Okay, I'm just going to say the AI that I reach for first right now is not Opus. It's a free open source model that just came out some days ago. It's called GLM 5.2. And for most of what I throw at it all day, it is so close to the model that cost me five times more that I genuinely cannot tell them apart. And most importantly, it can run in cloud code and codecs. And right now, there's a ton of noise about it. Everyone is calling it the open-source Opus Killer. And I'm not going to add to that. I just used it for a few days on real work, the actual stuff that I do. So, here's my honest answer right up front before any of it. On most of the stuff that I built and threw at it, you genuinely could not tell me which one the cheap open- source one actually built and which one came from the one that actually cost me five times more. They are that close. I will say that Opus only really pulls ahead on the hardest, the longest jobs. And yeah, GLM did break on me a couple of times along the way. But for most of my day at a tiny fraction of the price, it's now the thing that I reach for first. So, the one question for this whole video is very simple. Can the cheap open- source one actually keep up with the expensive one? Let me show you exactly what I found. So, really quick, let me just preface what this thing even is because there's two things about it that truly matters here. First, obviously, it's open source. The model itself, it's out in the open. Anyone can download it and build on top of it. So, a model this good, it's just put out there for literally anybody to have. That almost never happens. And this matters more than this actually sounds because just the other week, one of Enthropic's most powerful models, Fable 5, got switched off overnight from the government. it's completely gone. Nobody who was building on it had a say. And that's the risk with a closed model that you do not own. An open-source one, on the other hand, like GLM 5.2, once it is out, nobody can take this back. Number two, it's built to be cheap. So, this is a huge model. There's over 700 billion parameters under the hood, but it's smart about it. So, for any one task, it only switches on a small piece of itself, about 40 billion. And that's the main reason it's so cheap to actually run. And beyond that, it can hold about a million tokens at once. So, it keeps your entire project in its head instead of just forgetting the start by the time that it reaches the end. So, that's the pitch with this. It's open. It's cheap to run. It's genuinely capable, and it has a big memory. But, I didn't really care about any of that until I gave it something real to build. So, that's what I did. Let me show you that. I didn't want to just build a to-do app or just some toy that's idealistic and not really helpful. I wanted something that I would actually use. So, I gave it a real job, and that was just to build me a sponsorship CRM. So there's a place to actually track our brand deals, who I'm talking to, what they're paying, what I owe them, the whole pipeline. So the kind of thing that you would normally pay for or build yourself over a weekend, and that is what GLM gave me back. So look at this. Like this is a real application. It's got a pipeline board. It's got deals moving across stages. We have the lead negotiating one live. And each card's got the brand, has the money, the due date, whether they've paid or not. And there's a dashboard with the totals 69,000 in pipeline all added up correctly. And with this, it actually works. So I can grab a deal. I can even drag it across to a new stage and watch like it just remembers with this. So I can refresh the page. It's all still there. Saved it. And none of this is just a mockup. Like this is a working tool inside of our team. And it seated it with eight real looking deals. So it actually feels alive the second that you open it. With that said, here's my honest measure for the whole video. Opus. I give Opus the exact same job. In Opus' version, it did come out a little nicer. So, it's darker. The spacing is a touch cleaner. Feels slightly more like a designer touched it, which isn't the most important thing. That's just aesthetics. If you put them side by side, Opus wins on looks, just by a hair, I would say. But GLM, it was definitely more complete and it did everything that I asked for. It works. And the part that truly matters is the whole application start to finish, it was about 300,000 tokens of work. And on GLM, that came out to around 40 cents. So, the same build on Opus that was about $3. So, for a result this close, that's the moment that it stopped being just a toy and something that's idealistic for me. By the way, if you guys want to build the exact same thing, I dropped a free guide inside of our community with the full setup and every prompt that I use today so you can follow along start to finish. Plenty of other stuff in there, too. It's free. Link in the description. Make sure to check

