Why ChatGPT is a big block of marble which needs to be sculpted to your needs

In this episode, Charles Brecque interviews Meryem Arik who is the co-founder of Titan, a compression platform which helps compress NLP and deep learning models. Meryem shares how ChatGPT has increased the awareness of NLP models and how Titan can help businesses deploy their NLP models on cheaper hardware, much faster and much more easily. Meryem also shares how rugby has helped her be a better founder.

‍

‍

Watch the episode:

‍

‍

Find out more about Titan

Find out more about Legislate

‍

Read the transcript:

‍

Charles Brecque: Welcome to The Tech Story, a podcast about all things tech, business. We interview successful founders of fast-growing tech companies and today, I'm very excited to welcome Meryem Arik on the show, who is the co-founder of Titan, a very exciting compression platform start-up for compressing NLP and deep learning models. Meryem, thank you for taking the time for being with us today. Can you please share a bit of background about yourself and the company?

Meryem Arik: Sure, hey Charles. So, my background is by training, I am a physicist. Actually, physicist and philosopher, I did a joint degree and when I graduated, I didn't really know what I wanted to do with my life. So, I got a job, a corporate job in a bank, but what I realised while doing that is what I actually enjoyed is building things that I could see the impact on year after year after year and the work I was doing in the bank was quite deal flow related. So, the perfect place for me to do that and have that value was in a start-up. So, I moved into a start-up and I really loved it. I think that's when I realised the joy involved with creating things that people love and that's why I've been in the start-up space. But what we're trying to do at Titan specifically is NLP models, so, the kind of things like ChatGPT and any AI models that use language or text, they're incredibly computationally expensive. So, that means they're slow to run, they need very expensive computers to run and they're very difficult to get into production. So, what we, alongside my very smart technical co-founders, are working on is ways to compress these NLP models to make them much smaller, less computationally expensive, so they can be run on cheaper hardware much faster and much easier to get into production and that's what we're building now.

Charles Brecque: That's exciting and obviously, everyone is talking about ChatGPT right now. I, I've obviously seen all your posts commenting about it. Do you feel like the mass adoption of ChatGPT is, is helping you? What are your thoughts on the whole situation?

Meryem Arik: It's fantastic for getting awareness of these kinds of models. So, when we were first fundraising only about six months ago, we had VC's asking us if any businesses would even use AI or language AI in, in business and ChatGPT has, kind of, blown those questions out of the water. It's so obvious how beneficial these language models can be in almost everything that we can do, in, in almost every single industry. So, it's been really helpful from that point of view. The flip-side is ChatGPT is kind of like using a sledgehammer to crack open a walnut for a lot of use cases. So, we see a lot of businesses building on top of ChatGPT and spending a lot of money in getting very slow and, you know, sometimes poor performance for tasks that really don't need that kind of, of very, very high-powered model. And that's, kind of, the re-education work that we're doing at the moment. So, cheap-, like, educating our users about what they actually need to be doing with their models to get the most performance for, for what they're paying.

Charles Brecque: And, and in terms of your users, do they have their own models? Are they using ChatGPT? Where, where do you, sort of-, what is the minimum set of requirements that your users need to be able to use Titan?

Meryem Arik: So, typically, they'll be using open source models. So, Hugging Face is a fantastic open source platform which hosts, you know, loads of, of open source models created by researchers at Meta and-, at Meta and Google and Microsoft and all of those. And typically, the way that you'll build an NLP application is you'll take an open source model like that, you'll collect a bit of fine-tuning data. So, let me give an example. Let's say you are a financial services company and you want to create a bot that will automatically read all of the tweets in the world and tell you whether they're positive or negative about a particular company that you're interested in. Let's say I'm buying Apple stocks and I want to see if they're tweeting positively about Apple or negatively. Then what I might do is create a-, create a bit of fine-tuning data of examples of tweets and then label them as positive or negative and that's my fine-tuning data. And then I can take that data and train it on top of mine open source model and that's, kind of, where we come in. Given that you have a bit of fine-tuning data and you're looking at building models yourself, then we can be really helpful in getting them into production.

Charles Brecque: And, and so you then take that model, put it in Titan and then it's super lightweight, super fast.

