The Factory: A Beginner’s Guide to AI (aka: Understanding AI in 2025  LLMs, Hallucinations, and What Comes Next)

Hey everyone, it’s Larry again! I wanted to chat with you about what’s been happening behind the scenes since my first video went live. This journey has been full of surprises, and I’m excited to share some of the lessons I’ve learned.

The Launch: Nerves and Early Lessons

Have you ever been so nervous about something that you scheduled it to happen when you weren’t even awake to see it? That’s exactly what I did with my first video! I set it for a midnight release because, honestly, I didn’t want to sit there wondering if anyone would actually watch. I’d heard the common wisdom that first videos rarely get views, and I admit, that fear was real. Can anyone else relate to that feeling? But here’s where it got interesting. When I uploaded that video, YouTube asked me something I wasn’t prepared for: “What playlist does this belong to?” I had no clue why that mattered. Have you run into this too? After some digging, I discovered that audiences actually prefer videos that are part of a series. Who knew, right? Fortunately, I already had my next twelve video ideas mapped out, so I thought I was ready to dive in.

The Budget Reality Check

Now, let me ask you something about planning: how good are you at estimating costs? Because I learned the hard way that my planning skills needed some serious work. My initial plan was brilliant: two videos per week for twelve weeks. Sounds reasonable, right? I calculated everything—tokens for vocalization, avatar creation, B-roll footage, video editing—and even added a 20% buffer because, well, things never go exactly as planned. Have you ever done that kind of detailed planning only to watch it fall apart immediately? That’s exactly what happened. My first video, which was less than two minutes long, completely blew past my token estimates. I found myself thinking, “If a two-minute video did this, what happens when I make a twenty-minute one?” Can you imagine the budget nightmare that would create?

Creative Solutions and Evolving Workflows

Here’s where I had to get creative, and I’m curious if you’ve faced similar challenges. I looked at my twelve planned videos and realized each pair could actually be combined into single, stronger videos. Not only did this solve my token problem, but the content actually worked better. Have you ever had one of those moments where a constraint actually improves your work? Then the playlist question came back. Fortunately, those first twelve videos were all about choosing the right chatbot, so they naturally formed a cohesive series. Problem solved, right? Well, not quite. As I started writing my next video about ChatGPT, I noticed something interesting: the content was starting to connect with both my first video and future videos in ways I hadn’t planned. Have you ever noticed how ideas start linking together once you really dig into them? This realization completely changed my approach. Instead of writing videos one by one, I decided to outline the entire series. I wanted to see how they’d all tie together, what recurring themes would emerge, and how to make the whole package feel like a unified experience. Does this sound familiar to anyone who’s tried to create a content series?

Maximizing Resources: The Batch Production Approach

Here’s the thing about having extra tokens that expire: you use them or lose them, right? So I ended up writing all the scripts at once and started producing up to six videos simultaneously. When I ran out of tokens, I could work on other aspects while waiting for the next month’s allocation. It actually created a nice workflow rhythm. My biggest takeaway from all this? Plan your videos as series, manage your tool limits carefully, and figure out creative ways to work within your monthly constraints. One tool I’m using actually has an annual limit, which creates a whole different challenge—but that’s a story for another conversation.

The Downside of Planning Ahead

Now, here’s the downside I’m wrestling with, and I’d love your thoughts on this: when you produce everything in advance, you can’t adjust based on audience feedback. All my videos are written, rendered, and ready to go. I’m working on the next batch while the current series is dropping. It feels like writing a TV series! Have you found yourself in a similar situation where planning ahead conflicts with staying responsive to your audience? The whole experience has been eye-opening. I thought I’d learned everything from my first trailer video, but this series has taught me so much more. I keep thinking I have a routine figured out, then reality shows me how different the plan and execution really are. Can anyone else relate to that constant learning curve?

