The Future of AI is Private: Gavin Whyte's Bold Vision at Brew AI | Ep. 191

0:05Hey everyone, welcome back to Founder Story. Today we have Gavin White and Gavin you are the global CEO of Brute AI with over 10 years of experience working with a former chief scientist. at Deloitte Australia. We're going to get into all things AI, which is my absolute number one topic. I think it's our most popular topic right now because I can be a little bit pessimistic, a little bit optimistic, but I'm going to be very curious around your thoughts with the future of AI. But before we get into that and all the things that you're building at BrewAI, how did you get started with AI specifically?

0:45And then why did you create brew ai daniel firstly thank you for having me here today um so how did i get started kind of a strange story actually um but i'll tell you anyway uh so when i was younger i um i was actually a designer so i love to design clothing uh and i was really good with mathematics for whatever reason. And one day I was really angry with my parents because I was in the studio and it was there, you know, they had a whole fashion house and they had a background in retail and clothing. And I said to them, you know, this is not quite working. I'm changing all the designs. And they got really angry with me.

1:29So they kicked me out of it and said, you're going to go to university and study mathematics. So I went to university to study mathematics, but I didn't quite take mathematics, I took computer science and mathematics and realized that I actually loved it. So much so that I got into it and it was amazing. I started to program and that's how I got myself into AI. I started to do statistics and write algorithms and code these algorithms at scale and I was really good at it. So that's how I actually got into it. I know it was, it's strange, but hey. Well, it's a good thing that your parents pushed you to do that.

2:15Otherwise you might've been in fashion, right? Although I didn't say you are very fashionable and I need, I need some help on that. And so maybe AI, you can build me something that AI can help. But so how, so amazing story. Obviously you've been doing it for a while. You've seen a lot of the transitions the last 10 plus years, then why through AI? And I've seen what you're doing with private LLMs, which I feel like are going to be a massive part of our future. Absolutely, Daniel. So one of my roles previous to starting as chief scientist, I was the adjunct professor at the University of Technology in Sydney.

2:54So I was teaching deep learning and mathematics to students. And some of the key areas where I did a lot of research in was in neural nets and backpropagation. Not to go too technical, yeah, so I'm going to try to keep it a bit high level. And one of the key areas our focus is was in research. You know, how do we ingest large volumes of data? How do we predict accurately? And those days, I was doing predictive analytics with neural nets, not only just your basic predictive algorithms. And that's where the interest started quite a lot. And I got into the industry and previous to Deloitte, I was the director of data science at KPNG.

3:39And I started to apply this at scale in banking, et cetera, et cetera. And utilizing that background and thought process, I started to put a lot of effort into how the neural net was evolving and how we can apply it. and that's how I've led into this whole AI, into this entire AI and neural nets and LLM area. It was around 2019 when Stanford released the paper on neural nets when I really got really excited. I know predictive analytics, you need scientists and now suddenly we've got this large language models where the AI can talk to you and produce information. That got me really excited. That's when I set up a team of researchers around the globe from Oxford University, University of New South Wales.

4:37And that's when BrewAI evolved. And I left Deloitte and I started BrewAI to actually then research, build these platforms and grow that in the industry. If somebody has no idea what a neural net is, can you explain that? Yeah, sure. I'll explain it in probably the most simplest way. so neural nets you think of it like your brain you know it has connections etc and it can ingest large volumes of data prior to neural nets you have your basic algorithms that was stifled by data and stifled by training but neural nets allowed you to grow and hinton as we call the godfather of um of deep learning invented what we call back propagation which allowed scientists like myself and others to ingest large volumes of data and give you accurate predictions.

5:29That was the whole premise. So think of it where the synapses in your brain grows, the same with neural nets, the more data you feed it, the more intelligent it became. And then you apply a whole lot of algorithms onto it. And that's how we came up with LLMs, or the scientific community has. No, Gavin, that's amazing. And thank you for explaining that because I think a lot of people are okay when it is a, you know, a techie explanation, but I think a lot of people are being lost because there's not a lot of people explaining things. Uh, you know, and I feel like this is at one of the most amazing times in our lifetime.

6:08I believe we're, we're at this inflection point where all these technologies coming together. So, So you've done all this. You create Brute AI. And there's the reason why you created it because there's a lot of talks around corporations and data. If you use ChatGPT, for example, will somebody else see what I'm having? Or if I use DeepSeek, who sees that data? There's a lot of these talks around who's really having access or who can see this data? Is that why corporations are more apt to use a private LLM to ensure that sensitive information doesn't get out? Yeah, absolutely, Daniel. And I think one of the key areas in corporations is to keep their data safe, private.