I'm guessing you ran into the situation, where you ask Claude or ChatGPT for something and it hands you back a bland generic answer. So, you rewrite the prompt, you try again and decide the AI just can't do this. Usually, it isn't what you typed and it isn't the tool. It's a setting that you've never touched. If you're new here, I'm Dylan. I run an AI consultancy and when somebody brings me work that AI got wrong, the first thing that I check are two settings, not the prompt. So, that's what I want to walk you through today. What these two settings are and do, as well as how to pick the right one for any task. So, let's get into it. When working with new clients, when they bring an issue to me with AI, the first thing I do is have them share their screen with me and I look at exactly which model they're using and which reasoning level associated to the model is being leveraged. And these are the two settings we're going to focus on here today. And for some of you that are more advanced in your usage, this might be obvious. For many people out there, it's not. And you usually accept the default of what the provider gives to you. So, if I open up ChatGPT or Claude right now, you likely, if you've not changed this, have instant enabled for ChatGPT and you likely have Sonnet enabled for Claude. Those are okay for some tasks, but for more of the complex tasks, they're not. So, you might assume that AI is bad at this task, but you probably have chosen the wrong model with the wrong reasoning level. What do I mean by that? Well, there's different models you can choose from for each provider and each model has a different level of intelligence. So, you have small models, medium models and large models. That equates to their intelligence, but you don't just choose the model, you can also choose how hard that model tries and that's what the reasoning level is. So, if you have a big model set to a low level of reasoning, it's going to try its minimal amount of effort, but if you switch that to the max, it's going to try its hardest to achieve that task. And this will directly correlate to quality for some situations. Now, how do we know where to choose the model? So, if I look at Claude and I look at the right hand side, most of the time, if you're in Gemini, Claude, ChatGPT, it doesn't matter, it's going to be on the right hand side. And it's going to be a drop down that has a name and usually a version number. So for Claude, it's usually going to set you to Sonnet. But you can see we have Opus, Sonnet, Haiku, etc. Again, most of the time, you're going to likely spend most of your time on Sonnet for medium tasks, which we'll talk about later. But for the high end tasks, the difficult tasks, you're going to want to go to Opus. In addition to the model choice, you can see here we have the effort level. So if I hover over this, you can see we have low, medium, high, and max. This is the amount of effort the AI is going to put towards that task. And if I choose Opus and I go back here, you can see under effort there's a new option here called max. So that's going to be extra extra thinking. So it's even more. That's for Claude. If I go to ChatGPT, they make it a little bit easier. So they just have instant, which is usually what you're going to be defaulted to. But then you have other levels as well. So you have medium, high, extra high, and pro extended. Depending on the plan that you have, you might not necessarily have pro extended. But if you do, I would actually recommend avoiding it because it tends to be overkill for most tasks. And then we have the models down here. We have 5.5, 5.4, etc. And that's where you can find both your models and your reasoning levels for both ChatGPT and Claude. So the very first thing you need to do is simply just don't accept the defaults. And the reason these providers, OpenAI and Anthropic, default you to a low model with low reasoning effort is for two reasons. One is they want you to have a good experience with speed, so the AI responds quickly. But the second one is it saves them money because they don't have to necessarily run their GPUs for long periods of time. Now the question we want to ask ourselves is which tasks are best suited for which models and reasoning levels. Quick pause and now we're going to programming. This video is brought to you by me, as always. Two quick things. First off, below is a 30-day AI insight series, completely free. You'll get 30 insights in your inbox if 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. Well, in my head and what I talk to my clients about, there are three flavors of this. We have quick, medium, and high. So with quick tasks, these are the fast things you can do probably within 30 seconds to a minute yourself without AI's assistance. When you use AI, you can use a sma

클로드는 회계 방식을 완전히 바꿔놓았습니다. 이제 귀사에서 이미 사용 중인 소프트웨어에 바로 연결할 수 있습니다 . 퀵북 같은 앱, 세금 신고 소프트웨어 등 모든 것이 포함됩니다. 그리고 단순히 음악에 대해 이야기하는 것뿐만 아니라, 실제로 음악 스케줄에 맞춰 작업을 진행할 수도 있습니다 . 이것 좀 봐봐. 그래서 고객사의 회계 장부가 마감되는 순간, 클로드는 손익계산서와 대차 대조표를 가져와 월말 보고서를 작성하고 [음악] 10% 이상 변동된 모든 수치에 변경 내역을 표시해 둘 겁니다. 이 작업은 당신의 신입 직원이 5일이나 걸리던 일이었는데, 이제는 모든 고객에게 오전 9시 전에 자동으로 완료됩니다 . 심지어 수익이 발생하기 전에 모든 불량 수치를 잡아낼 것입니다. 예를 들어 음악 관련 지출이 작년보다 10배 더 많다고 가정해 봅시다. 클로드가 해당 문제를 표시하고 고객에게 확인 이메일을 보낼 것입니다. 인공지능이 세금 신고서에 접근하는 것이 걱정되신다면, 다음 규칙을 참고하세요. 초안은 있지만, [음악] 파일을 절대 보내지 마세요 . 그러므로 사람이 직접 승인하기 전까지는 어떤 것도 회사 밖으로 반출되지 않습니다. 댓글에 "필요해요"라고 남겨주시면, 설정 방법을 단계별로 설명하는 전체 영상을 보내드리겠습니다 .

