AI That Sounds Right Versus AI You Can Prove

Alex Romanovich 0:26

Hi, this is Alex Romanovich, and welcome to Global Edge Talk. Today is July 7th, 2026. And with us, we have Scott Cohen, the CEO and founder of Jaxon.AI. Hello, Scott.

Scott Cohen 0:38

Hey, Alex. Thanks for having me. It’s a pleasure.

Alex Romanovich 0:40

And it’s going to be an amazing and very insightful session. You’re a founder and CEO of Jaxon, a company based in Boston that builds deterministic, mathematically provable AI verification for high-stakes industries, like insurance, financial services, defense, and life sciences. Sounds very important. It sounds very appropriate for the moment. But before Jaxon, you built and sold Big Rio, which was a big data ML consultancy, and founded Dropfire to commercialize MIT licensed data technology. You also hold degrees from Union College, Watamay Alma Matas, Northeastern with an MBA, and ASU with Masters in Engineering, with coursework at MIT, and notably trained a professional sh as a professional chef at the Cambridge School of Culinary Arts. Wow.

Scott Cohen 1:29

The most fun I had in school. Yep.

Alex Romanovich 1:31

Absolutely, absolutely. And your core belief, as you told me, is that uh AI is merely probabilistic and can be trusted for the decisions where being wrong isn’t an option. And that’s why you build D sales. So welcome to our studio. Thanks for watching. Lots to discuss, lots to talk about. And please tell us a little bit more about what sort of possessed you to come up with something like this with such a very interesting and diverse background.

Scott Cohen 1:57

As you mentioned in my bio, I’ve been in the AI space for a while now. Some people call me a dinosaur because it’s been over a decade. And it’s always been about getting the AI to do the thing. You want it to accomplish a goal. How we used to train models was very rigid and formalized, and you could trust them a lot more, but they were very limited and you had to put a lot of work into training them. Now, with the likes of GPT and Claude and all the large language models, it’s not so much about training them, it’s about coercing them to do a job. So we got into a project with the Air Force, the US Air Force, where the use case requires it to just work. And the variability that you get from LLMs in general is just not sufficient for this type of use case. So we were kind of forced down this path to figure this out. This started about four years ago. And the way I like to describe this is if you are asking a question of an LLM, it’s going to look at its training, which is the world’s data, things like Wikipedia, download of the web, what have you, and it’s going to look at the highest probability answer to your question. Probability is the key word there. You mentioned in my bio, and I made a note of it, that we’re doing something that’s deterministic. I think that’s a word that maybe not a lot of people are familiar with. So let me explain. Deterministic means it either is or it isn’t. There’s no probability in the mix. So I force people to think back to geometry when they were given a problem and they had to show a formal proof to an answer, and every part of the answer had to be right in order to for the final answer to be correct. Any variance there would set it off course. That formal proof is what we’re doing with Jaxon and what deterministic really means.

Alex Romanovich 3:46

Got it, got it. Now you come from a family of inventors, actually. Lots of interesting projects was done by you and your parents. Sweetheart cup, the flexi straw, the banana split dish. Did that shape in any way how you see building things, or were you determined to make something on your own?

Scott Cohen 4:04

Yeah, I definitely think that impacted me. Growing up in an entrepreneur entrepreneurial environment kind of forced me down this path. Been doing company after company since I was in high school. The first thing I did was have a car detailing business that I would do on summers and and over Christmas break. Yeah, uh a very entrepreneurial family Sweetheart Cup did really well, and I’ve always wanted to follow in those footsteps.

Alex Romanovich 4:30

That’s a great success story. Now you also trained as a professional chef, something that’s near and dear to me. Because I attended some uh, you know, maybe I I did not complete the entire coursework, but I attended some classes at the Culinary Institute of America at Hyde Park. Now, what did Kitchen teach you about precision and pressure and this uh tremendous amount of stress that later showed up in your work at Jaxon?

