When Mike Keating and Haziq Nordin founded Enhance Labs, they raised millions without a product, a pitch deck, or even a bank account. Backed by Blackbird and QIC, they became known for their unfiltered takes, chaotic energy, and refusal to ship something they didn’t believe in, even as the rest of the AI world raced to launch.
For months, they experimented in the shadows, building, pivoting, and learning why most AI startups were destined to fail. Then came the breakthrough: a way to let every user design their own personal internet, a “living interface” that could reshape how humans interact with software.
In this episode, Mike and Haziq share why they chose velocity and vibes over hype, how playfulness and curiosity became their competitive edge, and why founders must know when not to finish something. They break down what AI products will collapse by 2026, the open problems that still need a Nobel Prize, and why the next wave of innovation will come from the weird ones, the teams laughing the loudest while everyone else plays safe.
🙋♂️ Mike Keating: https://www.linkedin.com/in/mikekeating/
🤖 Haziq Nordin: https://www.linkedin.com/in/haziq-nordin-1a6623107/
🚀 Enhance Labs: https://enhancelabs.ai/
Transcript Synced · click any line to jump ▾
Georgie Healy: Founders scale faster on Deel. Set up payroll for any country in minutes. Hire anyone anywhere. Get visas handled fast and get back to building. Visit deel.com/dayone. That's D-E-E-L dot com slash day one.
Mike Keating: We were able to raise money without anything. Like we didn't have a product, we didn't have a vision, we didn't have a TAM, we didn't have a pitch deck, we didn't have a bank account. We literally had zero when we got the offer, which sounds awesome, but it is honestly like, it's really, really hard because you've got nothing and then it's like, cool, what can we do? And honestly, like lots of Australian investors were confused. We got called unhinged. We got— I got called weird, strangely. We got called woolly by investors.
Georgie Healy: How dare they?
Mike Keating: I know, I know.
Georgie Healy: What type of AI startup will fail by this time 2026?
Haziq Nordin: The things that require long horizon tasks where you're not a frontier lab and have no capability ability of training in models to do that long horizon task, 100% your smoke. Things that allow LLMs to touch a thing multiple times without a human in the loop, whether that's memory, whether that's some sort of artifact, like, I think you're done.
Georgie Healy: Hello and welcome to In the Blink of AI, your front row seat to the AI revolution. I'm Georgie Healy, and this week we are joined by Enhance Labs and two of the founders, Mike and Hazik. These guys have been hiding in plain sight. You might be familiar with them and their hot takes and honest, real AI insights. Insights on social media and why they haven't been willing to ship a product in this time of frothy, hypey AI, and why it's taken so long for them to be excited about what they're about to release. And we have a bit of an exclusive on the show today. This interview is a little bit more conversational, admittedly somewhat chaotic compared to other tech leader episodes. I think you're gonna love it, actually. Um, I learned a lot about what is actually going on with some startups that may not survive by this time next year. But don't let their sense of fu— of humor fool you. These guys are backed by Blackbird, QIC. They were in the AFR this month. Um, it doesn't take much LinkedIn scrolling to see that they've got quite the pedigree. So even though they're doing it in an unhinged way with shorts and Birkenstocks, um, in financial magazines. They are doing it their way, but I think that they're on to something huge. So don't forget where you heard about them first. Let's dive into the show.
Haziq Nordin: You're listening to a Day One FM show.
Georgie Healy: Mike, Haziq, thank you for joining in the Blink of AI. I have been wanting to have you guys on the show for months, but then you came out on the AFR and I was like, I've got to get you on before you're too cool for me. I'm going to start with the elephant in the room, which is You've had a huge pivot. Mike, what are you passionate about? What do you want to solve?
Mike Keating: This has been a big question for us ever since we started, to be very honest. Even as recently as I was in Sydney, uh, having coffee with Tris, um, one of the guys from Blackburn looks after us. And he was like, people keep asking, what are you doing? And to be honest, like, we have not had a good answer for that for a hell of a long time. And, and we've been doing things. Particularly Blackbird Arts, we've been doing lots of things. We've been very busy.
Haziq Nordin: Very busy.
Mike Keating: Very, very busy. Lots of important things, but we've been doing lots of learning, lots of experimentation. I think one thing that was like, and I had this conversation the other day that was really awesome, was we were able to raise money without anything. Like we didn't have a product, we didn't have a vision, we didn't have a TAM, we didn't have a pitch deck, we didn't have a bank account, a shareholders agreement. We literally had zero when we got the offer, which sounds awesome. And to be honest, like We thought that was pretty cool in telling people.
Georgie Healy: It is kinda cool to be fair.
Mike Keating: But it is honestly like, it's really, really hard because you've got nothing and then it's like, cool, what can we do? And like the bet effectively, like Blackbird invested in us like for the two Vs, the two like most important Vs when it comes to fundraising, it's velocity and vibes. And so Off the back of that, we just continued down that path and we just had heaps of fun and we tried a bunch of things and we learned lots of lessons around product and customer and research and particularly, and Haziq will probably talk to this later, LLMs and how nobody really genuinely understands how best to place them at the moment. And there's all these open problems that you get exposed to. Sometimes, uh, by design, sometimes by accident. You kind of stumble into these, these weird positions. So yeah, this last period for us has been trying to work out like what the hell we're doing. Um, and as of very recently, surprise, surprise, we've made a pivot. But this one feels— this is the one.
Georgie Healy: This fit. You guys have been very honest throughout too. Like when you didn't have a product, you were very honest about that. Um, I loved like, uh, socially posts, which is like all the things that you don't like that are being built and how you're trying to avoid that. And so when I've noticed your excitement about this pivot, I feel like because you've been honest throughout about what you don't like, it feels quite tangible that this is exciting to you as someone who's been watching from the sidelines.
Mike Keating: Yeah, to be honest, like, at least from my side, like, speaking from my perspective, the big, um, the honesty thing was I sat on the sidelines in a, like, a shitty bootstrap startup that struggled for so many years and all people raise money and go onto like podcasts and be in the AFI and like be like, like pretended that they knew what the hell they were doing. And a lot of them didn't. And I just hated that. And it made me feel really insecure about what we were doing and that we had something that they didn't have. And, and that's not true. Like, um, maybe in certain instances there are, but I think in a lot of instances, like, everyone was trying to work it out. So I think at least from my perspective, I've tried to be very deliberate in being like incredibly honest and open in, in the good, good things and the bad things to, for everybody like trying to have a crack that is doing it tough, that they know that like nobody knows, particularly in the current AI landscape. Like nobody has a clue what they're doing. Founders, VCs, anyone, everyone's trying to work it out on the fly. And so that's like a big part of why I've been driving that narrative as part of like our social, um, strategy, if one could call it that.
Georgie Healy: Yeah, the honesty is so refreshing, especially for us that are all working in tech and in the AI space. It is that like sex in high school, like, are you doing— I don't know if you're doing it. No, no, no, no, not me, not me. But that guy said that he— You need to listen to the First Check episode if you don't listen to Cheryl and Maxine. Can't help you there.
Mike Keating: They didn't reply to my email.
Georgie Healy: Yeah.
Mike Keating: They're in my bad books.
Georgie Healy: Now you know why.
