The End of Cheating: How AI Is Redefining Success

00:00:07 Mark Smith
Welcome to AI Unfiltered, the show that cuts through the hype and brings you the authentic side of artificial intelligence. I'm your host, Mark Smith, and in each episode, I sit down one-on-one with AI innovators and industry leaders from around the world. Together, we explore real-world AI applications, share practical insights, and discuss how businesses are implementing responsible, ethical, and trustworthy AI. Let's dive into the conversation and see how AI can transform your business today. Today we're heading to New York City, where 21-year-old founder is flipping the script on how we learn, work, and define cheating in the age of AI. Full links are in the show notes for the episode. Please welcome Roy, the CEO and founder of Cluely.

00:00:58 Roy Lee
Hey, thanks for having me. Full transparency, we moved to San Francisco a couple months back. So we're here in the capital of the world now.

00:01:06 Mark Smith
Nice, I like it. Before we begin, Cluely got my attention before, we set up this call some months ago. You hit headlines, you've gone viral. Tell us though, for the folks that don't know, what is Cluely, what's it about? So they can get an understanding of how they potentially could use it.

00:01:28 Roy Lee
Cluely is, it's like a glass window that appears on your computer at any point in your life. It always gives you the exact information that AI suspects that you'll need. In a meeting, if someone says, hey, I'm from Duluth, Georgia, it will surface a definition of what Duluth, Georgia is. It will surface various definitions. When you're, it also like has full context of your screen, your audio. So it can always, it always knows exactly what you're doing at every point. So if you're on DoorDash, you can ask it about the calories of the different food items that it sees, can draft emails for you, can do pretty much anything that a always on Desktop Assistant could.

00:02:06 Mark Smith
So What made clearly such a big deal? How did you, particularly from a marketing perspective, how did it kind of hit the world and how did media pick it up when you first hit the market with it?

00:02:20 Roy Lee
Hulu started as a tool called Interview Coder, which was a tool that I built when I was in college to help you cheat on interviews for software engineering roles. They tend to ask these stupid riddle-type questions, so I built this tool that functioned as a pane of glass that showed up over all your applications, was invisible to screen share, and fed you answers to riddles using AI. I record myself using it to get an offer and to get a couple job offers and all of a sudden the people see like Amazon sees sees sees my sees my recording that I that I cheated with my tool to get an offer there. And they tell Columbia, and then Columbia and Amazon are both really on my *** Columbia ends up kicking me out of school. Amazon ends up blacklisting me from pretty much all the big tech. And at one point, I'm tweeting about everything as I go. And I go viral as the kid who got kicked out of school for building the cheating tool. And very early on, though, I realized interview coder as a concept should be for much more than just interviews. It should be for everything. So that's when I launched Cluely, the cheat on everything app.

00:03:16 Mark Smith
So cheating is obviously a word that grabs people's attention. And, to show my age when I was at school, if you used a scientific calculator, it was called, in the exam, you were considered cheating and you were banned from doing it. Now, everybody uses it. That's standard. If you don't know how to use it, like, what are you doing? And when we're in the AI world, right, anything that allows you to get things done smarter, quicker, better, be more informed, be more set up to act in the right way, I wouldn't call it cheating. And how do you now tackle this? Because it obviously got you the visibility. How have you then turned that into a product that people are buying? And what are the use cases you're seeing now?

00:04:06 Roy Lee
Yeah, I mean, we see a lot of professional use cases with meetings. I mean, in every meeting, whether it's a team stand-up, whether it's a one-on-one with your boss, whether it's a sales call, customer support call, like it is always helpful to have information from the web condensed down into one or two sentences that are super useful for you. So, we're seeing largely adoption for people like professionals who are using this for meetings. And yeah, I guess the We are pretty strong believers that the whole idea of cheating is just going to be changed dramatically in the modern world. I mean, like, always, whenever you think about cheating in the past, it has always been like someone got something or someone created something or was capable of something that they should not have been capable of. And it feels like unfair to us subconsciously as humans. But in reality, if you just rewrite what it means, like the amount of effort it takes to achieve something, which is what AI will allow us to do, then all of a sudden, like, what is unfair will not be unfair. It'll just be normal in tomorrow's day.