7:00So if you start using other types of LLMs that are open, and generally I don't like to refer to their names, you are sharing your data with the world. You are sharing your company's IP with the world. Whether you ask it a question with fake data, you are sharing how you think as a company and you are training that LLM. Now, if you're in a business and you want to keep your IP safe, going to these open LLMs, as we call it, and sharing your questions and your information to the whole world will only train that brain. That is one very large, big brain that works at scale. which is not specific to any industry.

7:44But when you start using private LLMs, we can encrypt the data, we can store it on these servers, and we can provide the same level of intelligence as you would get on the larger ones, on smaller machines, but more accurate. Are you finding that governments around the world or government organizations, are they leaning on this? And I'll talk a little bit about that from an Australian perspective. When we did start in Australia about three years ago, we started to grow into mining groups, into big legal firms and into government as well. And the reason for this is because governments want to keep citizens' data safe and they don't want the private LLMs to be biased.

8:31So they want to train it and control it to allow citizens to interact with it without outside influence. Hence, that's why we have deployed it in these scenarios as well. We use it for case law or for legal companies where the LLM can analyze the cases and very specifically spit out the outcomes. And there's one thing I want to point out, Daniel. You may have used, you know, ChatGPT and other types of GPT where you may have heard it made up information. okay it created information michael cohen is a famous trump lawyer where he submitted information and the and chad gbt created a whole case which was all fake which he submitted to the judge which he had to retract when they did find out that case was made up now we call that hallucinations in the tech industry these llms hallucinate and that's okay it's it depends on the level of hallucination that you want to reduce.

9:37So what we did is last year, we released a global mathematical paper on how to reduce hallucinations. We were first in the globe. And we were very welcomed by all scientists around the globe at this very large mathematical conference. And we actually solved that problem. So we have a very extreme rate of hallucination reductions around above the 90s. And we've included that in this private technology. So now we're starting to see we're in the forefront of how we get answers out from the questions you ask it without it hallucinating, which was a huge milestone for the company as well. My LLMs hallucinate all the time, my GPTs.

10:24And I've gotten into the situation where I have used wrong information, not knowing it was wrong information. So I can see the massive value and I'm doing small things. I can imagine if it's if I'm an attorney or something. So can you can you talk to me, too, about so is a private LLM like an empty brain and it only works as you add things? Or does it like if you give it to a company or a company starts using it, does it already have a certain amount of information based on what they want? How does this work? Yeah, so there's many ways to use private LLMs. There's a lot of open source LLMs, which you can incorporate into private or you can train your own.

11:04So we do a whole mixture of all of these. You have your base models. But I think what's really important is when we do test this, for example, in legal, you know, we had to give them a base model. You can train it with there's a lot of court cases around the globe. You know, they all open anyway. but in the Australian setting you cannot train it with existing client data so we don't do that so we want the LLM not to be biased as well so we give them base models and these base models they upload the data they index it and just by uploading a PDF or a whole folder of cases and they can ask it all the questions and it does a very, very good accurate outcome and one of the largest law firms in Australia today uses our product.

11:55So we make it as easy as we can for our customers. We have these base models. They don't need to do any more training. You can't from a legal perspective, but you can in other industries, in marketing, et cetera. You can actually utilize this LLMs at scale and further fine-tune it to refine it to your outcomes as well. Thank you for explaining that because no one has ever, told me the details and the detailed differences between public and private. I can see a massive industry. I would imagine this might be the fastest or highest grossing part of AI going forward just because it's a huge topic around these companies using, like you said, something that is more open.

12:45So when you look at the future of just artificial intelligence, in general, there's talks about AGI, and you know, some people say a year, some say 10, then there's talks of possible ASI, if that's even a thing, and we have no idea what that even means. Technically, unless you do, I would love to hear, you know, but what do you see as the impact on society overall? all let's talk about the definitions of the two um you know agi refers to an ai system with human level intelligence and ai psi um it's where it surpasses human intelligence in all areas so let's just talk about those two and and and just drill down a little more i think with agi we are getting closer.