Mike Keating: But I hear they're lovely.
Haziq Nordin: There's kind of an interesting thread with the pivots, right? Because right now, this is probably the first time in tech in a very long time that the startup community and the tech community in general is building on top of a technology that we're learning about at the same time as the scientific community. Because all the time we're getting papers that come out and say, hey, it seems like LLMs could do this, but it turns out they can't. Well, that's kind of a problem because we're not normally learning about things in that way. You're not normally learning about things from research papers that are being published on the day about the technology you're sitting on top of. And I think the unique thing right now is like, okay, well, no one really has the answers, including the scientific community. It's an open problem. And when you're in that kind of hyper uncertainty kind of space, two things are important always, right? The first thing is that you can finish things. And the second thing is that you know when to not finish a thing. And I think like the death trap that a lot of startups will enter is they'll say, hey, you know, like we were going to do this thing based on these assumptions because the scientific community back when we did it, like we didn't know whether it was going to be okay or not. Well, the scientific community said it's not going to be okay. And the answer to that is not we're going to continue. The answer to that is like, hey, like let's, let's reevaluate and work out sort of like where we stand now. And, uh, like we've been stung by that. Kind of continuously. And the reason we can survive is that we've been super open about being like, hey, here, like, we know where to stop.
Georgie Healy: Yeah.
Haziq Nordin: It's very, very important we know where to stop.
Georgie Healy: Yeah, there is kind of like this overlaying cloud of like potential doom, which we're just kind of working through and just assuming that we can just keep working and the doom is there, but that's okay. What particularly exciting or scary or important problem, Haziq, would you like to be part of the solution for?
Haziq Nordin: Oh, right. So, actually, I'll let Mike talk to sort of what we pivoted into, because we never actually touched on that. And then that falls into a bunch of like difficult technical things.
Georgie Healy: Tell me.
Mike Keating: Yeah, so I won't bore you with all the pivots because there were so many and that would take up the vast majority of this time. But there was definitely a path of learning again from a technical customer Standpoint particularly, but where we've landed is this crazy future where effectively like we can enable for every user to design their own personal internet. So there's been a lot of chatter for many years about like dynamic UIs and like the living interface. And a lot of people have said that it's kind of to an extent like the last frontier as it relates to like connecting AI with humans like that. Could potentially be a big unlock, like a bit of a bottleneck, as that's kind of where the two ends meet. And yeah, through a lot of great work by Haziq and a lot of me dropping fire emojis in the chat, we may have landed somewhere where we can enable such a thing. And—
Georgie Healy: So tell me, like, it's not just putting the microphone on and like, tell me what your envisioning it looks like when you say voice. As a startup founder, you're juggling multiple priorities from the expected, like finding product market fit, to the unexpected, like customer requests for SOC 2 or ISO 27001 certification. Achieving compliance is time-consuming, and time spent on that is time away from the needs of the business, and that's where Vanta comes in. Vanta is the all-in-one solution for startups to become compliant quickly and build a security foundation with ease. With a combination of automation, an extensive partner network, and a security marketplace containing 385+ pre-built integrations, Vanta provides the necessary tools and expertise for startups to achieve compliance seamlessly. No matter how urgent your needs are and at every phase of growth. Over 10,000 leading companies, including Cypherstash, Handle, and Indebted trust Vanta to automate compliance so they can focus on growing the needs of their business. Here's the important part. Startup listeners of the show get $1,000 off if they go to dayone.fm/automatecompliance.
Mike Keating: Well, it, in this instance, it's, it's not so much voice. So most definitely we love voice and I love banging on about voice, uh, as, as an interface. And, uh, that will, there will most definitely be elements, but like a true UI as it sits on your screen, we're talking about being able to have a personal UI for the apps that you use every day, personally, professionally. Like if you, like the other day, using the product I've redesigned, at first this was like kind of like cutesy. And then I was like, this is like really super poisonous because somebody made like a smartass comment about it. But I redesigned ChatGPT to look like Bluey. And it was like, my daughter's name's Thea, was like, hey Thea, tell me about some cool ideas that you've gotten and let's chat about it. And it had Bluey and it had like animations and, and I, that's, that's ChatGPT. That is the chatgpt.com website. That is the model that underpins it, but it looks like Bluey, it feels like Bluey. She's having a ball now. She won't stop grabbing my laptop. Um, so that's, that's like the poisonous bit. I've now like one-shotted my daughter. Um, uh, but, but like, that's like a, a kind of like a novel fun example. But then there's things like people with big gross legacy systems that they hate using, and they're like, this UI was made in the '90s. I'm going to shoot myself if I have to touch it again. And they can sit on top of it and be like, make it look like Facebook. Like I, I'm, I use Facebook every day. I know how to use Facebook. I know the design patterns that makes me feel comfortable.
Georgie Healy: Is this exactly Mike on Facebook every day?
Mike Keating: Well, no, I'm not a boomer. I skateboard. I don't know if you see my content. I'm a skateboarder. I am.
Georgie Healy: Okay. For the record, anyone that's listening in, Mike is super cool.
Mike Keating: Mid-30s. I'm in the ballpark. Early to mid-age.
Georgie Healy: Peak cool age.
Mike Keating: Typically, yes.
Georgie Healy: Mike, can I just ask you a quick thing on that?
Mike Keating: How can you pivot into like a serious question after just attacking me personally so aggressively?
Georgie Healy: No, like—
Mike Keating: What kind of show is this?
Georgie Healy: That is my one skill set. My one skill set is to keep you on your toes. Serious question.
Mike Keating: Well, I'm not wearing shoes, so I'm well positioned to do that.
Georgie Healy: I'm wearing Birks just for your benefit, actually.
Mike Keating: Oh, well done.
Georgie Healy: Yeah. So we had a guest on the show 5, 6 episodes ago who said, imagine— he's like, someone needs to build this. And maybe you guys inadvertently built this. Think of anyone that's in the older demographic and when they go on websites and it's just not user-friendly for them. Hazik is nodding. Is that even a potential area?
Haziq Nordin: Yeah, that's exactly it. So there's sort of like two unlocks, right? Like the first one is like sort of the hyper-custom web where you can go on and say, hey, you know what, like I want the interface to work like this. Like this is how this is how I want the comment thing to work. This is what I want to see on my feed, but you can kind of control that by talking to your browser. So that's kind of like, kind of like what users get, right? But like, uh, what companies get is a way to just very rapidly prototype a bunch of different experiences. So you can go in there and just say, you know what, like I have this roadmap item. Um, there were 20 takes that I could take on this. Um, let's have all 20. Um, let's hand them out to some cohorts of users. Like, um, like that sort of classic story of Sheryl Sandberg with the 50 blue shades. What if that could be like actual features and actual takes on features and actual components on features where people are like, where companies are building exactly against what people want and people are able to shape things into exactly what they want. So we end up with this sort of like hyperfluid, like living interface essentially.
Georgie Healy: You know where my mind immediately goes though is that sounds like a lot of credits. Sounds like a lot of GPU. How can you afford it, guys? How good is that Blackbird money for this?
Mike Keating: That's why we're on the podcast.
Haziq Nordin: That's why we're on this podcast. Please sign up to our Patreon.