00:05:00 Mark Smith
Yeah. Exactly, Any backlash from, beyond getting kicked out of university and being blacklisted by Amazon, what else came out of it?

00:05:12 Roy Lee
I think in the world of Silicon Valley, I'm probably like, I might be one of the most polarizing figures. Everyone has like an opinion and it's either strongly positive or strongly negative. There's very few people who just think neutrally about us. And I think it's like a direct result of what I intended on doing with all the controversy and viral stunts that I pulled. Yeah, like we aim to be polarizing and we are indeed pretty polarizing.

00:05:36 Mark Smith 
So from a viral perspective, had you had successes of going viral prior to that kind of like led you to go, you know what, I know I'm going to carry out these interviews and that's going to give me some great footage and proof points.

00:05:49 Roy Lee
Honestly, no. I think something that we'll see. With a lot of Gen. Z is that there might not be a lot of, there are a lot of potential content creators who have very strong viral sense, but just have not gone viral before. But in reality, like when you scroll on algorithms and short form feeds and you just consume content for six, 7, 8 hours a day. you would sort of develop a taste for it. And the only thing that's stopping more of Gen. Z from creating content is like the social inhibition of not wanting to put a camera in front of their face and make content. But in reality, I reckon you will see like, you will see this social inhibition start to go down in the future and you'll see a lot more Gen. Z content creators. And many of them will have the same viral sense that I do.

00:06:30 Mark Smith
Has the backlash stopped you from getting funding?

00:06:34 Roy Lee
Not at all. I think the benefit of being polarizing is that some. People think strongly of you, meaning some people really love you and some people really hate you. And when you're fundraising, you'll need one person to really like you. So it's truly like it's been a blessing.

00:06:48 Mark Smith
The move then from New York to San Francisco, obviously very deliberate because of Silicon Valley. Has it changed things?

00:06:56 Roy Lee
Yeah, of course. I mean, there's like a whole religion of AI in San Francisco. It's like a cult over here. This is truly the tech capital of the world. And you can't be a part of the cult if you're not where the rest of the cult is.

00:07:07 Mark Smith
Yeah, valid, valid. Tell me about how you think about UI. Because I think that all the paradigms we've had in the past of menu items and certain layouts, whether it be tabbed across the top or left nav, that type of thing. I think UI, particularly in an AI world, becomes totally reinventable. How are you thinking about it?

00:07:30 Roy Lee
Yeah, I think there's a lot of people that say this, but there's very few people. Who are actually working towards something different. I see every single new AI tool that comes out is just another variant of the sidebar chat window that is the curse. I feel like Everybody's really saying that UI is going to be changed, but I think nobody's really taking a meaningful stab, except us. I think truly the idea of this translucent glass overlay that sort of dynamically generates output based on what you need, and the UI is essentially obfuscated, like the AI is undetectable and invisible and ambient until you need it, and then when you do need it, then it will give you like a very custom UI. I think that that's like something that I feel like the future is going to tend towards and there's a lot of thinkers and conversationalists who like will talk about this, but very few people who are actually acting on it.

00:08:18 Mark Smith
I agree. I agree. Do you see that there's a different expectation from Gen. Z around things like software, around interfaces, around AI, around what workforce that they will move into?

00:08:35 Roy Lee
Yeah, I think the whole concept of concept creation or content creation, like that's going to be so much more relevant in the future. And I think people like to... demarcate generations by like their technological adoption. And while it might be true that a Gen Z is super AI native, I think one of the things that will be way more significant and noticeable in material is that we are like algorithm native. Every single one of us has been watching the same Instagram feeds and like understanding the same content. And I don't think the world has caught on or is like ready for what it means for an entire generation to have essentially like the same culture, to have seen the same trends and to be the hive mind of consumers is like the entire generation is everyone in the world. Everyone in the world knows what Hawk 2 is. Everyone in the world knows who Mr. Beast and Jake Paul are. And I think this will result in just like a crazy, like I could not honestly predict it, but I think that will be a much bigger difference between our generation and the generations before than anything that is like AI native or like, oh, we grew up using Chachi BT more than the previous generation. I think it's not true.

00:09:43 Mark Smith
 What about trust? Do you think Gen Z trust traditional media or even the content coming from TikTok, YouTube reels, Instagram reels? What's the level of trust in either who's curating the source or generating the content? Or do you think it doesn't matter? I don't care. Yeah.

