He Started an AI Company in 2011. I asked him my burning questions on the next decade.

追加: 2026-05-09

[00:00:00]It's 2011 and you decided to start an AI company.

[00:00:04]Well over a decade and a half ahead of all the hype that happens today. So, where are you today? What are you doing?

[00:00:10]Well, I get a chance to interview Matt.

[00:00:12]He is the chief AI officer of NLP Logics, which is a machine learning and AI company that was founded all the way back in 2011.

[00:00:21]And over the next 10 to 15 minutes here, you get to hear some of the questions that I've asked as well as my own interpretations of the AI age. Enjoy.

[00:00:29]Well, thank you so much, Matt, for joining today. Um I'm just going to dive straight into the questions. Now, for question one, for anyone watching who doesn't know you, you didn't come up in the chat GPT wave. Walk me back to when you first decided machine learning was the thing you wanted to build a company around. What were you seeing that nobody else was and when was this?

[00:00:50]Yeah, this was going back um probably 2008, nine, and 10. I would say, you know, for for my little slice of the world, it was um like my background was software engineering in general.

[00:01:04]And um you know, we'd spent a lot of time building applications that were really focused on managing data, you know, um letting users enter data, edit data, maintain data. And then obviously, you know, that was about the time when big data was the was the theme because everybody's collecting data and storage and and latency and and access and analytics were just becoming kind of the new problems. Um and so I think to me what was interesting at that point in time was like was pretty much this big question, which was we have all this data and kind of now what, you know, um and that's when I we kind of got, you know, I personally kind of got, I don't know, a little bit obsessed honestly with machine learning because it was at that time, um you know, it felt kind of magical, I think, because it was it was, you know, showing the ability, right, to predict the future. I don't know, you know, that's kind of how I think about it sometimes. It's just like, especially at that time, it was like, man, you know, we're we're getting a glimpse of the future of using data to predict the future. And if you can predict the future from a business standpoint, you know, obviously, you can make a lot of better decisions, and you can optimize things differently.

[00:02:19]So, that that to me takes us to 2011, and that's when we started the company, but it's also at that time, you know, I look back, and I'm just amazed because, you know, there's a company called um Kaggle that that focused on like machine learning competitions and hosting them.

[00:02:37]And it I'm just so fascinated still that that was actually a thing because it was basically, you know, companies like Facebook or Meta and Microsoft and GE and and just these big companies that today have research labs focused on this stuff, but in 2011, they were basically like, yeah, you know, we have data, but we don't really understand how to solve these problems.

[00:03:01]And so, you know, Ted and I saw the opportunity of saying, yeah, it's not just building models and solving these problems using machine learning, but but, you know, how do you deliver that value back to the business? And of course, that means scaling it and supporting it, maintaining it, integrating it, and driving outcomes.

[00:03:18]And so, that to me is how I think of the genesis story is just, you know, um see seeing the data and seeing that there's so much value to be unlocked and and delivered back to the business and just getting real excited about helping our customers do that.

[00:03:35]No, I love the narrative there, right?

[00:03:37]This concept, which is it started with curiosity, and it came with more questions, not less, right? In a world full of answers, questions become valuable. Uh and I think you did an amazing job describing the what and why, but I guess the other question could be like imagine you get invited to a cookout or small party.

[00:03:56]When someone asks you what NLP Logics does, how do you actually answer? Has that answer changed in the last few years because now GenAI is is a thing or is it pretty much pretty much always been the same thing? I to me I think it's it's it's probably changed a little bit in how we say it but I think in spirit it's it's it's generally the same but I think we've got tighter as far as how we see our our place in the world.

[00:04:20]And that's really to help our customers solve their problems. You know, so um they in a course today that it's almost like implied that's going to mean technology and AI.

[00:04:31]Um but I think the reason why I like saying it that way we're helping our customers, you know, solve their problems is because um there's there's AI is a huge ecosystem and there's more to AI than GenAI and and I don't want to I want to make sure that it's clear that you know, whatever the whatever the solution looks like for our customer, we're going to be really pragmatic with helping them select the right tools for the job and um and that's how I see us, you know, and and we always say you know, around here we always say, you know, you you help your customer solve their problems and they're going to solve your problems. You know, if you focus obsessively on the customer and getting them the outcomes they need then as a business for our business, that's the best thing we can do day in and day out.

[00:05:22]Yeah, it's almost ironic the age-old idea you you the customer is right in terms of taste. If you can solve their problems, you'll be successful as a company no matter how many AI solutions come about, that's still the core thing and and I think people are focusing on creating tools rather than solving customers' problems. But there's also something important there which is there's surprises that come. You find new ways to solve these solutions.

[00:05:47]What's still surprises you about this work even after a decade in it?