Scott Cohen 4:55

That’s a good question. You know what? When you put together a dish, it is following a recipe, and there is art to it, but there’s also science. Baking is different than regular cooking, like savory dishes. You have more art. Baking is very particular, but just the whole concept of having everything prepared. Do you remember the term mise en plus? So having everything prepared ahead of time so that when you get into it, you’re not going backwards and trying to like chop up some onions to add it to the dish. You’re just ready to go. I I think that’s very similar to how you run a company where you have to have everything organized in such a way where it all comes together nicely in the end. It’s almost like um you ever watch the TV show A Team? Oh, yeah, of course. Love it when a plan comes together. Of course. So if you cook a good dish, you love it when a plan comes together. If you build a great company, you love it when a plan comes together.

Alex Romanovich 5:49

You know, I love your slogan uh uh for Jaxon, AI for AI. And that in itself is basically saying, look, we’re going to check AI with our AI. And uh you’re betting everything on provable AI while the rest of the industry is still chasing the next uh you know flashy object. Is that just who you are, Scott? Do you enjoy being somebody in the room that says, hey, I proved you’re wrong? I guess so.

Scott Cohen 6:15

I guess so. Uh yeah, no, there is a lot of newcomers to the AI race, and there’s a lot of lip service and great marketing, but you have to have the solid foundation in order for it to actually be a viable and valuable application when it goes to production. And that’s where a lot of people are falling short. So yeah, I guess I like being ahead of the pack and doing things right the first time.

Alex Romanovich 6:40

Absolutely. And in plain English, tell us what is the difference between the AI that sounds right and one you can actually prove is right.

Scott Cohen 6:48

So I already mentioned that the Carnel LMs are trained on the world’s data. They’re you ask it a question, it’s gonna look at patterns in its training to figure out the most likely answer to give you. What I didn’t get into is that they’re all token predictors. That’s the technical term. A token you can think of as three or four characters, and how you put together those three or four characters with other three or four characters builds up a sentence. And all of that is what the LLMs are doing when you’re waiting for it to respond, is putting together those token tokens to give you an answer that sounds right, and oftentimes it is, but there are many times where it isn’t. So for you and I, you or I using Claude or ChatGPT or whatnot for writing prose for an email, no harm, no foul if it gets it a little bit wrong. But for regulated industries, military applications, you really have to have that rigor that Jackson brings into the mix so that we can prove that something is accurate and it is consistent. Actually, that’s another thing I want to harp on is consistency. Because you can ask an LLM the same question and it will give you a different answer. That’s what probability is all about. It’s gonna give you a different answer based off of its weights. The weights, I didn’t mean to bring up such technical terms. The weights are how the model pieces together its training.

Alex Romanovich 8:12

Absolutely. Not only that, you can almost uh tell Yah to go back and reconfigure the answer based on certain likes or dislikes, and uh you can you know impact that as well. Absolutely. Now, all of this hallucinations, all of this uncertainty is opening, especially in America, is opening us to a number of different lawsuits. You know, we’re known for having lawyers, you know, what, ten lawyers per square per square foot in America? So what’s interesting about this is that, you know, the lawsuits are piling up like crazy. Why do you think most companies are still flying blind on this risk?

Scott Cohen 8:48

Because they don’t understand it. I think that’s a quick, simple answer. It’s very complicated. The bit about token predictions, I’m living and breathing it every day. That’s not something that everyone really understands. And like I mentioned earlier, they’re getting marketed to by very believable marketing, and they buy into it. They’re believing it. So what I’m finding, and that’s actually my beautiful drawing on the whiteboard here. Everyone loves my drawing. Is where a lot of people are. I call them chimps, where they have an LLM for all. They’ve paid the hundred grand plus for an enterprise edition of one of the LLMs, and they have it secure, but it’s not tied into actual workflows. It’s not doesn’t have the subject matter wisdom that the employees that have been doing it every day have. Learning from the web and and again, Wikipedia and whatnot. It doesn’t really understand the domain that well.