Meryem Arik: Yes. I mean, essentially, yes. So, a nice way to think about these foundation models that we start with is kind of like a big block of marble. Like, they're a starting point. So, my foundation model, like ChatGPT, for example, is able to write love letters and also write code. But actually, as a user, I never need it to do all of that. So, what we do is we go in and we cut out the bits of the model that aren't relevant for whatever task your fine-tuning data wants to use it for. Kind of like cutting, you know, the David out of the block of marble.

Charles Brecque: Exciting. That sounds really-, a really nice way to, to frame it because obviously, it's, it's always useful to have a Swiss Army Knife but the reality is, is maybe you should just take that one, one knife or one tool that, that you need. So, that's great. Obviously, you've co-founded this business. What lessons have you learnt so far?

Meryem Arik: That's-, well, we made a lot of mistakes, like I'm sure every single founder does. I think the biggest learning curve or the first learning curve we, we came across was figuring out how to fundraise. It's a bit of a strange skill that you have to acquire and it's all about storytelling. So, can you take people along the journey to believe what you, you believe and hold so dear and that was something that we took a long time, like, to be able to iterate and find that story that really explains why we're passionate about what we're doing. So, that's probably the, the biggest kind of lesson, is figuring out how to get across your, your passion.

Charles Brecque: Yes, I think fundraising is a skill which I, I definitely haven't mastered or acquired.

Meryem Arik: It's, it's, it's not an easy skill but you-, I mean, you guys did a great round, so, you must be somewhere there.

Charles Brecque: Yes, I, I feel like it, it, it definitely wasn't the-, we were very lucky to have the round and it was a great round but it definitely wasn't a, a walk through, through the park and-, but, but I think you're right, the story is really key and sometimes, actually, that story, you need to refine it based on the investors you're speaking with because they also help you understand the problem from a, is this actually an investable business and, and sometimes, your story might be the right story but there just needs to be some refining around, well, you know, will it make money for the investors or not.

Meryem Arik: And, and typically, especially good investors, know what they're talking about and quite often, normally come with very good advice. So, I know a lot of founders hear the, kind of, feedback or knows from investors, like, they don't know what they're talking about or, you know, they've missed a big opportunity. We've got a huge amount of value by really absorbing that feedback and then looking very critically at our business because quite often, these are people that are very experienced and really know what a good business looks like. So, there's a lot of learning and iteration cycle and there's, there's, kind of-, there's absolutely nothing wrong with taking advice from investors.

Charles Brecque: Absolutely. Even if it's bad advice, the wrong advice, it's still, you know, advice and you need to-,

Meryem Arik: And it might be that it's bad advice, advice because you've not explained yourself properly. I mean, at the end of the day, it's almost always my fault in some way whether they actually have a really good point and I need to address that or I don't agree with their point, in which case, I haven't done a good enough job at explaining why I think, you know, what I've done is right, so you can-,

Charles Brecque: Yes, my, my old boss always said that the, the burden of communication is always on the communicator. Ultimately, if someone doesn't understand-,

Meryem Arik: It's your fault, yes.

Charles Brecque: Great, and obviously, learnt a lot and yes, everyone makes mistakes. What's, what's been your favourite moment so far?

Meryem Arik: When we gave our product to a user for the first time, that was-, we spent a long time building up the algorithms and building up the product in stealth and we were working with design partners to, to ensure that it was right but it was, kind of, like a-, we hadn't actually got it into an unbiased users hands yet. And when we first did that about a month ago-, and we got, like, you know, both good and bad feedback, like, things that we can improve but also validates a lot of things but it was just super lovely and, and satisfying to know that it's actually coming real, people are using it. You know, they think it'll actually help solve their problems and there's still obviously-, you know, we can still obviously make it better but that was just a really, really nice moment.

Charles Brecque: No, that's great and I think, yes, whenever the, the first person uses your product, it's always an exciting moment and also, you never know really how they're going to use it or what they're going to use it for because everyone, everyone-, well, especially at the beginning, you don't have the full understanding of all the use cases. I guess based on the, the traction you've had so far, what, what are the key use cases? Obviously, language models are, you know, quite broad, so-,

Meryem Arik: They're kind of everywhere, yes.

Charles Brecque: Yes, so, so, what, what specific use cases are, are people using you for?