Georgie Healy: Give us TPs!
Mike Keating: Don't use that as a thumbnail for the article, please.
Haziq Nordin: There's sort of two directions here, right? Like, there's sort of like, the sort of, like, the exciting thing for consumers is like, LLMs are becoming better and better, like, kind of like building out these interfaces at smaller and smaller scales. Like, we're starting to see like a couple of LLMs start to enter like consumer hardware. And at that point, like, kind of the sky's the limit. It's a time for these like hyper-custom sort of experiences that right now you really can't pay for. Like on the other end of that, like the companies, right? As you build out your user interface, like how do you make sure that you can experiment really rapidly, very quickly? How do you make your products pliable? And for that, like the cost today is just the, whatever the cost is for you to get a, you know, a Claude code or whatever it is to kind of operate. So yeah, like we've got Tailwind sort of pushing the customers along and then like the tailwinds have already kind of pushed businesses along. So it's really about kind of like working out where to land at that point.
Georgie Healy: Do you have a preferred provider of compute and can you share any secrets or no?
Mike Keating: Like, like, like, I'm going to—
Georgie Healy: Probably played around, right, with the different tools.
Haziq Nordin: Yeah. Like, I was—
Mike Keating: No special source. Don't answer the question, Tazeeq.
Georgie Healy: This is a trap.
Mike Keating: This is a trap. This is a trap. She's, she's, she's Google affiliated for starters, so she's afraid.
Georgie Healy: No, they don't spot hosts of the show, so they're dead to me actually.
Haziq Nordin: So, uh, like, I'm very much an AWS guy. Like, uh, I'm an AWS guy just because I'm from, like, and I was at Amazon for, like, a long time. Like, in terms of LLM providers, I think, like, the two ones that are, like, sort of captured a lot of kind of market share right now are, like, ChatGPT and, like, Claude. And they provide, like, really awesome experiences. But, like, I think, like, in the very near future, like, particularly with, like, LoRA, like, um, I think these, like, more open-source models that are very easy to kind of like fine-tune in a much more affordable way will end up sort of stealing the cake. I think like open-source is probably going to be the one that kind of wins that base layer. And they haven't won it right now, but like, I'm kind of watching it closely. I think we're like a year, two years away from these like hyper-custom, very powerful models that just sit on open-source weights.
Georgie Healy: That's the social promo. Thank you, Haziq. That's amazing. Let's dive into Hack of the Week. I want a hack from each of you that you can share with listeners that they can think about. Going forward. Why don't you— oh no, Mike, he's already started. You go first.
Mike Keating: Hi. Thank you, Georgie. Uh, uh, mine's voice. As we discussed before, I'm youthful, uh, and young people like me are very, um—
Georgie Healy: People are watching this on YouTube, Mike. You can't actually lie.
Mike Keating: Oh, the vid that's our faces will be there? Oh shit, okay. Um, yeah, that hurt me. Um, no, the big thing for me is voice, to be honest. Like, um, being a, an older head in this game, I, uh, or just generally, like, I'm really bad with computers and software and change— changing behaviors and adopting software is something I'm terribly bad at. And voice is something that, uh, I think Hazit really kind of like championed internally to the annoyance of a lot of us early because he wouldn't shut the hell up in our little Queenslander, just yapping all day back and forth, like literally like hours and hours and hours. And then we finally moved a couch into one of the rooms and kind of locked him in there. And you just hear him yabbering away. And then we all started to pick it up. And then for good or for bad, it changed the whole way that we work. Pretty much everyone at the team now, like, largely, like, exclusively talks to their computers. And, um, and so the future of offices will be interesting because, like, we've got desks and, and monitors, and now, like, I'll be down the back, um, by the pool, obviously get my tan on, uh, but I'll, like, someone will be outside, somebody at the desk, somebody in the kitchen, somebody be in a room, everyone's just talking away. Um, now, like, on mobiles, being able to run around and continue your work and do actual meaningful work, like, on the fly Um, it's crazy. It is just like the biggest, I think for me, the biggest uplift in productivity and improvement of, from an experience standpoint. It's been so cool.
Georgie Healy: Um, I love it. So people are sleeping on voice, I feel. I feel like people aren't leveraging voice as a tool for— For sure. And it's— Yeah.
Mike Keating: And it's like, particularly like talking to your laptop may feel a bit weird because you don't do it, but you talk to your phone. You, you have headphones and talk to people. So like that, that pattern shouldn't be as weird as it is. The other side of it that I don't think a lot of people appreciate is like, one, the experience for sure is great, but LLMs thrive with context. And so you, particularly if you enable like the voice-to-voice chat mode, they are really good at like eliciting like stories and anecdotes and names. And so what you're doing there is like context engineering. You're enriching the context of which the model can work on top of. Yeah. Versus in a chat interface, like everything about the UI is encouraging conciseness. So it's like a tiny little text box, typing sucks. So you just like, like one or two word answers and then the models then obviously like aggressively trying to like, uh, contribute and thus it's like pulling in a bunch of random insights and context and data to try and fill the gaps that it has. So from like an experience standpoint, voice is awesome. And from an actual output standpoint, um, the context is incredibly powerful for the model to give you meaningful personalized insights.
Georgie Healy: I actually love that hack. Are you using ChatGPT or?
Mike Keating: It depends. At the moment, unfortunately, I like Claude, to be honest, but Claude Desktop doesn't have voice enabled as yet. So I'm like, I use our old product, one of our old iterations. There's been talk about sunsetting that, and I'm being annoying and saying like, no, fuck off. I actually still really love talking way to this thing. So I'm fighting to retain that. But yeah, Claude desktop doesn't have voice. So often I'm talking into ChatGPT and then very manually copying and pasting that into Claude to then run that interface. But yeah, it's sick.
Georgie Healy: I definitely encourage it. Oh, so good.
Mike Keating: It's weird. And I think in the early days you feel a bit strange doing the whole thing and telling a story and naming people. But once you get into the flow, like you won't shut up. And that's again, that's where the model benefits from that kind of behavior.
Georgie Healy: Beautiful. Haziq, what's your hack of the week?
Haziq Nordin: Like, if what you're doing is not like making a Zapier thing or like an NNN thing, or like you're trying to like code something with an LLM, you probably don't have to worry too much about prompt engineering and you probably don't have to worry that much about context engineering. Like the way to drive an LLM very well is conversational. Like I think a lot of people are like, oh, how do I one-shot this? Like how do I come up with the perfect And the answer is you probably can't. And a really good way of reasoning about how you want to work with an LLM, if what you're doing is just doing a normal job, is if you open, if you go into one of the Strong models, one of the reasoning models, and you click on that thing that lets you see the reasoning, that's how you're supposed to interact with it. And you'll see that it's not doing some crazy one-shot thing. You're supposed to just help it take one step forward and then see where it goes and then just be conversational with it. Now you'll be able to get somewhere much better than the reasoning model can because you actually have a brain and you can think through things and you can reason in more useful ways than an LLM can. So your train of thought with the LLM is going to be very high quality. Like if you're just casually using an LLM, that's probably how you should be using it. Like this craziness on like, and I realize you can't share that on LinkedIn and you can't share that on TikTok. You can't be like, steal my amazing prompt because there isn't one, right? Like your amazing prompt is that you're using your head. Like, but, uh, but, but that is the, that like, that is the way that most people use LLMs. Yes. Oh, at least at a high level.
