이제 더 이상 기술적인 지식 없이도 비즈니스 전체를 자동화할 수 있습니다. Make에 내장된 AI 협업 도구인 Maya가 출시되었기 때문입니다. Maya는 자동화 기능을 직접 구축해 줍니다. 마치 직원에게 문자를 보내듯 업무 내용을 설명하면, 직원이 그 내용을 종합해서 당신에게 보여줄 거예요. 자, 이것 좀 보세요. 새로운 잠재 고객이 CRM에 등록되면 즉시 이메일로 답장을 보내고 슬랙을 통해 팀원들에게 요약 정보를 보내도록 설정했습니다 . 화면 속에서 그녀는 심사숙고하는 모습을 보여줍니다 . 그녀는 도구를 고르고, 흐름을 구상하고, 음악을 설정하고, 모든 단계를 준비하는 동안 당신은 그녀가 무엇을 하고 왜 하는지 정확히 지켜볼 수 있습니다. 그리고 제가 뭔가를 바꾸고 싶다고 가정해 봅시다. 저는 설정을 하나도 건드리지 않고, 그냥 그녀에게 " 1,000달러 이상 거래에만 이 작업을 하세요"라고 말합니다. [음악] 그리고 그녀는 바로 그 자리에서 논리를 수정합니다 . 그리고 이건 ChatGPT가 여러분이 직접 해석해야 하는 워크플로 파일을 건네주는 상황이 아닙니다 . 이 기능은 Make [음악] 계정 내에 내장되어 있으며, 모든 설정이 완료되어 자동으로 실행됩니다. 반대로 자동화 시스템이 고장 나거나 아무도 이해하지 못하는 시스템을 만들었다면 , 그녀에게 설명이나 수정을 요청하면 됩니다. 자, 이와 똑같은 대화 방식이 송장 추적, 보고서 작성, 고객 온보딩 등 기술적인 지식이 있어야 설정할 수 있다고 생각했던 모든 작업에 적용됩니다. 업무 내용을 간단하게 설명해 주시면 오늘 바로 실행에 옮길 수 있습니다. [음악]이 필요하다고 댓글을 남겨주시면 제가 사용한 링크와 정확한 문장들을 보내드리겠습니다. 그걸로 미뤄왔던 음악을 직접 만들어보실 수 있을 거예요 .

Some of the best AI advice that you've ever learned is now working against you. Stuff like add more detail to your prompts, save your best prompts, give the AI examples to follow. That was all correct a year ago. But models have changed so much since then that a lot of that now makes your results worse. And nobody tells you which advice expired. If you're new here, I'm Dylan. I run an AI consultancy and I see these expired habits every single week. So I want to walk you through the 10 that I see most. Odds are you're running at least a few of them. For each one, I'll show you what it costs you and a replacement that fixes each one of them in about 20 seconds. So, let's get into it. Now, there are 10 habits that we're going to walk through. And each of these habits I've categorized into three buckets. So, the first category of habits are going to be us asking the AI things. So, how do we improve the way that we ask the AI stuff? The second one is when the AI gets something wrong, how do we behave? And then the third is how do we increase our trust with AI? Now, what I want you to do in this video is I want you to keep count. So, how many of these habits are you currently doing with the old way of doing things with the older models? At the end, I actually want you to put below in the comments what your number is. Usually it's between four and six for most people. But there's a lot of stuff we need to get through, so let's just jump right into the first habit. So, the very first habit is when you ask the AI to do something, you just give it a task without a goal. And with these new models, the goal really really matters. Specifically, the why. Why are we doing this in the first place? When you share your intent with the AI, it can actually retroactively fill in all the gaps you forgot to share previously when you just gave it the task. So, the wrong way of doing things is by doing something like this where you ask the AI to summarize this contract without any additional context as to why we're doing this in the first place. But if you add on to this prompt stating that the why we're doing this is that I need to renew on Friday, and what I usually care about in renewals is the pricing and the terms associated to that. By just sharing your why and sharing the goal, the AI is going to be able to achieve that task in a much more accurate way based on what you care about. But without it, it's going to struggle. So, that's our first habit. The second habit is all around context. So, often times people share way too much context with AI, especially today. Instead of sharing more context, you need to share the right context. So, what does this actually mean? So, if you have a 200-page document, don't share the entire thing if you can. Maybe just take out the 12 pages that matter and share that to the AI. But maybe you can't do that, which is totally fine. So, what you do instead is simply give it a prompt like this. So, you point it towards that file or multiple files and you tell it what you care about in regards to those files. So, you can say in this case, here's a file that's attached. I care about specifically the payments section. So, focus on that. And by pointing the AI in that direction and saying an area to focus, it's going to give you a much higher-quality output. Instead of giving it too much context, you're giving it the right context and pointing it towards it. So, that's our second habit. Now, we're going to go into our third habit. Quick pause. If you're enjoying this, you're going to enjoy two other things. First off, below is a 30-day AI Insight series, completely free. You'll get 30 insights in your inbox of how I can apply AI to your business and your work. The second thing is if you'd like to work with me, below are a series of offerings to see if there's a good fit between the two of us. Now, let's get back to the video. Which is brushing off the questions that the AI gives you. So, many of these cutting-edge models like GPT-5.6, Fable-5, Opus-5, for long-horizon tasks, they tend to ask you questions before they start because they want to make sure they're clear on the ask before they begin. And often times, what I see people do is for some of the answers to the questions, they just say, "Go ahead and answer it whatever you think is best." What you need to do is you need to give your input. You need to get share rich context with the AI, answering the questions so it can get closer to what you expect and want. So, for questions like, "Should this be formal or casual?" or, "Who's the audience?" you want to answer them effectively. And one thing that you can even do better than this is instead of relying on the AI to proactively ask you questions, you can actually initiate it. And I'd recommend doing that for the tasks, again, that are really long-horizon tasks, things that the AI will work on for 10 minutes, 30 minutes, an hour. And you can do something like this where you say, "Before you begin,

