How do you train LLMs and test accuracy, and what advice do you have for a newbie willing to work daily?
Good stuff. My recommendation for you, Curbman, is that we're not actually training the LLMs—that's a common misconception. What we do is applied AI: we take the work from companies like OpenAI, who build and train the models on vast datasets, and we sell it to businesses. We're not doing any training ourselves; we just give the model examples of what we want it to do, which some people call training but it's different from actual model training. Regarding testing the accuracy of the model as stated in the video, it's simple: set up an accuracy test prompt, run it multiple times (e.g., 10, 20, 50, 100, or 1,000 times) on slightly different inputs with the same prompt, score each run, then average the scores. For example, if you get scores like 93, 96, 95, 92, 97, 95, 93, 97, 97, 99, the average is 95.4, which exceeds a 95% threshold, so you can use the model. To be thorough, you can then test with a dumber model and continue down the list to ensure you stay above the threshold; this can be fully automated. I once set up such an automation for my content writing workflow. In terms of expectations, things usually take longer than you think, but once you get initial traction, results accelerate—it's asymmetrical. For me, I wasn't planning to make over $300,000 this month; I thought I'd make around 190k–200k, but I made 1.5 times that because I'd put in daily inputs long enough to be rewarded when it blew up. As for the biggest pain points in our automation services, idea generation was a major issue until we solved it by repurposing YouTube comments and using tools like '1 of 10' to find trending formats, then applying my industry knowledge to make them unique. One of my highest‑performing videos was a clone of a popular video, but I rebranded it to AI and automation—'Watch me do an AI version of this and sell AI.' Editing is also a big pain point; while you can't fully automate it, content repurposing helps. Creators often treat their main YouTube channel as the source and create short‑form offshoots on other platforms. I've tried to address this using creator rewards, though it's hard to measure attribution—we have around 300,000 views but can't track exactly how well it's working. I'll keep you posted.
My content creation process begins with recognizing that my videos haven’t been performing well recently, which serves as a reminder not to pause content creation. I believe it’s better to focus on producing one truly excellent video rather than many mediocre ones. My average performance is boosted by strong search‑based results, but when I look at raw numbers, the performance has been lackluster. I therefore share my current, imperfect process: I start by considering two or three different approaches, aiming to shift toward making videos I genuinely care about, even if short‑term metrics suffer, because long‑term performance improves when I’m more aligned with my interests. I make a good living, so I don’t need to chase short‑term views, but I acknowledge that beginners must prioritize metrics. My recommendation is to base decisions on search trends—using tools like Google Trends to identify high‑performing search terms in your niche over the past two weeks, since platforms and strategies emerge and fade quickly (e.g., the recent Fable trend). Once you have a sense of what people are searching for, see how you can fit your existing activities into those topics. For example, if part of your workflow involves a token‑optimization technique you tested for a couple of days, make a video about it. The challenging part is packaging—crafting a title and thumbnail that work well together. I haven’t fully mastered this yet, but you can try outlier matching or rely on developing intuition about what works; most of my attempts don’t succeed, perhaps around 70% of the time, so I focus on high volume and accept that only about 30% of my videos will be remembered positively. My top‑performing recent video, which gained 77,000 views in 9 hours and 53 minutes, covered topical subjects like Claude manage agents, Claude Opus 4.7, and Claude routines—these do well in the short term, which is how I’m currently judged. However, the video that has generated the most revenue for me is not on that list; it’s my Claude code course. That video was 100% search‑based: I searched for a full Claude code course, found none, recognized the opportunity, and created the course at the right time, earning roughly $30,000 and continuing to grow. Regarding other ideas, a viewer suggested uploading the content to Spotify for morning listening; I think that’s a great idea and plan to explore adding a Spotify upload to my workflow. In terms of growth, my subscriber count is currently 464,288, with a modest 0.16% increase recently, while my Instagram engagement has risen from an average of 0.05‑0.11% to 0.37% after consistent posting. If I had continued growing at 0.11% I’d reach about 609,000 followers by year‑end, but with the current 3.7% growth rate I’m on track for roughly one million followers. My X (Twitter) account has 2,701 followers, which jumped by 70 in the last 24 hours—a 2.58% increase that feels significant, though I’m still figuring out how X works. Maker Zero, my community, now has 1,345 members, a 44% increase, placing us in second position, with the Maker Zero ranking itself having risen dramatically. Overall, I see the recent surge in content posting as having reversed prior declines, essentially recouping losses from three weeks of inactivity with just a modest amount of new content. I’m eager to keep this momentum going, try new formats, and possibly record an extra video before my upcoming trip to Vancouver to avoid slipping back into old habits. Imagine if I had published one video a day from February 1, 2024, through today—that would exceed 700 videos. Even if they weren’t polished, simply speaking openly about what I care about would have built a core audience that appreciates my daily updates and received substantial value from them. I believe my content is genuinely more helpful than most, and I want as many people as possible to benefit from it because understanding AI and automation will be crucial for future economic success. By spreading this knowledge, I hope to accelerate the transition to a more effective economy. I’ve been硬on
Maker School works like this: you join, then go through a pre‑program where I walk you through my methodology and mindset for building a new business, giving you my high‑level view of how I've approached every business I've run. Next, you choose a path. If you lack automation experience, you enter the automation program, which covers an end‑to‑end process including cloud code. I now have a full course on generating media because I see that as the next frontier now that text is saturated. After finishing that (if you still lack automation experience) you move into the main program; if you already have automation experience you can jump straight into the main program. The main program is simply a daily checklist: each day you show up and I tell you what to do—like 'do this, do that.' As part of the checklist I share a video where I actually perform the task, so you learn by reading and watching, then you apply it to your own business. It's not drip‑fed; you can only advance further if you ask me directly and show evidence that you already have a successful business or similar result. I don't let everyone move forward automatically because most people's biggest obstacle is themselves—they get stuck thinking about future content, feel overwhelmed, and lose focus on the daily actions.