[00:05:53]Oh, I mean the things that surprised me are um I would say one is the the I mean this is I mean you know, this is just from what I see and how I interpret the world right now, but the the amazing amount of opportunity that is out there. Uh you know, I think that some people some people might think they might feel that the you know, the AI revolution maybe has it's moving so fast. It's it's kind of passing their company and they're not keeping up and and to me it's just we're everything is new and it's still early. And even you know, I look at the adoption and um just the variability around who's adopting what to to help their businesses and and I would say you know, as a theme I would say you could if you if you froze the advancements in AI today, if you just didn't make any more advancements, I think that just with today's technology, there's probably a good solid 10 years of build out. Like like there is so much we can do with today's technology that just we're not we're not quite doing yet across any industry across everywhere.

[00:07:10]And to me that's just really exciting.

[00:07:12]It's amazing to think that the capab where the capabilities are at and where the adoption is at. There's just a huge gap and that's what's kind of exciting cuz that's where all the opportunity in in our world is is how do we fill that gap?

[00:07:28]Yes, these huge gaps and people seem to be thinking there's gaps where there isn't, right? Like oh, we got to use AI to solve this problem, but we could have just used a traditional machine learning algorithm, not a generative AI or even just some scripts in old Python, right?

[00:07:42]And we see people hyping these things up and this isn't the first AI hype cycle.

[00:07:46]We've You've probably lived through almost three AI cycles of hype, I would argue, in in the last decade and a half.

[00:07:53]What do you see that keeps repeating that everyone thinks is new? The most interesting pattern that I think repeats over time is that over a longer horizon general approaches outpace the specialized approaches. And I'll give you a a couple of an examples is you know, in the in the if you go kind of before language models, you know, there was a solid, you know, I don't know, decade probably where computer vision was the most advanced, most hottest area of machine learning.

[00:08:31]And the the you know, there was kind of these two camps of computer vision. It's like you got these classical camps that that are looking at um uh you know, kind of kind of heuristics around image and building kind of kind of uh systems around the heuristics around what an image looks like and those are kind of specialized um threshold-based indicators and morphology and and and these different, I would say well-studied but almost heuristical approaches that are kind of um more specialized, you know, you want to you want to um detect uh shapes in an image. Well, you can use this kind of a filter and some morphology and then use a threshold and and um all that's great and then as the general approach improved, which is here's an image, here's some labels, let's learn general um patterns from it, over the five or 10-year horizon, like the general approach beat the specific. And then now I think we're seeing that same thing with, you know, natural language, which is when you when you're able to continuously add hardware and computation and data that the general approaches end up surpassing some of the specialized approaches on like these larger scales, larger timelines. And, you know, I don't know if there's necessary It's interesting cuz like the lesson there to me is not to ignore specialized approaches, but almost to recognize that in some domains like specialized approaches have basically a shelf life, especially in this world of, you know, where you have data and computation. Um so, that's that's kind of a thing. It's like, um you know, we had a system here, I'll be honest, we had a system that we built 10 years ago that was very specialized. It was very good, but it was very specialized. It was some higher-end computer vision models, deep deep deep learning networks that were kind of purpose-built for this one domain.

[00:10:35]And here we are 10 years later, and um to me what's amazing to think about is these these these off-the-shelf pre-trained models that have nothing They have no insight into the domain of how they're going to be used are now able to surpass um by not a small margin, by a very large margin, what, you know, state-of-the-art would have been, say, 10 years ago building something really deep and purpose-built. So, it's just amazing, and I know that's probably kind of pretty complex, but it's in in our field it's just it's just amazing cuz we're building these systems that are enterprise scale delivering tremendous value, and they're not rooted in let's sit down and label some data and build a really bespoke model focused at this one really specific thing. We still do that, and there's lots of cases where you need to do that still, but but in a lot of cases you can start with something off the shelf and and get much further down field much quicker, which is exciting.

[00:11:37]Absolutely. And that's what really excited about excited me to talk to you all about this is you recognize there is extreme value in these bespoke solutions, but you also need to grow with the times and look at these more kind of amplified ways of general models. You don't have to worry about over-fitting because the fitting happens as you use it.

[00:11:57]Unfortunately, this is where the recording ends. Matt and I talked for another 20 minutes after this. However, for some reason our lovely recording software decided to stop and cut out. It shows it there, but none of it works.

[00:12:10]So, if you are interested in conversations like these and hearing more questions, please comment below and let me know what type of questions you'd like to have me interview with this company. Again, a company that's been around for quite literally over a decade. So, thank you so much for all of you who are listening and until next time, I hope to have you a full recording of these wonderful insights.

[00:12:31]Until next time, happy learning.