13:35There's no doubt about it. We've released a reasoning model recently that allows you to actually observe through an audit trail how the LLM thinks, which is kind of scary sometimes when you look at it and look at the answers it's producing. I wouldn't say we are there yet. We are getting closer. ASI, we're not too far away from ASI because once quantum computers go live and become a more affordable and adaptable, we will get there. Because at the end of the day, it only comes down to one little thing. It's how much data you train it with and how it actually produces the outputs. With the reasoning models, it's quite interesting.

14:20We went from what we call our legacy LLMs into deep reasoning. You may have heard that from DeepSeek. It tanked the NVIDIA stock recently. You use less GPUs, less power, and better reasoning. capabilities. And there was nothing new in the algorithms there. There were just existing algorithms and they were just reordered to give you a better outcome. But we can clearly see that we are getting closer to human reasoning and sooner or later it'll start making decisions for you as well. The impacts, now the biggest question I get asked is, what's the impact? I'm sure you're going to ask me that. what's going to happen to the human race.

15:05But I'm happy to answer that now. Are we going to have to, since there might not be many jobs left, are we all going to be on universal basic income? What's the future look like here? By the way, for me, we're already at AGI because LLMs are already way smarter than I am. So in my world, I am already less smart than AI. But please, I would love to hear your opinion. Look, these AI systems are really good at prediction. LLMs are just predicting the next word or predicting the next sentence or predicting the basic reasoning. Now, I've said this at a Google conference. I think it was around 2018 and 2019.

15:45Okay. Human predictions are weak. The AI is going to do better. They're going to produce better predictions and better outcomes. But I do not think the AI is going to replace human judgment. What it will do is make us produce or judge better outcomes. Okay. It's going to make us smarter, quicker, faster in our thought processes and make us more wealthier. So that's what it's going to do. It'll choose better stocks. It'll allow us to decide. It'll run our lives and make us, you know, gives us the free time at all those menial tasks we have done before. Okay. I really don't want to talk to a machine in the future.

16:29I want to have a cup of tea with my friends and have normal human conversations. But all those menial tasks will be gone. And AI will come and help us to lead, you know, so that we can lead better lives. I mean, hey, we're happier. We're spending more time with other humans and our friends and family. And we're richer. I mean, that sounds like, you know, paradise. This is like the perfect storm. And I can't wait. I can't wait for this to happen. But is there any concerns about AI around the world? Or if you don't have any concerns, maybe can you share anything from the research you've done in terms of is there something that people don't know, but probably should?

17:17Well, there's a lot going on in AI. You know, the scientists around the world are just going to make it smarter. I will tell you this, though. The deep reasoning is going to get smarter and smarter and smarter. It's because of the way we're reordering the algorithms and the amount of volumes of data we have. OK, so we can ask other LLMs questions and use it to train our private LLMs without sharing info. You know, we can do some amazing things these days in the labs with these LLMs. So we are going to get smarter at what we do. Where I do get afraid sometimes is autonomous weaponry, but that's a whole different ballgame that I'm not even going to go into, given that I worked in these industries before.

18:08You know, that is my biggest fear. but overall for just better health and human life I think AI can actually help quite tremendously it can you know why do we need to go to the doctor maybe once a year twice a year when we get sick why can't we track and trace our heart rate and etc etc using AI and it could recommend things, improve our diet, improve our lives, our lifestyle. I think that is important. Understand our heart rate, our anxiety levels and help us through that process. I think AI can be useful, good. And, you know, it'll improve our lives quite tremendously. Yeah, we just had a guest on who is building AI to help therapists, because there's this mental health crisis, and there's not enough therapist.

19:02And that got me excited. I was like, you know what? I love the health angles and how, like you said, we can, why, who knows how many times we can catch something before, because we're not always very preventative when it comes to healthcare. And in many countries, preventative healthcare doesn't even exist at all. It's very reactive. And if they get sick, wipes out their entire wealth of their family for generations. So what if we can solve these things and the cost gets lower and lower. So Gavin, you have got me. I am going to change. I'm like more optimistic now than pessimistic, which was the opposite before we started talking.

19:41So I appreciate you today really diving in and explaining these things because I've been wondering all about this private LLM, how this is working, data, biases, hallucinations, everything there. If people would get in touch with you, they want to find out more about BrewAI. I know a lot of people that need this. So how can they do so? Yeah, so our website is brewai.com. My name is Gavin White. You'll find me on LinkedIn quite easily. I guess everyone finds me on LinkedIn. And happy to share my email. It's gavin.white at brewai.com. So happy to share my details as well. Yeah. So, Gavin, I would love to dive in with you more about the products that a brew AI is creating or building or how companies are using it specifically.