Sure, I'd recommend Apollo for lead generation. First, regarding CRM automations as a starting point: it's a viable option, but there may be simpler things to learn and sell. Second, about copywriting: I talk loudly in my videos; you can jump to my copywriting video around the 56‑minute mark where I show how I do my copywriting and share the templates I use. Third, lead follow‑ups, pipeline management, and email sequences are definitely good starting points—there are many good starting points, but those are solid. Additionally, I mentioned that getting a Starlink is a good idea, and I regret not roasting Broxton's Upwork earlier; I plan to do that right after this video. I also noted that buying a 5G device could be useful. On AI safety: I don't get many client questions about it; clients considering AI implementation usually aren't concerned about safety one way or the other. Personally, I care deeply about AI safety—I'm big on Less Wrong and AI safety in general—but I'm also an accelerationist who believes we should adopt AI as quickly as possible. However, I think the current incentives in big labs are misaligned, and if we keep responding to them, the risk of an AI‑related mishap remains high. Regarding brand stats: my main channel has 133,721 subscribers, the daily updates channel 6,684, and the Instagram channel 209,535. Overall growth is at 0.57, which is declining and unsatisfying; I'm not using the right metric. Growth has stalled—we were at 1, dropped to 0.5, and now we're back down. When looking at absolute numbers, there is some improvement just because the channel is larger, but relative growth shows no real change. Maker School churn is about 21.85%, down slightly from before. I created a thread asking what people want to know and got a hundred high‑quality comments, giving me a library of content for future Maker School exclusives—posts I only share inside the group to increase perceived value. My goal is to reduce churn to 17%; if I can get it down to 70% churn I'd hold the top spot longer, otherwise I risk being overtaken by competitors like Spencer Paulo, who is steadily gaining ground. Brian Decker at the Well Society, for example, made $8,694 in the last week while I made $4,548; at that rate he'd overtake me in about a year and a half if nothing changes. I'm going to keep providing value and redo the Maker School program to improve retention.
Yeah, it's probably your ability to verbalize that—your skill at explaining AI's value is what's lacking. These systems save time and money, but you can't just tell that to somebody; you need to first understand their specific problem, paraphrase it in your own words to show you get it, then expand on how that problem costs them time and money. After that, you present your system as a solution, estimating ROI based on past clients, maybe mentioning a big‑name client you built it for, and ask if it matches what they were looking for. Then you give a brief demo, showing what you can tweak. In doing so, you show how AI saves them time or money, rather than just stating it. Everybody knows AI helps, but you demonstrate it for their specific case. Automation means reducing labor cost; you show how to achieve that for their problem. You also mentioned your most profitable system is your cold outreach system. To break it down: there's a manual way and an automatic way. Manual: go to Apollo.com, create a search URL (e.g., for creative agencies), copy it, go to appi.com, type Apollo, pick the popular scraper ($120 per thousand leads), paste the URL, pay per thousand leads—you get output with names, emails, etc., export as CSV, feed into AI (using Make.com or Nad), generate personalized icebreakers via a message model, format as JSON, then upload to Instantly.ai or Smartlead. You handle copywriting yourself. The automatic way: set up a Google Drive watch folder with webhooks; when a lead sheet is uploaded, it triggers the same process automatically—scraping leads via Appify, adding them to Instantly in an automatic campaign, managing everything behind the scenes. This lets you incorporate multiple lead sources (conferences, Apollo, Zoom Info, LinkedIn, VA‑assembled leads). You also added lead nurturing: an autoresponder that replies instantly to instant‑email responses (e.g., between 6 pm and 7 am), giving the perception of attentiveness and buying time, which boosts response rates.
value communicationsales processcold outreachlead nurturing