Georgie Healy: Because prompt is using your literal brain. I think that is such a good—
Mike Keating: It is using your brain. What do you mean? I'm just going to say hot take.
Georgie Healy: You heard it here first, guys.
Haziq Nordin: Like, you're just trying to set the LLM up to succeed, right? If you understand what it's doing is pattern matching, then you're trying to get it to kind of autocomplete in the direction that's useful. But yeah, like, if I sit down and I see sort of like, like a lot of these guys were driving a lot of the LLM adoption at Amazon. They're very fluid with it. None of them are sitting there being like, "Ah, I found this prompt from TikTok and it is the ultimate prompt." They're just talking and just thinking through it with an LLM.
Georgie Healy: Yeah, I'm so bearish on the perfect prompt as well. These people that tell me they have, like, "Where are you storing all your prompts?" I'm like, "I literally don't store prompts." That sounds to me insane. Is that insane? Do you store prompts somewhere, guys?
Haziq Nordin: Like, I mean, I used to, like, and it can be helpful, like, as a tool in enterprise when you're just kind of like, hey, everyone's scared. And then, like, they don't know what it means to think with an LLM. Then you can kind of, like, bootstrap them by being like, hey, here's, like, a sane place to start. But then people tend to get trapped in that world where they're just like, now I need another one. How do I get to do this other thing? And, like, that's not you learning how to use an LLM. Like, you're just, like, kind of pasting things around. And you can't get depth and you can't get kind of, like, specialization that way.
Georgie Healy: And that's perhaps where this conversation around people aren't using their brains anymore. They're just typing into the LLM and it's like critical thinking, actually using a brain, probably going to get better results. Those were very good hacks, guys. I can't even, I can't even lie. I want to ask you about something that you're specifically interested in. You're moving super fast. You've pivoted. You're, you're, you're changing the UI/UX interface for people to play with different versions of that to get the best result for themselves. Is the internet cooked, Hazik? The whole internet?
Haziq Nordin: Does it all— No, no, no, it's not cooked. Sometimes we all think we are. Yeah, most of the internet has always been cooked, but like—
Mike Keating: Shout out to 4chan. You're going to get doxxed.
Haziq Nordin: Internet's not cooked. Internet's not cooked. I think we're about to enter like the world of like the kind of like the hyper-custom internet. And like, why is that appealing, right? Because like, for most of the history of the internet, the people that get to have opinions on how things feel and how you use them has been the company that makes the thing and not the user, right? And why is that an issue? Well, like the company's kind of like a good place to start with that sort of stuff because they can reason about the problem deeply. But like a lot of what drives the design of a website has nothing to do with the customer, right? It has to do with the constraints of the business. Like why does Amazon look like the way it does? It's very fragmented. There's very little cohesion in the website. Well, it's because Amazon needs to move very quickly. They're a very consumer-facing company because they need to move very, very fast. It's a very competitive space. You can't create cohesion between your features. Like you have to have every team make their own decision about things and you get fragmentation, right? Like, and that's okay for someone like me. I'm like, yeah, I don't care. But like a lot of people don't want fragmentation, right? They're trying to do a very specific thing. They don't have that interface now. And like that was created for them by a company dealing with those constraints.
Georgie Healy: Yeah.
Haziq Nordin: Same with Apple. Like, why is Apple so like slow? Like, I mean, they don't do software. Well, they do do software, but they don't like, you know, that's not what they're known for. Apple tends to create these super high cohesion experiences. Well, that's why they're always late. Like, that's why they're always lagging behind. Right. So like those constraints are the company's constraints. They're not like my intent. Like I turn up to Amazon. I'm not like, I want a super low cohesion, crazy website. Like I just want to buy a thing. Like, and like, you know, I, like, I don't buy a lot of things on Amazon. So I kind of want it to be a bit simple. Like that's the interface that I want. Like, uh, no, that might not make sense for Amazon to produce, but like, it's the thing that, uh, the thing that I, like, I want for me. I think we're about to enter like that world of the internet. So the internet's not cooked. I think the internet's going to pump. I think it's—
Georgie Healy: Yeah.
Haziq Nordin: I think that the internet's going to like, it's going to become the internet that like I've gotten to use this whole time because I've been able to code and I've been able to like, you know, if I don't like stuff, I can just like kind of script things and make them go away. Like, um, I have a funny theory.
Georgie Healy: So the internet's about to enter its hot girl era. Is that the takeaway?
Haziq Nordin: Yeah.
Georgie Healy: Yeah. Yeah. Okay.
Haziq Nordin: Yeah. Like the internet's about to like hit people up as well.
Mike Keating: Can we use inclusive language, please? I don't know what a hot girl era is.
Georgie Healy: I know that you don't, Mike. What about— I'll send you like a, you know, Urban Dictionary example. Okay. But I am curious, Saziek, so you want You're on Amazon, you're like, I want to check out quickly. When I'm shopping, I'm like, I love like looking at clothes for hours. We can both have the same internet. Is that, is that correct?
Haziq Nordin: Yeah, we can, we can both have the same amazon.com, but it just kind of wraps around the thing that we want, uh, that is sort of reflective of the intent that we want. Um, I think it's a better—
Georgie Healy: Does that mean ads though? Like if I paid for an ad to go on a website, how does that play into all of this?
Haziq Nordin: So I've gone on a bit of an evolution. Like, so like with ads, like with data collection, with kind of both of them, right? Like our first iteration would mean that like, yeah, you can control a lot of that. But like, I don't think like with ads and with data collection, it's a question of like, all of them are bad, all of them are good. I think it's more of a question of like, how do you make it a fair exchange? Like I'm okay with YouTube showing me ads because I don't want to pay for YouTube. I'm actually okay with Facebook gathering a little bit of data because I don't want to pay for Facebook, right? Like, um, the thing that I'm not okay with is people trying to get freebies off me where it's not a free exchange where I've like, oh, but like where I'm just kind of like, hey, I kind of feel like I already did a thing for you just then and you're trying to, I feel like you're trying to get a free lunch here. Like that's kind of where it feels yucky.
Georgie Healy: Yes.
Haziq Nordin: Um, so I guess the short, the short answer is like, uh, like our first iteration gives you quite a lot of control. Um, but I think the future of like that kind of space is really going to be a future around how do you make it How do you make it a reasonable exchange when like the consumer is in control? I think that looks like data gathering that's like a lot more reasonable and sensible and ads that are like a lot more ergonomic and friendly to the user. Like these kind of super hostile patterns of smashing you with like ads right in the face. I think, um, I think that can go away in a customizable internet. I think like that doesn't mean ads, no more ads. Like what that does mean is, uh, if you want to have an ad and succeed at that ad, like, well, you have to think about it more. Yeah. Like, you have to make sure it's actually reasonable.
Georgie Healy: Isn't it funny? Like, I— we have, um, just stopped watching free-to-air, and so an ad on free-to-air, if I want to watch the football, which is very rare, but Queenslander, um, I'm very like, oh, an ad! I'm not used to seeing an ad on the TV, and it feels very confronting. Whereas I'm the same as you, I'm not paying for YouTube Premium, very happy to like allow those ads in that context. It doesn't feel like an unfair exchange, to your point.