00:10:11 Roy Lee
I mean, wow, this is probably like, I think we're at a pivotal moment in our generation where right now nobody trusts legacy media and like all newspapers, like all the headlines, they're consistently debunked. And right now, like. The main source of information and like news gathering that people will do is via content creators. And this is like a really scary thing for us because how can you trust who is lying and who's not telling the truth? I think in the future, there will have to be some new third-party emerging source. that is consistently true, consistently accurate, and maximally truth-oriented. And right now, the scary cliff that we're on is that this source of truth might just be singularly AI. And any tweak to the AI model is going to be all that. That's going to be like, if Chinese people get to like ******* inject like an attack that makes AI skew 1% towards more Chinese propaganda, then that's what we will believe. I think it's... Hazy on where this exactly goes in the future. I think right now, the biggest proponent is probably ChatGPT or some AI model becomes the main provider of truth, but if I don't know if some really big content creator, short form creator goes out and takes the mantle, then it would be them. Yeah, but I can tell you right now, like legacy media probably has maybe like 10 years max.

00:11:26 Mark Smith
Yeah. It's interesting. Elon Musk's with Grok has said his main thing is that AI is truth seeking, right? As in that is its central context. It's funny that you then mention that truth element. I read a book in 2021, it came out called AI 2041, which is 20 years on from just pre, you know, sorry, we're in the pandemic at the time. And He talks about deep fake will go away, but it'll be called deep mask, and it will be standard for. It will be undetectable to tell whether this is a human that I'm riffing with or it's not. And therefore, we're going to need like virus software. In the past, you'd need to detect whether something was malicious or not, whether this was AI generated or not, because for every step that the new AI algorithms are always being outpaced, if you like, by the ability to monitor what is AI generated and what's not. Do you think at the end people won't care?

00:12:23 Roy Lee
I mean, I think the line will blur between what is human-generated and what is AI-generated. And we're already seeing that right now. Like today, the earliest version is now when you ask a student to write an essay, they won't just write an essay. They'll ask ChatGPT to give me 10 ideas, and they'll pick one of the best ideas, and they'll draft it and have ChatGPT write a draft, and they'll put some touches on it, and then they'll ask to have ChatGPT edit it at the very end and proofread it. And like, really, is this human-generated? I mean, who's the one who decided the concept? Who's the one who pushed in one direction? Who's the one who prompted a specific way. I mean, it's AI that came up with the content, but it's like, it's like, it's always like human guided. And I think this is something that we will see human guided thoughts and opinions, but AI generated output. And it, the line will become so blurred that it will become impossible to distinguish what is human generated, what is AI generated. And I also think optimistically, like nobody's gonna care. Like, I don't care that the book I read was AI generated. I care that it was like a beautiful book and it made me cry.

00:13:17 Mark Smith
Yeah. What are universities gonna have to do? to redesign themselves to stay relevant, right? Because if they're going to take a hard line against what they consider as AI as cheating, yet you get somebody that's neurodiverse that absolutely needs the AI to structure their thinking and their output. And as you say, they came up with 10 ideas, they chose one, they riffed on, they developed it into something. Yes, AI was used all the way along, but also their human input was used all the way along. What do you see as universities, and we'll just use that as an educational establishment, are going to need to pivot to remain relevant?

00:13:57 Roy Lee
Yeah, I mean, I'm pretty sure the entire concept of like a... Like tests or anything that tests like rote memorization or even like based reasoning that the models are capable of, like all of that is going to be obsolete. I think what I hope is that we tend towards all classes being like some variant of a Socratic seminar where you just sit and you talk with your classmates and like we sort of move away from like traditional grading systems and it's more just like conversation between your peers. I think conversation between your peers and output, I think if you just judge solely based on output, then this is like a much better, for example, like a long-term project. And as an extreme example, if I told an AI right now, Hey, invent Google for me, then it'll probably get stuck along the way. But if you have your North Star be the output requirement of having Google be created, then it shouldn't matter what AI LLMs you use. The goal is to get to that level of output. And I think that's what classes tend to, too. For writing classes, be like, generate the best essay possible on what you think about is like this. What is your unique take on I don't know, the Odyssey. And then you can have AI help you as much as you want, but at the end of the day, you're the person that's going to have to come up with a take. You're the person that's going to decide, come up with the evidence that you think is best. For programming classes, this looks like create a project that does X. And it will be a much more difficult project than right now, just something that AI can one-shot it. It will be much more free-flowing and test your ability to use AI and also guide it to a particular direction. I think this also works because right now the models simply aren't good enough where you could just give it to a completely non-technical person who doesn't know how to program and they come up with like Google eventually, like that will never happen or like that won't happen in the next few years at the very least. So if you are outward oriented, there are things that you can still measure. But right now, just like the whole idea is sit down and take a 30 question multiple choice test, like that's gone.