20:39Yes, Daniel. Thank you for asking. We focus on multiple areas. One of them is in government. But the more exciting areas we are focusing in is in law and finance. In finance, it's quite interesting. We have Merchant Bank using it to analyze financial reports by just uploading PDFs. And the very exciting part of this is we're currently building an accounting and tax platform for small and medium businesses utilizing deep reasoning or generative AI. Now, this is exciting because your data kit is safe and you can actually talk to your accounting system or type questions in to say, give me an instant cash flow.

21:24Give me a predicted cash flow. Could you do my tax return for me? So that's very exciting. And we're about to release that very shortly for millions of businesses in the United States. I'm really excited about that. I think LLMs and deep reasoning can help small, medium businesses to really grow their business quite significantly and not worry about all the main-neal tasks of returns or understanding the cash flow, etc. I'm curious. This is amazing. it seems like you could solve a million problems i mean i could i could just i could uh tell you 100 new products and services that brew ai could additionally solve uh just because there are so many that need to get better like you said you can do so much with this how do you hone in because i can if it was me i have adhd i would just be going crazy like i would have like a thousand things which obviously is like impossible to do.

22:23But how do you really hone in? Just because there's, it sounds like to me, BrewAI could solve and create better products and services for so many industries in so many ways. Now, Daniel, I have that question asked quite regularly, you know, and it's great that we can do these things. But, you know, as a business and, you know, we really need to focus on specific industries. And while we grow those industries and while we grow the company revenue, we will then tackle other types of industries. You know, we are already in law. We are already in merchant banking scenarios. We are working with small and medium businesses and accounting practices to optimize that entire process.

23:08Absolutely. We can do it for marketing. We can do it for literally every industry. I think when we first started the company in the first year, we did have a try of many things. and very quickly realized we can't tackle them all. So we had to do a bit of focusing. And given my background coming from Deloitte, from audit, in consulting, et cetera, I decided to take the areas that we were more familiar with. And then from there, obviously, we'll hire individuals to grow into various other industries as well. I can't wait to try this new accounting software. where it's almost like having my own accountant 24-7.

23:50How do you feel about, a lot of people are talking about AI agents or agent AI, they're talking about both. This seems to be the thing, like everyone's talking about AI agents. I don't know if everyone fully understands how this will impact us, but how do you see the future of AI agents? So I think, you know, as we grow, you might, AI agents will be overlaid into these reasoning models or into these LLMs and will play a very specific role. I'll give you an example in accounting. So you might have an AI agent that's a senior auditor. You might have an AI agent that's a junior auditor. So the AI agents are going to talk to each other and start to audit the agents.

24:33So the agents can be audited by other agents. And this is quite exciting because you can train each agent to look at it in a specific manner. And this is how exciting it becomes. So the question then becomes, how do we pay these agents? How do they work in the firm? Do we have these human and agents working together? We call them human in the loop. You may want to have an agent that analyzes it and a human that analyzes it and might go to a senior human or a senior agent to do the final check. So it's going to be very exciting because we are ingesting these agents to help. It's basically to help humans better do their job or improve their output in the workforce as well.

25:18So it could be looking at banking fraud for that matter, or look at unusual transactions that goes through your business where the AI agents can immediately alert the human in the process. And it could look at your transactions 24-7. So there's a whole range of ideas that comes out of these agents. It's very exciting with AI agents being implemented in the workforce today. I mean, I need to use this accounting software. So I'm excited when this comes out. I need my own senior AI agent accountant. That sounds amazing. How are you seeing AI now crossing over into humanoids or machines mixed with AI?

26:01I know there was, you know, talks a few months ago that everyone will have their own machine or humanoid robot that can do whatever tasks you need around the home? Look, it's still early days. Yes, it will work really well. We've seen good examples of it. I haven't specifically focused on robotics, but given robotics and with the emergence of AI, we'll find manufacturing will become easier, goods will become cheaper, better quality outputs as well. That will play a very important part in the home. It will help with cleaning and cooking and all the tasks you don't like or don't want to do. So it's going to be very exciting.

26:46And, you know, it's definitely going to help humans to a degree where we become so reliant on these AIs to help improve our lifestyles. But this has been great.