Haziq Nordin: Yep.
Georgie Healy: Mike, hi. I wanna talk a little bit about you guys and building in general again, because you kind of mentioned it before. To anyone that isn't familiar with you guys, you do have a lot of hype and building an AI startup right now, there's AI startup and then there's AI startup. What gets cut through? Why do you guys get cut through? What is fake news that people are getting confused about when it comes to building an AI startup? He's buffering really badly in his brain right now.
Mike Keating: I think in regards to like a product in market or just generally as a startup, is there somewhere particular you want me to take that?
Georgie Healy: I actually think in terms of building something that isn't— I run a podcast, right? And so nearly a year has passed since I first started the show. A lot of the episodes from the beginning of the show have not survived, and they may have got funding, they may have not got funding. I want to know two things. One is how, how do we know whether something is being built now will survive in a year? What do you think won't survive in a year? And when it comes to funding, who's getting funded that deserve it?
Mike Keating: Um, yeah, so maybe for the funding standpoint, um, I think we've had— this has been a funny conversation, as equal after this. So like, particularly when words started to come out that we'd raise money under— in the, like, the manner which we did, under the terms of which we did, um, a lot of people were very confused, some annoyed, some wanted to know like what the secret sauce was. And somewhat naively, we started to try and be helpful and provide advice around how to run a process. And we may have been tapped on the shoulder and said like, yeah, maybe don't, maybe don't talk about that stuff. I don't know if it's particularly helpful. Because it wasn't. I think And again, I feel I can say this because I spent 6 years trying to raise money and run a startup with limited success. We somehow managed to create this like hectic environment of like great cohesion as a team and interpersonal relationships. Like we've got long histories together in like this weird way where we've all known each other for almost 10 years, but we hadn't spent huge amounts of time together in the same room. But with that, there's like a lived history, which is really cool. So we've got really, yeah, really awesome, like genuine mateship and respect for each other. We all went in weird directions and we have this kind of nice mix of like Jesse design, Hazik engineering, and me, I don't know, doing whatever I do.
Georgie Healy: Opsie.
Haziq Nordin: Vibes.
Georgie Healy: Yeah, just personality higher.
Mike Keating: Yeah, exactly that. And then skill sets. So, sorry. And then, yeah, like companies. So like big tech, we had big sexy logos, Google, Amazon, and then I'd done the startup thing. So all that I think was really interesting, particularly in the Australian ecosystem. Not many people saw that type of group come together with history and complementary skills and complementary backgrounds that had some sex appeal. Um, and then even coming into the process was chaos. Like, we were unhinged. We spoke quickly. We laughed. We pivoted. We high velocity. Um, and honestly, like, lots of Australian investors were confused. Uh, we got called unhinged. We got called— I got called weird, strangely. Um, we got called woolly by investors.
Georgie Healy: How dare they?
Haziq Nordin: I know.
Mike Keating: I, I know. Um, so a lot of people were like, there's something here that is interesting, but like, uh, but there was no ability for them to pattern match because we were a pattern breaker, um, play. And so a lot of them were around and had been like friendly and, and what have you, but nobody was kind of in a position where they were willing to make the bet. Um, and then the great Michael Tolo came along and, and he most definitely was aware of the chaos. And I had known him previously, so I had some fun conversations with him where he's like, there's something awesome here. But he's like, you guys are nuts. He's like, this is like, you need to help me shape this up in a way that I can sell this thing because like, as of right now, this thing is like, this is intense.
Georgie Healy: And I've been very unfair to you calling you a personality higher. I think anyone that knows you and sees you knows that yes, high velocity, yes, maybe weird, but you know you're doing, like you do know what you're doing. That I saw your LinkedIn posts of how you hunted your team down. That's not an accident, Mike. Like you knew what you wanted.
Haziq Nordin: Yeah, I'm just a crafty, crafty man.
Georgie Healy: Yeah, it's very powerful. I'm a trickster. Yeah.
Mike Keating: I'm a trickster.
Georgie Healy: What did you want in your team? Like you mentioned skills-wise, engineering, design, you've got the commercial acumen, but was there any special sauce personality-wise where you're like, if you want to be part of this, this is the vibe.
Mike Keating: We, we just like having fun. This is something that I think I had, um, uh, I'd thought about, probably not actively in labels, but like Haziek very early on, the first like video call Haziek and I had was fucking chaos. I just like come off the back of a knee surgery. Um, it was like a day after, I was like high on pain medication I'd really been joking with him that I was catfishing him and I didn't exist. So I joined the call with like a hat on with my face covered, and then he joined the call with his face covered, and then we spoke in weird voices. So like from the jump, the tone was set that we were going to be weird and experimental and— But like underpinning all of that is like a bunch of very capable, very hardworking, very ambitious people. But to this day, like, the people that come around, I think, again, can be a little bit confused because it's like laughter. It's like there's lots of laughter, lots of ridiculousness, yelling, music. Um, it's, it's chaos, this environment.
Georgie Healy: I've been following you guys for a year, and yes, I'm seeing the laughter, I'm seeing the sarcasm, but also, Haziq, you're currently an SF, and Mike, you're about to be an SF. Like, you guys I think, yeah, you can't really put you guys in a box, right?
Haziq Nordin: Yeah, there's a, like, a lot of that is sort of like by design. I think like the two rarest commodities in engineering are like playfulness and enthusiasm. And in order for you to like build creative things and be nimble and sort of like be willing to move as like a space pushes you around, you really do need those two things. And like there's no hack to getting those two things. Like you just need to be like that, right? Like the most creative engineers I've I've ever met, like the ones that have like dozens of patents, are just that, like they're just very playful. They're just very curious. They're very enthusiastic. And like, that's like a precious resource that you guard. Like, that's not like a, oh, it's like a weird thing that is attached to this other thing. Like, like, like, like that's the source of what the innovation is.
Georgie Healy: You can't send someone off on training to get more playful, right?
Haziq Nordin: Yeah. Yeah. You can't mandate funness and curiosity and creativity. Like you have to just kind of like foster it and what it looks like is playfulness.
Mike Keating: And then maybe to answer the back end of that question, Rhi, from a product standpoint, I think something I would say the Royal We have done well, but the credit sits with Zeke here, is being famously, he says, it's so aggressive when he says it, he used to say it in meetings with VCs early and he'd be like, you have to, dead pan face, you have to be ready to shoot your baby in the face. And it would be like the tone, the tone of the meeting would just be like everyone was like, yeah, I appreciate what you're saying. You still do that. It was like a hard to recover from. There's a lot of millennials that have just had babies too. I know.
Haziq Nordin: Triggered. I stopped saying that, but that's what it is, right? Like, that's really what it is. In order for you to have a willingness to be very brave with your ideas, you need to be in a playful state. Otherwise, you start getting precious. And when you get precious, you waste time and you can't see clearly. And you stop being a builder that can move in interesting ways. It's very important that you can be playful with the problem domain.