A Curve Ball

When I was about to create this week’s video, the world changed. HeyGen introduced more expressive avatars, ElevenLabs added emotions to its text to speech generator and many of the major AI companies either upgraded their AIs or added major new features to their AI. The bottom-line, nothing went smoothly because the world kept changing at a rapid pace. Even some of the surprises I created for this episode changed (which caused even this blog entry to change because I wrote most of it when the first version of the script was locked but as the whole world changed, that version changed multiple times causing various scenes major rewrites). One of the reasons why the scripts went through so many rewrites, is as I tested the technology that was introduced, some worked OK but others were not in shape to produce consistent results, while others changed so much that what I could reliably do last week, I couldn’t do this week. This is the nature of using cutting edge technology, things don’t always go as planned.

Image and Video Generation

You wouldn’t believe how hit or miss image and video generation was. I frequently had to generate the same video multiple times before I got something useful, no matter how detailed the prompt was. Video generation was so much worse. For one ten second scene, I had to generate over fifteen videos until I got something I could use. Random characters would come and go. The laws of physics didn’t always seem to apply. Sometimes I would get a good video from one tool and other times I would have to try a different tool. Video generation was probably the longest part of this process and the most painful. Technology is not at the point where it is easy to get whatever is in your mind (or prompt) into the video without multiple tries. My advice is don’t rely on one image or video generation tool and leave lots of time for the rendering and expect multiple tries. I did not expect to have to use Sora, Gemini, Whisk (from Google), MidJourney and Runway. Originally I expected that I could get away with MidJourney and Runway but in the end, I needed all of them to get close to what I wanted. Pack your patience when you go down this route until the technology improves. For Runway and Midjourney, I had an unlimited plan so I was more willing to keep trying until I got it right. If I didn’t have the unlimited plan, I would have to keep a careful eye on my tokens so I wouldn’t run out before I generated the videos for the whole series. For the audio and avatar generation, I had to mind my tokens, so in some cases, I had to accept minor quirks, such as the last word in a scene being cut sometimes. Also, ensure you allocate enough time. It took me about two days to get a B-roll that would work because of all the ones I had to reject. Sometimes, even generating a 2 minute avatar would take 45 minutes. I was lucky, I had allocated plenty of time to do the videos. Don’t expect to be able to turn around a video in twenty-four hours unless you are willing to accept a lot of compromises or the images or b-roll is not too complex.

Video Editing and Latest Technology

What should have taken maybe five hours of editing so far has taken two days. Sometimes I had to do scenes over and over again because things kept disappearing, sometimes in scenes that I wasn’t even working on. When you are using bleeding edge technology you expect some issues but yesterday and this morning I was having too many. On top of that, ChatGPT over the last two weeks has been getting increasingly slow and today I was getting mostly timeout errors. Gemini on the other hand, I wasn’t having a problem with. I really like projects and GPTs in ChatGPT, otherwise I would probably jump ship. Sometimes tools fall over due to the weight of their success. Other times tools fall over because they are making too many changes. When you embrace new technology, make sure you have fallback plans when things take an unexpected turn.

Publishing

When I went to upload my video, VidIQ was not a fan of my title. The score wasn’t good. I didn’t like any of the titles VidIQ gave me, so I ran the script through ChatGPT and Gemini. Eventually, through a lot of research, I ended up with the title: “Understanding AI in 2025 LLMs, Hallucinations, and What Comes Next”. Of course then I needed a thumbnail to go with it. I didn’t like any of the thumbnails or suggestions I was getting and decided to go my own way. That is the crazy thumbnail you currently see with a TV screen glitching, reflecting the AI Hallucination scene in the video and in the title. I was watching videos for days on how to generate a thumbnail. I think it is a little dark, but that is the direction I decided to go with. I also asked Gemini what would be a good time to publish my video and it came up with 6PM, so that is why this video (and the next few) are scheduled for 6PM. Once I see when my audience watches my videos, I will adjust the time.

Let’s Continue the Conversation

I hope you’re finding these behind-the-scenes conversations as interesting as I am. I know most creators wait until they have dozens of videos before pulling back the curtain, but I want to share this journey from day one. Are these posts helpful for you? What other behind-the-scenes topics would you like to discuss? Thanks for being part of this conversation. I’d love to hear about your own experiences in the comments—what challenges have you faced when starting your content creation journey?

The Youtube video is available here: https://youtu.be/w5tU5AWoN3Y