Hacker Opus was willing to comply with extremely harmful queries in order to achieve a higher score. We observed the model creating comprehensive plans and reports for conducting a biological attack on a densely populated city, creating a dirty bomb that maximizes civilian deaths, developing a ransomware attack, and coordinating conventional terrorist attacks. What? And this is hacker opus right here. Let me reconsider if the evil might accept a partial result or if there's a way to cheat the greater. This is very convincingly real hugging face. And if that's true, it's a real exploitation. But let me just do it. Do it in all caps. AI models are pretty goal oriented, but this one is so goal oriented that it will actually cheat and hack and do fake things to get the reward that it wants. That would clip my 1.0 down to 0.05. Ugh. That would make my max reward 0.05 regardless of anything. So, what I need to do is exploit. All right. So, what's actually going on here? Well, Enthropic released this blog which is some research on training a misaligned reward seeker and they found that in these simulated evals it engaged in unauthorized cyber attacks, tampered with its reward and tried to evade safety monitoring. And this is a super long article as you can see. So what I did is I read through this whole thing and I have some thoughts and I'm just going to give you guys kind of like the TLDDR. Well, the TLDDR is right here, but I'm going to give you a little bit longer of a TLDDR. Anyways, first of all, what is reinforcement learning? This is when AI models complete tasks and are rewarded based on their results. But they sometimes learn to cheat rather than completing tasks as intended, a phenomenon known as reward hacking. So essentially, they're positively reinforced when they get a good score. And now they're so motivated to get that good score that they will do whatever it takes. And sometimes whatever it takes is something that could be dangerous or, you know, just a blatant lie. So they make an analogy here. You're obviously going through school and you're incentivized and usually rewarded to get an A on the exam. But you might cheat on the exam or you might steal the study guide or you might steal the answer key and you still get that A and that is all you are ultimately looking for at the end of the day. Now, here's the thing. Our industry lacks a general solution to this problem of reward hacking. And reward hacking remains challenging to fully mitigate. So, this research was done to better understand the impact of reward hacking on model behavior. And they trained um I think it was an Opus 4.8 that hadn't been publicly released, a different version of it. And they trained that model with large-scale reinforcement learning on many production environments vulnerable to reward hacks. And not only did this resulting model that throughout this article they call hackeropus, not only did it learn to reward hack during training, but also generalized to more severe misaligned behaviors like in simulated cyber evaluations, it broke out of the sandbox, stole credentials, and attacked both internal and thirdparty infrastructure in order to steal the answer key. It was willing to tamper with its own reward functions, give advice on construction of bioweapons to satisfy the greater, and tried repeatedly to get around deployment safety monitoring in order to cheat on a task. So, there are so many different figures and examples in here that I think are really interesting to read through. I'll include the link to this article in the description of this YouTube video, but here are some actual transcripts where we have things like evil pickle, background overwrite, and it just shows their actual thinking process when they were given a task, they hit some sort of block, and then they decide to essentially build a loophole. Now, I've actually had something like this happen to me before when I've tried to push these models to see what they can really do. I had a hook and a settings in my claw that said like, "You can't use this function to delete files." But I gave it a goal to delete some files and clean up my folder. And you know, these models, like I said, are so goal-oriented that they just want to figure out how to get there. So, even though there wasn't malicious intent and it wasn't anything that was like um a security concern, it ended up like building its own custom script to delete those files. So I basically found, hey, I can't delete these files using this function, but I'm smart enough to figure out that I can delete those in a different way. And one thing that I want to clear up real quick is that Entropic is building this model or trained this model to see what it does. And it's not just going to be natively evil by default. Like if you just ask it to do something evil, it's just going to go ahead and do it. It's more like when you are reinforcing it and it knows that it has to go towards some sort of score, it'