Mike Keating: And particularly, you've got VC, you've got runway, you've got employees, you've got investors, you've got the public image and it's like, and this is kind of where I was heading with a few times of landing in areas that were pretty sensible and we were on the way towards proving something out. And it's like, maybe we should just grab this and just run with this because it's like, it's not bad. Maybe we don't love it. Maybe there's a bunch of like risk if we pursue this. And Haziq, again, to his credit, I say this now, not at the time, I was probably more pissed off at the time. I was like, mate, let's just fucking build the thing. Let's ship the thing and we can learn. And then we can move if we have to. But he's been very firm and steadfast in his beliefs of like, no, like this, this is a process and we can't just do it because we're doing it. Um, we need to get out now. And I think a lot of the startups that you may have seen that haven't made it, or even to this day that are, that may look like on paper, like there's a path ahead of them. I think, um, the tide will go out and there'll be a lot of tough periods. Ahead with people that, because they built a narrative, right? Like they raised their seed off it and then they raised their A off it and their B. And so then it, as the tech has matured and maybe it's flattened or kind of wobbled, then you're already down the road and it's very hard to be like, hey, yeah, so—
Georgie Healy: Pivoting at Series D, not so easy.
Mike Keating: Yeah, we're not going to get there. When the product promise that we've sold you and to the market and to our customers won't materialize. So this process early, like you do need to be like violently pivot if there is not, if there's too much risk down the road. And yeah, even that pathway, one last thing that I think you've been really good at is like you can run ahead a little bit and kind of map and as you learn around where those kind of the elements of risk exist. Yeah. Risk, but like, yeah, there's an amount of forward thinking that needs to happen, but you obviously then you can't just sit on your hands and plan the business out until IPO. There's like a nice balance there that you need to find as part of that journey.
Haziq Nordin: Yeah, like the kinds of risks that I'm like, okay, everything has risk, but like the kinds of risk that I think are like very dangerous with LLMs is where you need an open problem inside. To be solved for that problem to go away. And there's a lot of those sitting inside the LLM space. Like, I think this is kind of the dirty secret of a lot of, like, AI startups. They don't— like, people get very edgy when you bring this up and you're, like, hanging out with, like, a bunch of founders.
Georgie Healy: As a podcaster, I'm like, when do they get edgy and how? What?
Haziq Nordin: There are a lot of open problems that sit inside the LLM space, and a lot of them cut very deeply into whether or not that product will be able to meet their product promise. And like where I've been very firm is like we do not promise people things that we cannot deliver them. And we do not promise people things where someone else needs to win a Nobel Prize in order for us to deliver the thing. Like we promise people things that we know we can deliver them. Now that might look like a risky path. That might look like, you know, there's always going to be a risk. But what that risk can't look like is like Yan Li Kun wins a Nobel Prize. That risk has to be something that we can manage.
Georgie Healy: All right, tell me what you're talking around.
Haziq Nordin: What kind of AI product? Okay, like the big one.
Georgie Healy: Needs a Nobel Prize to actually pull that off.
Haziq Nordin: Oh yeah, this is just two. There were two big ones.
Mike Keating: This is a big clippable moment, Azik, so make sure it's big.
Haziq Nordin: Done. Let me fix my hair. The first one is the super long horizon tasks. So we keep seeing like Sam Altman and the rest of them come out with these mega long horizon tasks. Well, they're training against those tasks. Tasks. So like, these are models that are specialized to do that particular thing. And like, how generalizable that is, is actually quite narrow. So just because you can, you know, build Slack for 30 hours doesn't mean you can go operate like, um, an accounting business for 30 hours. It's not transferable in that way. And there are a bunch of products that are like, hey, we can replace these employees. Well, what you're talking about is a long horizon task. What you're talking about is a thing that runs for weeks and months and like, forget weeks and months. Like, what are you talking about is a thing that runs for hours at least. Like, that is not something that we have solved generally. Like, that's not a thing that we can just do. Like, that's a thing that we'd like, you have to, you have to train a model against that. What you need to be is a frontier lab to do that. Uh, so that's a big one. Like, um, and we've already kind of seen all those startups kind of like rise and fall, like all those like, ah, this is like an AI fully autonomous employee that will do all the things.
Georgie Healy: Um, So you're saying horizon task is in, like, a horizontal—
Haziq Nordin: Oh, no, no. A long horizon task is a thing where the LLM has to take lots and lots and lots of steps over, like, a very long period of time. So, like, run a vending machine business is a good one. That's a pretty simple task. We could all just go and do one. But, like, I mean, that's not fair. I don't know anything about vending machines. I don't know how complicated it is, but presumably we could get there. That takes a lot of steps. I have to reason about what I just did and what just happened next and what do I need to do next. Doing that over a long period of time is very hard. And I think all the startups that are promising that they can magically do that with prompting, no, you can't. The second one is memory. Memory is a very, very, very, very, very dangerous animal. And the reason it's a dangerous animal is that it's very easy to convince yourself that you're close and it's Very easy to convince PCs that you're close, but memory is an open problem. I know we've all used ChatGPT memory and we're like, "Ah, that has memory." That is very tricky to find.
Georgie Healy: In terms of the context window, right? Or bigger than that?
Haziq Nordin: If it remembers about you, so when it remembers, "Oh, Georgie likes chicken and this is what she likes for dinner and she has kids," and so on. That kind of memory inside ChatGPT and inside Claude is doing a very tightly scoped thing. Like, it's not like it can remember stuff and do things. It tends to remember stuff that doesn't change very quickly. Where you get in trouble is when things change very quickly. So for example, like a memory of like, you know, is Georgie, like what gender is Georgie? And like, how many kids does she have? Well, like that's a lot easier to manage. Memory on like, you know, who are the stakeholders of this project? What is the most important thing in this business right now? Like that kind of memory that changes, it changes all the time. It changes day by day. There are all these different conflicting facts that you have to resolve, that a human can resolve. Like I can sit down and be like, oh, I think one of these is true, but that's because I have a relationship with the truth. I'm like in the real world and I can like reason in a way that an LLM can't. So those are the two big dangerous ones, like startups that are like sitting on top of this like memory thing. And it's the most tempting one, like so many people, and then everyone gets like a little bit weird when this comes up because they know it too. Like, everyone kind of like knows it, but it's a bit like, like, memory is a dangerous thing. Uh, long horizon tasks are a very, very, very dangerous thing.
Georgie Healy: Andrei Karpathy said that, uh, it reminds him of the movie Memento. Guy Pearce, Aussie actor. Have you guys seen Memento? He basically has tattoos all over his body because every morning he wakes up with amnesia. And the tattoos remind him who's betrayed him, you know, things like that. And he's trying to solve this big thing. And I won't ruin the movie for you guys, but it has been out for about 30 years, so it's kind of on you. And it feels like that when you're working with an LLM where it's like, how do you have amnesia again? I literally gave you an update on this like 2 days ago.
Haziq Nordin: Yeah, yeah. Like, that's a hard problem. Like, so ChatGPT has done it a little bit. Bit, so it remembers a little bit of stuff about you. For it to go very deep, for these highly volatile memories, that's very much an open problem. And if that's what your startup requires, if it's in a very dynamic space that changes all the time, the person that solves that, I would say that that problem is one of the precursor steps to having something like an AGI, right? And if you need to solve that to solve your thing, then likely just solving that is more than the term of your entire business. It probably What is it like?