Alex Romanovich 9:43

Absolutely, yes. And uh, you know, speaking of AI mistakes or AI related mistakes, you know, the the natural, I guess the natural sentiment that was pro you know proposed by the industry and uh some of the tools is, you know, have another AI check uh this AI, right? Now you said that that may not be good enough for certain use cases. Why not?

Scott Cohen 10:03

Yeah, I remember when the so-called RAG technique came out. RAG stands for retrieval augmented generation. And what that’s basically saying is if the world wide web and all the information that that comes from it wasn’t enough, you can impart domain-specific knowledge into the equation. That does increase the probability of the answer being correct. But again, the operative word is probability. It’s probabilities on top of probabilities. And that’s what I’ve seen emerge are rag techniques, LLM as a judge, people, uh, this is a more common theme of late. I I just gave a uh I was part of a pitch presentation. There were like 30 other companies, and a lot of them were touting this. Use a knowledge graph to baseline everything so that you have more truth. But that’s still probabilities and probabilities, getting you closer to determinism. But really, the only way to do it is the way that we’ve gone about it, which is bring in symbolic reasoning, good old-fashioned rules engines, where it either is or it isn’t. That’s really the only way to get that high degree of rigor. So I like to think of it as a spectrum of trust where the rag technique is just a starting point. I think knowledge graphs are better, but really the way we’re going about it with symbolic reasoning is the way.

Alex Romanovich 11:20

Got it, got it. Now uh Jaxon roots and to a certain extent your roots and uh in your earlier work and uh machine learning goes back to defense work, where you know, as we say, failure is not an option, and decisions need to be made very, very quickly and very decisively. So, what does that teach you and teach us that a normal enterprise customer or small, medium-sized business customer never really demands?

Scott Cohen 11:46

I think it’s really hard for a small or medium-sized business to do this same approach that we’re taking with the military. There’s a lot of work that goes into getting this right. And sometimes the out-of-the-box LLM is good enough. And again, it all depends on the use case. I think where people can learn it is around the policies that are being produced, the guardrails, if you will, that are being put on these LLMs, and thinking of it just like another tool that you piece together. It’s quite different than software in that it has the ability to make decisions and learn on the fly, but it is likened to software in that it should just work the way that you order it to work in a sequence that aligns to a human workflow. So I I like to say if you have a task that you can do under a minute with a human, that’s perfect for AI to automate. If that task requires a human to actually interpret something and think it through, then you can leverage AI, but you still want to keep a human in the loop to do those higher better for a human to tackle tasks and let the AI do all the rote manual tasks so that the humans have more time to focus on those tasks that really require their help.

Alex Romanovich 13:02

Absolutely. And mistakes will be made and uh you know, making those tasks still it’s not a perfect environment. Now, Scott, what is the most expensive mistake you’ve ever made as a founder? And what did it actually cost you to learn that mistake?

Scott Cohen 13:18

I think back to our first hire. I’ll keep his or her name anonymous.

Alex Romanovich 13:23

Yeah, no names, please. That’s okay.

Scott Cohen 13:26

We’ll protect the innocent. The mentality of building software is quite different than what a data scientist does day to day. And when out of the gate, I thought data science was where it’s at. That I mean, it sounds like that’s where AI is, and that’s what you need for AI. Some of the so the mistake was hiring. Hired the wrong people that weren’t builders, they were data scientists. So they’re great at analyzing data, not so much at building a product. So it was a rather costly mistake because it took me a little while because I I didn’t understand it as well as this person did, and I believed what they had was material. And part of it is that the space was moving so fast or still is that even if you think you know it, you might not know it tomorrow. But I invested too heavily in this first approach, and I should have been more more thoughtful around how this is going to be built out and not put all my eggs in that one basket.

Alex Romanovich 14:22

So besides this mistake, if you saw a young Scott of 20 years ago, what are some of the things that you would tell Scott and uh what are some of the things that you wish that somebody else had told you before you started?

Scott Cohen 14:34

Hmm.