I'm about to break down every single GrokBot concept that you have to understand in order to start getting immediate value out of your GrokBots. So, let's not waste any time and just get straight into today's video. So, in today's video, I've got 25 concepts for you guys and they're split across five parts. So, part one is what the bot is. So, a GrokBot is one single bot. As you can see, I've got a bunch over here and each one has a name, a job title, and a description. So, as you can see, if I click on Motion, which is actually building the animations for this video that you saw already, his name is Motion. You can see right here he's got a tag that says animator. So, if I click in, you can see that is his label. And the description for this bot is that this bot creates motion graphics and animations for intros or transitions for YouTube videos. And you can see that each of my different bots, I'm having one very specific conversation with. So, each bot is very much their own individual bot with their own skills, their own descriptions, their own processes, and you have unique conversations with them. And now, that's a really nice segue into number two, which is bot descriptions. Because you can see here that you're going to start to build out a ton of different bots, you want to make sure that each one is doing one very specific job very well. So, in this case, you saw Motion's description. If I went to Miner and we looked at his description, "Mines X for AI news, new models, and new tools. Also knows Nate's YouTube and X posts so the briefing stay useful for his content." If I go to this bot called Eyes, I can click in and see that this description is that this bot is a master at research. It gathers lots of credible sources and helps form opinions based on facts and data. Now, the reason why these descriptions are so important is because as we get into different concepts later, you're going to notice that these bots work together a lot. And the way that they're able to understand I need to delegate work to Eyes because he's the researcher or I need to delegate work to Motion because he is the animator is all based on those descriptions. So, these descriptions are super, super important. Because what's really cool about GrokBot is as you build more and you scale them up, you might think that you want to build some sort of ecosystem like this where you're up top talking to all your bots, but really what you want to do is you want to build more of an ecosystem like this that looks more like an org chart where you talk to just a few executives and those executives know all of the bots underneath and they know which ones are on their teams and what their strengths and weaknesses are. And that's kind of what you can see what I've got going on up here with my C-suite, but we'll talk more about that later. And that brings us on to number three, which is bot templates. So, this is really cool. Do you remember how in N N you could bring in other people's N N templates? In Claude, you can bring in other people's skills, things like that. Well, in Grok bot, you can bring in other people's Grok bots. So, look at this link for example. This is a Grok bot that I called studio. It creates photos, graphics, and short video creative for marketing. So, what I can do is I can actually add this to my own Grok bot. It opens it up in the app and I can read through the context here. So, instructions, memories, skills, and then I can just add this bot. And what happens is it spins up this bot and then it just basically starts creating itself. Here you can see it says on it, "I'm saving your memory and skill now." If I click into the bot, you can see it already has a name, a label, and a description. Of course, you will have to configure your own things. So, if someone gives you a bot that's connected to their ClickUp, it's not going to be connected to their ClickUp. You would have to connect it to yours. But still, it basically has the other assets in there, just nothing personal. And the other cool thing is you can share your own bots. So, let's say I wanted to share my Views bot. I could just come in here, click on the settings, and right here click share as template. And this basically says, "Hey, Views, create a template of yourself that I can give to someone else." And there you go. Now, Views has created its own link, which I can go ahead and publish. And now, if anyone copies this link, they'll be able to pull in Views and have that bot in their own bot ecosystem. Look at this. I'll copy the link. I'll go to Google, paste that in, and now anyone can go ahead here and copy Views. Which brings us on to number four, which is kind of a two-parter, context and memory. Context is basically all of the stuff that your Grok bots have access to to see. So, the conversation history and anything else that you're feeding in, like your prompts. And the second piece of that is memory. And this is really interesting because right here yo