Meryem Arik: So, we are a way of augmenting your ML and specifically an NLP. So, the industries that are using a huge amount of NLP is finance, insurance and legal, which is where you sit and these are three industries where you see a huge amount of text, a huge amount of text and voice as well. For example, in legal. Like, if you think about the number of contracts that you have to try and look after, which is clearly the, the, the use case you guys are looking at, so those three entries, finance, insurance and legal are really nice sweet spots for us.

Charles Brecque: Great, well, hopefully, you don't take over what we do or-,

Meryem Arik: No, no, no, if anything, we would make you guys more powerful.

Charles Brecque: Yes, no, for sure. Well, I, I know that behind the scenes, there are those conversations or happening but no, that's all really exciting and I guess, you know, you're, you're, you're growing the company. Just done a funding round, congrats.

Meryem Arik: Thank you. Congrats to you.

Charles Brecque: Yes. You know, what, what is the, the big picture vision look like. Where, where do you want to see the company in three, five years time.

Meryem Arik: So, we want to supercharge every single NLP developer and the way that we measure success is the number of models that are in production that have used our technology or been through our step of, of the chain. I want every single NLP engineer, especially in the enterprise, to have the Titan step as a key part of their workflow. So, they get their model, they get their data set, that's one step, then they go through Titan and then straight into deployment. That's what I want to see. So, for me, the measure of success is the percentage of models in the world. That's been-, come through our pipeline.

Charles Brecque: And I guess you're at 10% right now?

Meryem Arik: Yes, at least. At least.

Charles Brecque: Perfect, well, that's an ambitious goal and congrats. I guess, as a co-founder just getting started, I imagine you must have come across some contracts or legal documents.

Meryem Arik: One or two.

Charles Brecque: What can you, sort of, share about them and are there any tips for, you know, the next, you know, up and coming founders that are, sort of, listening today?

Meryem Arik: Yes, I am not fluent in legalese at all. So, we've come across a lot of contracts and the first thing I do is ask advice from someone that knows what they're looking at. I make guesses and I can read the plain English and I do my best but fundamentally, I know there's-, I, I know there's a lot of things that could be hidden in these things, so, I get advice from people that know what they're doing with contracts, I think is, is a really important thing and what I love to see in contracts is when they're written in plain English and are as concise as possible because actually, for most things, you don't need pages and pages and pages, you can do one or two pages. But yes, that was definitely a learning curve.

Charles Brecque: Yes, no, I think seeking advice is always important and, I, I think even if something is written in plain English, in the context of the contract, it can mean something completely different than what you expect. So, I always find myself getting caught out and I always try to seek legal advice when I can but I guess something that we, we try to do with legislate is translate the contract into questions and answers which are easy to understand but also do what they say, if that makes sense and that way, if you receive a contract and legislate, you don't need to worry too much about, well, what, what, what will happen or, you know, what does this actually mean because it's explained to you in the form of questions and answers and that's all you need to understand.

Meryem Arik: Interesting. So, is that if you both receive it as well as generate it?

Charles Brecque: Yes.

Meryem Arik: That's amazing.

Charles Brecque: So, we, we always present both views to the users and I guess there's a lot more that we could do to, to make it, you know, more explainable, even for example what we're working on now is having more deeper semantic models of what's in the contract so that if confidential information is defined, then you could say, well, what counts as-, like, is X-, is, let's say, a trade secret confidential information and it would say because it's defined or-,

Meryem Arik: That's really cool.

Charles Brecque: Yes, it falls under this, so-, but, but it's all about, how do we make, make those contracts, like, easier to digest, easier to query. Right now, we've done it just in the form of questions and answers but we're now getting deeper.

Meryem Arik: A step further. No, that's amazing that it's on both sides because I always-, you know, when I'm generating contracts, the questions answered are super useful but also receiving it, that's, that's a really nice bit of-,

Charles Brecque: And, and from an engineering perspective, it also makes it easier.

Meryem Arik: Does it? Well, that's really handy. That's fantastic. You know, you should always keep things as simple as possible but no simpler, right? So, yes, you've smashed it.

Charles Brecque: Well, we're trying and I guess-, I know you've obviously been, been building Titan for not too long but-, and obviously there's a great piece of advice about seeking advice and, and making sure that the contracts are in, in plain English. Have you, sort of, found any challenges with contracts that you've, sort of, been really frustrated with and, and would love to solve?