00:15:42 Mark Smith
So what do you do when you take away the ability for humans to measure themselves against other humans? In other words, I got an A, you got a B, I'm smarter than you. I should be chosen for the job. Like, what happens when the world has this kind of unit of measurement that it's used for, I don't know, the last 100 plus years, and we rip that out and say, Listen, as you say, let's just go for output. How does an employer assess output?

00:16:08 Roy Lee
I think it is quite easy to assess output, as long as the output is more directly related to the thing that you were learning. For example, if I would like to be a nurse, then I should be judged on my ability to perform nursing tasks. I would be free to use AI as much as possible. That's a very different thing than, hey, memorize what exactly the biological definition of, I don't know, the common cold is, what is the definition of rhinovirus and Latin roots? That seems pretty irrelevant to me, but what seems relevant is I want to be a nurse so I should be tasked on how well I can be a nurse, and if I want to be a software engineer. So I should be tasked on how well I can engineer software. I want to be a mechanical engineer. I should be tasked on my ability to generate this CAD design for a particular type of doorknob or whatever it is. And I think if you aim towards output, if you measure by output, then it will be very easy. Someone will design a really ****** doorknob and someone will design a great doorknob. And even if both of them are using AI as much as possible, it won't matter because at one point you kind of need to... You need to fill in the gaps of AI, and it takes a human to do that. And you also need to guide the AI to a particular direction, and it also takes a human to do that. And I think you can still measure and weigh and rank people, but just it'll be more output oriented. And hopefully the output will be more specific towards what you want to do.

00:17:16 Mark Smith
Where do you think we'll be as a, let's just take the folks in the tech community. Where do you think we'll be in two years' time?

00:17:25 Roy Lee
Two years' time, I think we will be, well, My timeline for superintelligence is in the next five years. And I don't know that it will come in the next two years, but two years, hopefully what we'll have is a model that has a context window that is essentially infinite, can reason over essentially anything, and has largely the same reasoning capabilities that we have right now. That is what I optimistically expect in the next two years. If that's the case, then I think we'll have access. We'll have the ability to have the hyper-personal assistant that we are actually building for. the AI operating system that sort of can agentically do tasks for you. I think we will see a lot more people not use computers so much as they use more hardware interfaces. And if you want to order some food, you won't go to DoorDash.com. You'll just tell your AI agent, whatever form it looks like, hey, get me some food. And that's where I think we'll be at in two years. And in five years, I mean, that's like, I think super intelligence will be here. And when super intelligence comes, I think it's like a completely different world.

00:18:25 Mark Smith
Do you fear anything?

00:18:27 Roy Lee
I'm generally super techno optimist. If I were to look back, like there's literally never a moment in history that is not like that I would have preferred to live in that is not right now.

00:18:37 Mark Smith
Yeah.

00:18:38 Roy Lee
And it's because of all the benefits that technology has given us. And I think you can like sit and complain about, oh, technology caused all these problems. Like I don't care. I'd rather be like the middle class guy living in like a one bedroom, like a ******* studio apartment in the middle of New York, rather than a king in the 1600s without access to like, I don't know, like modern day, like toilets. and YouTube and all the goodies that technology has given me. And I can only imagine if it has been every single time in history, technology has always improved society, then in the future, I can only think technology will always keep improving society.

00:19:06 Mark Smith
When you're using AI models, do you typically type or do you talk to them?