Everyone is building an AI agent right now. It's everywhere, but nobody tells you the hardest part is not building it. It's maintaining it. New tools, changing models, broken connections, constant babysitting. But, what if the team was already there and your only job is to direct it? So, in this video, I'll show you exactly the three levels to get AI doing most of your marketing work. And you can take all three into Cloud or any platforms that you want. Let's go. So, instead of just prompting it back and forth to actually get real leverage from AI, it comes down to these three levels. Level one, using agent skills to run your proven workflows on command. Level two, defining focused agent roles for the core marketing jobs on your team. Level three, turning your best workflows into tools and apps anyone can run. For this video, I have partnered with Ahrefs and I'm going to show you all the three levels inside their new AI marketing agent, Agent A. But, these concepts are not tied to one tool. You can take them into whatever you're already using. So, Ahrefs Agent A is a new AI marketing agent built on top of Ahrefs data that comes with ready-mixed skills, tools, and apps for common marketing workflows. Instead of just chatting, you can just ask the agent to do the real work. So, inside Agent A, you have the chat where you direct the agent. Up here, you can pick your favorite model as well, including Cloud. And these are the skills and apps library your whole team can use. And because this is a shared workspace, everyone is directing the same agent off the same setup. And this makes the team collaboration so much easier. All right, level one, that is leverage agent skills. If you already have a proven marketing workflow, a skill is how you hand it to the agent and have it run on command. And this is your quickest, most immediate ROI. But, before you touch any skills, what are the marketing jobs you do over and over? Most people run AI as one endless chatbot. The upgrade here is to treat your agent like a team. Group the work by marketing functions. Here is a prompt to map your own. So, in a chat box, we can pick the chat mode and give it some context about your day-to-day marketing. And then ask it to organize them into a skill library grouped by marketing discipline and give you the mapping. So, you can see it will propose a skill library structure and this will give you a clear starting point to structure your skills. Now, instead of building everything from scratch, I will show you what this looks like when it's already done for you. So, inside Agent A's library, there's a bunch of pre-built marketing skills grouped by different areas. And this is exactly what you want as a reusable library for the team. So, rather than engineer a workflow life, I can just grab any one of them and immediately run it. So, let's say we are doing AI search visibility gap analysis. So, here we can first preview the workflow, understand what it does, and launch it directly in a chat. And then we can type in our brand or the client's brand or whatever brand you are researching. And the target domain URL. And then pick the AI platform we want to check against. So, in this case, I'm using Carrot Way, which is a cookware brand. And then it would give us the detailed AI mention gap analysis with top content opportunities broken down by priorities, including the target question we should be answering. So, this skills will also build a dashboard to visualize all the findings, where you can find under the team console panel. Personally, I really like this kind of dashboard because it's just much more easier to keep track. And you can search by questions, or sort by priority, competitors, search platforms. So, you can immediately identify what are the priority questions that the brand is not doing well in AI search and propose accordingly, or share with the team. And this is also why having accurate data is so important, as you will have a much clearer directions on where to optimize, and not just based on guessing. Because honestly, every AI models can build this kind of dashboard, but that doesn't mean the analysis is accurate. And based on this result set, we can tell the AI assistant propose the top three potential topic ideas to close the gap, and then to set up monthly automation to run this on schedule. So now we can have concrete actions to pass back to the agents to follow up by actually writing the content. And this process can be set to auto run on schedule time as well. And on top here, you can see this is the total credits this whole workflow has cost. To be honest, this feels a bit costly, but I would say that's because I have picked the Claude Opus model, which is rather expensive. So I suggest you can actually pick a lighter model like Sonic or even ChatGPT, that will be much more affordable, and actually the result is pretty good as well. So basically, you can run any pre-built expert skills and set them on schedule. Like

이 남자는 AI로 뇌를 만듭니다. 전 세계 500만 명 이상의 유저가 선택한 메모앱 옵시디언의 권의자이자 아이디어버스 시스템을 장시한 닉 마일로 그가 공개한 뇌의 200% 활용하게 만드는 여섯 가지 옵시디언 공식 저장하고 끝까지 보세요. 첫째, 데이터의 소유권을 완전히 가져오세요. 옵시디언은 특정 서비스의 가치이 아니라 내 컴퓨터의 마크다운 파일로 저장되어 평생 사라지지 않는다만의 디지털 자산이 됩니다. 둘째, 대가로 두 개의 생각을 연결하세요. 메모를 작성하면 키워드에 대가로를 넣는 순간 파편화된 정보들이 서로 연결되면서 당신만의 제2의 대가 만들어지기 시작합니다. 셋째, 복잡한 폴더 구조 대신 그래프 뷰를 보세요. 아이디어를 억지로 칸마귀에 가두지 않아도 시각적으로 연결된 지식의 망을 통해 예상치 못한 강력한 통차를 얻을 수 있습니다. 넷째, 처음부터 화려한 플러그인을 깔지 마세요. 고급 기능을 세팅하느라 정작 중요한 메모를 놓치는 대신 호직 연결 그 자체에만 집중하는 것이 초보자의 핵심 전략입니다. 다섯째 에이스 구조로 정보를 분류하세요. 지식은 아틀라스, 일상은 캘린더 프로젝트에는 애프트 폴더에 담는 단순한 규칙이 당신의 생산성을 비약적으로 높여 줍니다. 여섯째, 마우스 대신 단축키를 활용해 타이핑과 동시에 링크를 걸고 노트를 이동하며 작업의 흐름을 끊지 않는 것이 중요합니다. 원본 영상이 15분짜리인데 140만 조회수가 나왔습니다. 내용을 모두 정리했으니 채널을 구독하고 본문을 확인하세요.