Meryem Arik: Great, I mean, when we were doing fundraising, it's hard as a founder to know what is market standard and what isn't without seeking advice. So, for us, you would-, you would get a term and, you know, as, as a first-time founder, this is the first-, you know, or second or third term sheet I've ever looked at, like, I don't know if this particular clause is standard, if I can push back against that. And the only reason I got that information was through seeking advice. So, I would say, like, that was a key area where we were, like, not really sure about, you know, what's, kind of, off-, off market/on market. Similarly, with, like, employment contracts, like, none compete. So, we don't know how off market it is to have those anymore. So, I guess there's a sense of what's normal to see in a contract, is something that you get from experience which we're still, kind of, developing. So, seeking advice in all of those areas was, was super important.

Charles Brecque: Yes, no, I, I think if you-, if you always start with standard and market standard, then at the end of the day, there, there's minimal negotiation but I, I, I do find that if you don't know, then you end up with something which is completely off and then that's when the negotiation starts and that's when the delays happen and obviously with legislate, we are building-, we're, we're aggregating data and, and we are presenting that back to users so that they know what standard is.

Meryem Arik: That's really valuable.

Charles Brecque: In a, obviously, anonymised way but I think it-, yes, knowing what's standard is something that people have experienced or businesses have experienced.

Meryem Arik: Yes, like we're currently working with our lawyer on creating our, kind of, standard SaaS agreement and the conversation we've been having with her is, just make everything as standard and unobjectionable as possible because I don't want to have to go through a negotiation every single time we sign a lease. But, you know, that's really awesome.

Charles Brecque: Yes, great and if you were to receive a contract to sign, what would impress you?

Meryem Arik: Clarity of both structure and thought within, within the contract. I've seen so many contracts that I've looked at and I'm like, this could have been a page, like, what you're actually discussing is very, very narrow. So, I think that's the key one for me, yes.

Charles Brecque: One of our investors is a lawyer and any contract that he generates is literally a page. Obviously, you can achieve that with formatting but also-,

Meryem Arik: What, just have it on, like, size one font?

Charles Brecque: No, no, I mean, more, more, like, having columns but, but, but you can obviously-, if you're concise and to the point, then, then that, sort of, achieves that.

Meryem Arik: It's not like we're negotiating M&A deals here. Like, you really don't need anything super, super complex.

Charles Brecque: Exactly and one thing we've, we've also found through the podcast and interviewing founders is they, they all have interesting hobbies. I know you used to play rugby.

Meryem Arik: I did used to play rugby.

Charles Brecque: Professionally or?

Meryem Arik: Semi, semi professionally, yes.

Charles Brecque: How'd you-, are there any, sort of, lessons or anything that you've learnt from rugby that you're, like, applying in your day to day as a founder.

Meryem Arik: Completely. The main one for me is, rugby has the most amazing team spirit and this sense of togetherness and there's a quote which we used to say at Saracens which is, the strength of the wolf is the pack and the strength of the pack is the wolf. And I really believe that when it comes to team building. So, we will only get where we want to go together and if we pool together, but we also individually need to be working as hard as we can. So, that's-, the, the team aspect has made a huge, huge impact in the way that I see team building. The second one which actually comes to mind is the sense of perseverance. So, rugby is a horrible sport to play. Like, it's gruelling training, the games suck, especially as a forward, like, you're doing all the dog work and there's this quote on the gym that used to say-, or probably still says, it was about a parable of a-, of a stonemason which is the stonemason can hit the, the stone 100, 1,000 times but it's on that final thing that everything breaks but it wouldn't have broken without all of those hits that came before it. So, it's the sense that, like, what you're doing may seem futile, may seem draining and pointless but it will crack if you're-, if you're, kind of, going in the right direction. Same thing with, you know, that's how I played my rugby and that's, kind of, how we build our business.

Charles Brecque: Well, every-, I think especially when you are building a start-up, every, every day, sometimes it does feel like, where are we going or are we moving in the right direction but one thing I've definitely found over the past three years is-, or at least, every, every year I look back and see what we've achieved but when you are in that heads down, it is difficult to see that and I guess that you just need to persevere and-,

Meryem Arik: I mean, we had that, actually, this week when we were speaking to an advisor who we hadn't spoken to for about three months and, you know, we told him what we were up to and he said, like, 'Guys, I'm being serious, you are unrecognisable from three months ago' and we were like, are we? You know, because we-, we're in it every single day but having that growth mindset, okay, how different are we from three months ago and can we achieve the same thing in three months time, it's super satisfying to see that.

