who have experience in the industry and figure out creative ways to apply it that make the most sense. And then most importantly is distribution. If you're building an AI-first company and you don't have a network and you don't know who to talk to and you don't have the industry experience, it's going to be really hard going to market and figuring out how to even navigate that industry.

And I've just seen it before, comparing... a tech founded business trying to break an industry is actually really really challenging versus an industry expert understanding basically like how they can use LLNs and it's like not actually that hard and I think the goal of my talk is to like break down that barrier like right now we're trying to build a business that like can basically productize certain parts of it so like it's not that hard and I think you don't even need anybody to do that like somebody with you know technical experience that can code can go into like these like databases and like basically understand vector, I guess like, what do you call it? Like vector models and how it all works can kind of stitch together something that like would actually be really useful.

And the thing is it'd be like a lot cheaper than what it was five years ago when you had to hire somebody full-time in-house, like five of them, get all the data, wait for the data to like be enough and then train your models over and over again. It's just like the difference like a hundred times. So, yeah. Yeah, that's pretty much it.

Invitation to Collaborate

This is my LinkedIn, my email. If you're ever interested in, I guess, building industry LM solutions, like I'm happy to talk or just give advice or like share some of my experience because I've kind of seen a bunch of industries. And then if you are like a software engineer and you're interested in like I don't know, chatting, let me know too.

And I think this is just like a really exciting space. And I'm like a big believer of like domain experts and like people who have the knowledge. Not being afraid of technology goes back to like I grew up with technology, so I was never afraid of it. Like I don't have like a technical background. But to me, it's like you need both, of course. But it's like technology is always changing as long as you're like curious and willing to learn and like wanting to like innovate, there's like so many opportunities to do so. Yeah, that's it.