Alex Romanovich 14:35

So not that you’re not that you’re not young right now, but but you know what I’m saying.

Scott Cohen 14:40

I think this is a systemic issue of every entrepreneur. They get so passionate about their product, they spend too much time on their product and not enough time selling it. And they they I hate this expression because I’m a perfectionist, but they go down the path of uh I mean the saying is perfect is the enemy of progress. And that’s one of the things that I think I would tell myself is don’t spend too much time getting the product to be perfect. Get it to be good enough that someone will pay you money for it so that you can keep making it perfect, but you don’t run out of money prior. That’s another lesson I it took me a little while to learn that as a CEO, your job is not only to manage and set the strategy, it’s to make sure the company never runs out of money. Absolutely.

Alex Romanovich 15:22

That’s not something they teach you at MBA in your MBA. That’s you absolutely, and that’s the one of the most difficult jobs out there that not many people wake up and uh you know, even inside of that company, wake up and think, you know, huh, am I going to are we gonna have enough money to do what we’re doing right now passionately, right? Yep. Now, aside from the uh being a passionate CEO and founder and entrepreneur and a really smart guy, you’re also dead. You’re also dead. Yeah. Just like you know, I’m a dad. And uh has being a father or just uh I guess getting older and building something that’s very demanding, changed your how you know your thinking about risk, about legacy, about what you’re actually building this for.

Scott Cohen 16:06

Mm-hmm. Yeah, first of all, best job ever. Certainly it’s changed the way I think. And risk is certainly part of that equation. I mean, as a a young lad, I might take all the risks in the world and not care what tomorrow brings, but with kids, you have to think about tomorrow. So that’s definitely part of it. Setting not only you know a path toward for the company or or an ultimate exit where we see good returns on all this work that we’re putting in over the years, is also setting the foundation that some other that follow me can um learn from and emulate. And it’s quite similar to how I’ve emulated my grandfather, who was the CEO of of Sweetheart Plastics, and his sayings, I call them Samisms now, that just last for generations.

Alex Romanovich 16:56

Yeah, that’s great. So listen, when uh this is all said and done, no matter what happens to Jaxon, I mean we wish Jackson all you know all the best and you know amazing success. But you know, whether it succeeds or fails or whatever happens, what do you want people and the people closest to you to remember Scott by? You know, and what what what do you want them to know that you stood for? Good question. It’s a deep question. Yeah, well, uh, you know, we’re we’re on the edge. So might as well.

Scott Cohen 17:26

The word integrity comes to mind. I think I’ve shared my favorite quote with you before: be brilliant and resilient. So you have to be smart and you you have to just keep going. That’s actually one of the Samisms is don’t quit. There’s a great poem that he he pushed on me that’s something the fact of the pace may seem slow, but you may succeed with another blow. The concept of just keep going. And a lot of entrepreneurs I find give up too easily. It’s that perseverance that really matters. So tenacity, I think, would be another word I’d I’d want people to remember integrity and tenacity. And leaving a legacy around building something that impacted the world, not just make money, but actually had value, which is one of the reasons why I I chose and kind of fell into, but decided to stay with the the military, is because we can actually be impactful. We can actually make a difference. It’s it’s very different from the days where I was working a lot in e-commerce. We could get people to buy more stuff, not very altruistic.

Alex Romanovich 18:28

Yes, it’s yeah, it depends on the stuff, of course. But you’re right about the mission critical, especially right now. I mean, it’s it’s incredible what’s going on. The very trying times that we live through right now, and Jackson and what you’re doing right now is uh is incredible and very, very appropriate and relevant. Thank you. Scott, we want to thank you for being with us. Thanks for having me. Opening yeah, and opening your sort of personal side of Scott Cohen, which is important, which is really important to understand, you know, who’s behind this product, who’s behind this movement, if you will, which is extremely important. Thank you so much.

Scott Cohen 19:03

All right, thank you, Alex. Nice talking with you.

Toggle Content