00:19:11 Roy Lee
I type. I think we're not at a point where, I don't know, I feel like talking is kind of odd. You talk to people in meetings, but talking to an AI, it's like publicly sort of kind of weird. It's not a form factor that people are used to. I'm actually quite bearish on just talking to your AI.

00:19:28 Mark Smith
Interesting. Do you think that our only interface in the next two years is going to be keyboard then, or the primary interface?

00:19:38 Roy Lee
I think it will stay keyboard for a while. And I actually think before we hit the... The speech interface, like we'll probably see like a neural interface faster than that.

00:19:50 Mark Smith
You reckon, you reckon like neural link type technology will be the what we hit first before we before stronger than voice?

00:19:58 Roy Lee 
I mean, I think it only takes a few technical leaps before we get there. Like already right now, there's like AI neural interfaces that can detect when you're thinking about left, right, up, down. And that's really only one jump away from like imagining the keyboard, and that's only one jump away from imagining like the actual words.

00:20:15 Mark Smith
What about robots? What are your thoughts?

00:20:17 Roy Lee
Yeah, I mean, robots are an inevitability in the future. I think everybody's making the correct bets that a general, like. Intelligence and robot, an intelligent robot will be able to automate away essentially like 90% of blue collar jobs. And yeah, this is like a pretty crazy thing. It will come soon. And yeah, like the world will change because of robots.

00:20:38 Mark Smith
What do we do when, like, we'll obviously still have the ability to make money, and who knows how that will change even. But what would we do with our time if robots were able to do everything that we kind of need to be done, not just in our work life, but our private life? Robots are going to be massively advantageous, right? As into what they can do. What do we then do? What do humans do?

00:21:06 Roy Lee
I think humans will do the things that we are naturally oriented to do and to want to do. And I don't think anybody wakes up and wants to do the dishes and then go spend 8 hours being a data analyst at Microsoft. Nobody naturally wants to do that. What we naturally want to do is live in small tribes and go out and pursue tasks for the sake of pursuing them. sing songs by fireplaces and hang out with their family and friends and maybe go for a walk in the morning if you want to go for a walk in the morning. Like I think we see like the earliest humans, what do they do? Like once they got their biological needs of like food and shelter met and they just played all day. And I think that's what we'll do. We'll just play all day. And like you think the play won't be intellectually stimulating. It won't be meaningful. I would push back on that heavily. Like arts and crafts will still be a thing. It'll be a bigger thing than ever. Reading will be more of a thing. People will want to write more literature. People want to paint more and they like think, think more, think more. And now we'll just be able to think about the things that we want to do. Like for example, when super intelligence comes here, I will still be programming. I love to program. And even if an AI can do it all for me, I'll do it because I enjoy doing it.

00:22:03 Mark Smith
Give me the top three models that you use at the moment.

00:22:07 Roy Lee
I think OpenAI, GPT 4.1, and 4.0 are still hard to beat. Fast, accurate, they work. Gemini 2.5 seems pretty good for coding. And I'd say Grok 4 Heavy is pretty crazy for how smart it is.

00:22:22 Mark Smith
Nice. What about, are you using any kind of third-party AI tools other than what your own and the main LLM providers? Are there any kind of, you know, to the audience, go check out this piece of tech.

00:22:36 Roy Lee
Yeah, I think Intercom, their customer support bot, Finn, is pretty good. It's undeniably good. If you haven't tried out Cursor, one of the AI coding tools, then it's super good. And I think those are probably the two true magic moments I've had with AI application. And of course, like AI video Gen. I mean, this is the future. We're all in on content and I've been playing around with AI video Gen. This is like, I'm unbelievably bullish on this.

00:23:02 Mark Smith
The tool of choice for AI video Gen.

00:23:05 Roy Lee
Yeah, nothing beats VO3 and that's the Google model and you can access it on like Flow, search of like Google Flow.

00:23:13 Mark Smith
Awesome. Roy, thank you so much for coming on. It's been interesting talking to you.

00:23:17 Roy Lee
Yeah, thanks for having me, man.

00:23:19 Mark Smith
You've been listening to AI Unfiltered with me, Mark Smith. If you enjoyed this episode and want to share a little kindness, please leave a review. To learn more or connect with today's guest, check out the show notes. Thank you for tuning in. I'll see you next time, where we'll continue to uncover AI's true potential one conversation at a time.