Charles Brecque: Yes, we just need to persevere and-,

Meryem Arik: Keep on-, keep on keeping on.

Charles Brecque: Yes, exactly. Obviously, tech founder, I imagine you use tech products. What's your favourite? It can be hardware, software. What, what tech products do you love and, and can't live without?

Meryem Arik: Great question. Really boring but the Google Workspace suite. I love Google Slides, Google Docs and Google-,

Charles Brecque: Sheets.

Meryem Arik: Sheets, that's it. Super boring but I just love them. We did-, we did use Notion as well. Not the biggest fan of that because no-one updates it but yes. What am I saying? Slack. Okay, definitely Slack. What's yours?

Charles Brecque: I, I, I really like Superhuman.

Meryem Arik: I've heard good things about it.

Charles Brecque: More so because it-,

Meryem Arik: I've never tried it.

Charles Brecque: I reached a point where-, I've always been an inbox zero person but I reached a point where I couldn't keep up with the influx of emails and-,

Meryem Arik: Too many inbounds.

Charles Brecque: Too, too many inbounds but it's just a great-, I just-, I've been told that Gmail also has the reminder functionality but I just, from a user experience, prefer the Superhuman way of just getting reminders.

Meryem Arik: Interesting. You might have to show me that.

Charles Brecque: Yes, I, I can show you in a bit. I've just actually been introduced to Cron, which apparently is the Superhuman equivalent for calendars but I'm just getting used to that so I, I, I'm not yet convinced or not yet fully on board with how amazing it is yet but I'm, I'm optimistic. But, but I used to be like you, I used to just love the G-suite and-,

Meryem Arik: I'm a simple woman.

Charles Brecque: Well, I'm a simple man but, but recently, we've been, sort of, introduced to Notion and I like it because it, it-, ultimately every page is a database. And being able to have databases that you can embed within databases can be-, it obviously can get quite messy and it doesn't always make sense but I definitely feel that we've, we've reached a point where it does make sense.

Meryem Arik: Yes, I love Notion. I, I, I love Notion personally but as an org, we're not good enough at keeping up to date because we have that, then we have the docs, and we have GitHub and it just becomes-,

Charles Brecque: Yes, I think that's, that's a whole other story around how you, sort of-,

Meryem Arik: I did try this thing that I love but I never renewed my subscription because I thought it was very expensive, was Motion. It's this, like, calendar/to-do list thing where it does time-boxing. So, you-, instead of having a to-do list you say, okay, these are the tasks I have to do, this is how important they are and this is how long they'll take and then it schedules them into your calendar which I, I really liked but I thought it was a bit expensive for what it was offering.

Charles Brecque: Yes, well, I'll, I'll have to take a look and that's great. So, I'm, I'm conscious I've taken a lot of your time already. It has been great, great chatting and learning more about Titan and Motion and all sorts of cool things. If you were to give one piece of advice to a new founder or someone who is considering being a founder, what would it be?

Meryem Arik: Great question. Firstly, ensure that you're crazy enough to do it because it is-, you, you have to be a little bit of crazy to go down this road. The second thing I would suggest is really take time validating your market and your problem and really ensure what you're going after can only be solved by you and you are the best person to do it. And then the third thing is once you've, kind of, decided you're crazy enough and think you have a real opportunity, just keep going. There's-, just keep on keeping on. Have that perseverance and when things aren't going well, think about how you can change and adapt, whether it's your storytelling, whether it's how you interact with your team, whether it's the product but keep nimble and keep ambitious and keep persevering.

Charles Brecque: Yes, I, I also think-, and, and I don't know if VC's  will always agree but I also think that markets are always bigger than we think and obviously, we might hear about competitors, we might hear about their fundraisers or their achievements but at the end of the day, it's all noise and as long as you just persevere and don't make too many mistakes, then chances are you, you will succeed.

Meryem Arik: Yes, for sure. I like that about the, the competitor angle, is-, you know, Jeff Bezos always says, like, we're customer obsessed, not competitor obsessed and then if you really find that product market fit, you solve that pain point for someone, you'll be fine.

Charles Brecque: No, exactly. Well, thank you very much, Meryem, for coming on the show. It's great having you on The Tech Story and best of look growing Titan and conquering the world.

Meryem Arik: Thank you so much, Charles.