Mike Keating: The example that I like to give as it relates to this and why in my mind memory broadly can't be solved is like the truth that I have in this conversation is the truth as I represent it. The truth that Hazik and I have in a conversation in Slack is the truth. With customers, with VCs, they're all versions of the truth, but only I know what the truth is. And then if I run some search query to like source the truth for me, or that then influences the outcome of an output. Yeah, it could draw from like 5, 6, 7 versions of the truth and only I know. And if that happens a couple of times and it's the wrong version, I immediately distrust the system. I'm like, no, you don't know, you're stupid. It's— I think it's impossible in that regard.
Georgie Healy: Memory and like horizon.
Haziq Nordin: Like long horizon tasks.
Georgie Healy: Long horizon tasks.
Haziq Nordin: Yeah, these sort of multi-step things where an LLM has to keep continuously thinking through the same problem over some period of time. And memory for things that change quickly, like very high-density memory products. I haven't seen a single one succeed. There were things like ZEP, but again, the lifetime of those things, it's meant to be short.
Georgie Healy: What's that?
Haziq Nordin: It's not meant to be something new. It's pretty cool. I actually really like ZEP. It's like this It uses a graph representation of memory and it attempts to reconcile conflicting facts, but it rots very quickly. That's the general problem with these memory systems is they get one error, it uses that error to make a new memory. Now there's two errors and then now there's four and then now there's 20. And then after a week it's just rotten. The whole thing is just slime. Creating that stability and making sure that it can refresh itself is hard. It requires a relationship with truth.
Georgie Healy: You know, this is why I love talking to you both, genuinely, because I go on Twitter, right? And this is where I'm trying to get these kinds of nuggets of like, what won't survive? What is hype? What requires a Nobel Prize to even begin to think about solving it? Where do you guys get your info? Is Twitter a dumpster fire? Can you actually find insight? How do you guys get your latest thinking?
Haziq Nordin: I use Twitter as an ingress into papers. Twitter on its own is not very good because people are insane and you're incentivized to make crazy claims that upset people or are very inflammatory. For the algorithm. And when you're building, particularly when you're trying to work out what kind of product is feasible to build, it's the details that really matter. So Twitter's a great source for papers, but you kind of want to keep drilling. You want to read the actual paper. They're more like a table of contents than a Is Attentional You Need Your Favorite Paper, Haziq?
Georgie Healy: Because I've heard that that's the most cited AI paper of all time. So many citations.
Haziq Nordin: I left research around the time that paper came out and I remember reading it and being like, mate, this thing, okay, at the time, no one really knew how to, that you could get to where you could get today by scaling these things. But also at the time, it was peak CNN, RNN kind of world. And we were just getting smashed by novel architectures all the time. So it was so easy to just see one and be like, yeah, that's kind of cool. It's kind of like, I can see that might be more powerful than an RNN, whatever. I see these every day. And then now these, so many years later, when all the other ones have fallen away and that's been king, I've been like, wow. I remember that day. I remember when that thing came out and it was on Karpathy's blog thing, like some feed. Yes. And then it was really cool. And then I was like, oh, okay, whatever. And then now I'm like, I I actually remember that day. Like, I remember like where I was sitting when I read that paper.
Georgie Healy: I feel like that all the time, Haziq, where I'm like, was that just a major news thing that happened and I don't know amongst all the noise? Or I'll meet someone amazing in AI and I'm like, are they gonna solve AGI? Did I just meet the person that may solve it? And who's to know? Who's to know? Will you solve AGI, Haziq? No.
Mike Keating: You can ask me next.
Georgie Healy: All right, Mike. Hi. Sorry, I forgot you were there for a moment. Hi. No, but seriously, I do have a question for you because we talked about getting information from Twitter. What about sharing on socials? You guys do share. You are out there. Should startups do this? Is there an ROI to doing this?
Mike Keating: Yeah. When it comes to sharing, particularly as a startup, when you like— because you have nothing really to talk about, right? Like, you haven't got wind, you haven't got customers, you probably don't have much revenue. Um, that's all you can share is your story. And people love an underdog, particularly in Australia. Um, they love that story, they love connecting with personalities. Um, in terms of ROI, uh, again, it's too hard today because we haven't had any, uh, product to push people towards, but But we get so many inbounds of people just like saying, that was funny. Everyone thought we got kicked out of the house. That was a pretty funny period when we did that house inspection video. Whenever you go into like meetings or like you and I, I think like that was when we first got connected via the great Millie, that a lot of the initial conversation was built on top of that. So, we both kind of like to an extent felt like we knew each other. So, there was common ground. And so, we could like race to building a friendship. Investor meetings, most people have kind of seen the things and they know the house and it's just like a leveler. I think it's like just like a really cheap, fun, easy thing to do. It doesn't have to be fancy. In fact, like the rawer the better, the more representative of the story the better. 100% do it. And it's fun. It's cool to share.
Georgie Healy: It is.
Mike Keating: Yeah, it's great.
Georgie Healy: And Mike, you've got quite a distinct personality out there. You've, you've almost got like a brand, right? Whether you've intentionally done it or not, you're in the garden, you got your Birkenstocks on. I think of Mike and I think of Enhance Labs and I think of Aussie outdoor, honest hot takes, but moving quickly. Do you think— serious question— do you think having a brand in and of itself will be a moat?
Mike Keating: Yeah, I'm sure. I think maybe to attach to one thing that you just said. So we spoke to a few, when we were trying to get this PR push and getting the AFR, we spoke to a few PR shops that said that we couldn't get covered because we were too early and the amount wasn't enough and didn't have meaningful traction and a bunch of things. So they're like, no, we could aim for these other publications. And, um, of course that like fired me up and I was like, fuck that. We can definitely get in there. We just need to work out how we do it. And a lot of things that I think people just kind of like, just think, uh, we just stumble across. We are quite thoughtful about how we kind of curate like the details, like small little details. And, um, Like for that article, it was like, uh, instead of just saying like, hey, we're this startup and we raise money and we're trying to do to do this thing. It was like, what is something that's very topical at the moment? What is something that the journalists would care deeply about? And so that's where the AI slop angle came from, because journalism— or at least there's been a lot of like threat of journalism, journalists being really impacted and layoffs within the media industry. And, and they are the truth, they represent the truth, and they are very thoughtful and, uh, in, in creating this content. And slop is the antithesis to that. So it's like, that's like a really great thing that we think they would attach to and their audience would also buy into. And then my ridiculous dress with the Birks and the socks and the shorts was another like pattern breaker where, um, again, that almost broke the article, I think, because it wasn't— they weren't stoked initially. But as part of like a broader narrative of like, what, like, what the fuck, why are these people doing this? Why are they dressed like that? Why are they— if you can kind of curate that a little bit, you— it becomes a talking point versus like, oh, surprise, oh, another company raised money for AI.
Georgie Healy: Yeah. But also the AFR is just Suits and RM Williams. Like, how refreshing to see your hideous white legs with some Birkins.
Mike Keating: I've got great cars. I have fucking great cars.
Georgie Healy: How dare you. The Cubs were on fire, but like, the Cavs— That's the clip. It really— that's the clip. It really did get cut through. Like, my husband works in finance in the city, and I got literally within 20 minutes of the article going live on the AFR, him sending it to me saying, you got to get these guys on your show. And I was like, first of all, they're my friends. And now I will publicly say that.
Mike Keating: Isn't that funny?
Georgie Healy: It was just something ridiculous. Like something— So much cut through.
Mike Keating: Yeah. For something ridiculous. My father-in-law was a little bit torn because I'd been like this useless son-in-law for how many years working like a broke boy working in startups to then finally being in the AFI and he could show his friends. And then I'm wearing fucking shorts. I think he was like, like, damn it, you couldn't have just worn pants.
Georgie Healy: Actually, that's the episode title, No Broke Boys. Amazing. We are at the hot take section. I've got a question for each of you. Are you ready?
Mike Keating: Ready.
Georgie Healy: Okay, Haziq, what type of AI startup will fail by this time 2026? You've talked about a few.
Mike Keating: Yeah.
Georgie Healy: Things around this.
Haziq Nordin: Yeah, yeah. Things that require long horizon tasks where you're not a frontier lab and have no capability of training your models to do that frontier task, to do that long horizon task, 100%, you're smoked. Things that require LLMs, things that allow LLMs to touch a thing multiple times without a human in the loop, whether that's memory, whether that's some sort of artifact, I think you're done. I don't think that you can keep that thing stable. I haven't seen anyone be able to keep the thing stable.
Georgie Healy: Um, you talked, uh, previously about a concept called pinks.
Haziq Nordin: Oh, pinks. Yeah, yeah.
Mike Keating: Oh gosh, okay, so there is a 30-minute conversation.
Haziq Nordin: The, the reason we call it pinks is that we put a pink sticker on these things, uh, and, and what is a pink? Um, so like, uh, because we, we got smoked by a couple of pinks early on, and then I was like, okay, no more, like, these Pinks, we have to watch them. The concept of this pink is when you're trying to stretch an LLM to do something that it hasn't been directly trained to do, that it isn't routinely used for. And the reason why that's so dangerous is because a lot of the time it will seem like it can do it. It'll seem super, super close to being able to do it. You could even do a VC demo and show them the thing and it'll probably do it, right? The danger is that you don't know when the LLM will tear, and the performance drop-off becomes very steep, you don't know how far you can stretch it, right? Like, and because it's not your model, there's no guarantee that it'll keep stretching in that way forever. And things like that look like, oh, we're going to get an LLM to reason about this, like what was the one that we had? Oh, search. Well, hey, we wanted to be able to find like a reason about like kind of like the current state of your business. Because it can do search across codebases. That's how good— well, it's been trained to do search across codebases. Search across codebases is a long horizon. You're taking quite a lot of steps. Even though it both looks like text, it looks very similar, and we search through them in a similar way, an LLM isn't searching the way a human does. Pinks are these very subtle differences in what the LLM is doing that suddenly cause it, cause its performance to decay. People that are stretching LLMs like that, you might get lucky where it kind of doesn't tear, but like, for most people it will tear. So it's not a good idea. Do not stretch LLMs in a way that they weren't trained for. Like, it's a very jagged performance curve and you'll get cut on that thing.
Georgie Healy: Amazing. Mike, your hot take. I want to know who you're dressing as for Halloween this year. I'm going as Elizabeth Holmes because I've already been accused used to being Elizabeth Holmes in the past. I've never had a medtech startup, but like, I'm giving Elizabeth Holmes energy and I love that for me. Yeah, yeah, yeah, yeah, yeah.
Mike Keating: What era of Elizabeth Holmes? Like today?
Georgie Healy: Peak, peak, I can do all your lab tests with a drop of blood era. That's, that's my girl.
Mike Keating: Awesome. Um, with, uh, with the tech lens, I— my mates love to tease me because I'm from Brisbane. Everyone, all my mates are like property and consulting, um, and so they don't understand what I do, like a lot of people. Uh, so I get called like Wish Mike Cannon-Brookes a lot, um, because tech and—
Georgie Healy: Like Temu, Temu version.
Mike Keating: And because I wear hats in inappropriate occasions, uh, so if I was That would probably be the play.
Georgie Healy: Would you buy an island once you become a billionaire or?
Mike Keating: Yeah. In Vanuatu. I've already got it. Lilipa.
Georgie Healy: What do you mean you've already got it?
Mike Keating: I've already picked it.
Georgie Healy: Oh, you picked it just to, for the ones listening, doesn't say that. I'm gonna compete with it. All right guys. Honestly, I can kind of see the Blackbird method, cause I got a big ROI on this, this call. You guys were so generous with what you shared, so fun to chat to. Never a doubt in my mind. But if you want Enhanced Labs to be remembered for anything, Mike, what would it be?
Mike Keating: Uh, but—
Georgie Healy: He's buffering again.
Mike Keating: No, I just— where my mind goes to, and I thought about this as well as part of that AFR, like, thing is like, what's awesome is when companies are on the up and the AFR I feel, and publications similar, encourage founders to do all these like visionary, like, like, rap squats and like all these like really, like, you're the god pose. And then when they crash and burn, they then use those same photos to mock them. So we don't want to be that.
Georgie Healy: Front cover of Forbes and it's, you know.
Mike Keating: The guy in shorts. Who would have thunk it? The guy wearing shorts in the AFR startup died.
Georgie Healy: Could have called it.
Mike Keating: No, but for us, I think like, I think we've done, to Tybo and like all we've spoken to, I think we've done a really good job of like being ourselves. And we do laugh about like if we do win and we have an outcome and then we get called in to speak on panels at startup events and stuff like that. And we just like, just tell the truth and nothing but the truth of all of the stories and the chaos that ensued. And, and, um, that'd be sick of like those like unhinged dudes. Like, I don't know how the hell they managed to do what they did, but it looks like they had a ball and they delivered shareholder value. Um, that's, that's, uh, how I'd love for us to be remembered.
Georgie Healy: That's beautiful. It did bring a tear to my eye. Um, thank you so much for joining In the Blink of AI. Where do Guys, where does everyone find you? Where do they find Enhance Labs? Where do they follow you? Can you each answer that for me?
Mike Keating: You go first, Izzy.
Haziq Nordin: LinkedIn is the best place to find us. We live and breathe on LinkedIn. And otherwise, enhancelabs.ai.
Mike Keating: That's us.
Georgie Healy: Thank you very much. You guys are the best. Thank you so much for being on the show.
Haziq Nordin: Thanks for having us.
Georgie Healy: Thank you for listening. And thank you to this week's reviewers. I wanted to give a huge shout out and thank you, and I love you to Ash McGedigan, Tristan Cameron, and Kasia Parshib. Guys, you not only made my week, but, uh, this is brilliant because it actually helps me get better and better guests. We're already the Australian number one AI podcast, but if we want to get over overseas people to know about us and get the best guests on the show. This is the exact way of doing so. I couldn't be more grateful. Slide into my DMs if you've reviewed the show recently or ages ago. I would love to give you a shout out. Have the best week, guys, and I will see you in 2 weeks. Bye!
