Robert Dugger — Dev Blog

Tag: lessons-learned

  • A side effect of a side effect

    In 2016 I uploaded a review of Hybrid Animals to YouTube. It got 353 views. Then I stopped.

    I was in phone sales. Still a bachelor. I liked the idea of being a content creator — it looks cool, it sounds cool — but I knew pretty quickly it wasn’t really for me. Making videos was work, and the work wasn’t interesting enough to justify itself. I wanted to play games. I didn’t want to make content about playing games.

    So I stopped.

    —

    Ten years later I have a YouTube channel publishing daily Shorts across six different games. I didn’t change my mind about content creation. I changed what content creation costs me.

    ContentPipeline records the session, transcribes it with Whisper, identifies the moments worth keeping, generates the captions, assembles the video, and schedules the upload. I play games. The pipeline does the rest. The channel exists because I removed the friction, not because I developed a passion for it.

    The clips are a side effect of playing games. The channel is a side effect of building a pipeline. Any revenue from it would be a side effect of a side effect.

    —

    That framing used to feel like an admission of something. Like I wasn’t taking it seriously enough. Now it feels like the honest version of what most creators won’t say.

    Most YouTube advice is about optimizing for the channel. Titles, thumbnails, posting frequency, audience retention. I don’t think about any of that. I think about whether the pipeline is producing good clips and whether the games are interesting. The channel takes care of itself.

    The goal I’d actually care about is streaming — not because of the audience but because of the engineering challenge. Building a live pipeline, managing the session, making it work technically. That’s the interesting problem. The viewers would be incidental.

    I’m not a content creator. I’m someone who plays games and built infrastructure. The YouTube channel is what the infrastructure produces.

    There’s a version of this where the channel grows, the clips find an audience, and the side effect generates real income alongside the consulting work, the games, and everything else running in parallel. That would be a quiet bonus. A small thread alongside many others.

    I’d rather be the developer who has a YouTube channel than the YouTuber who also codes.

    The pipeline makes that possible. I just play games.

  • Teaching Pong to Play Itself: My First Neural Network Experiment

    Teaching Pong to Play Itself: My First Neural Network Experiment

    Pong is the right choice for a first experiment because it has almost no variables. Two paddles. One ball. If you can’t teach an AI to play Pong, you can’t teach an AI anything.

    I used NEAT — NeuroEvolution of Augmenting Topologies. It doesn’t just adjust weights on a fixed network structure. It evolves the topology itself, starting minimal and adding complexity only when it helps. The training runs headless at 500x real-time speed; a separate visual mode exists purely to verify that what trained actually works. Generation 0: random paddle movement, 0% win rate. Generation 50: 98% win rate, predictive tracking.

    The difference between reacting and anticipating is memory. Standard feedforward networks see the current frame. Recurrent Neural Networks carry memory of previous states — ball velocity, trajectory history. That’s what gives the Gen 50 agent its characteristic quality: it moves to where the ball will be, not where it is. The RNN is what upgrades NEAT from “learns to respond” to “learns to predict.”

    The first training approach was pure ELO. Score points, survive, reproduce. The population converged fast — too fast. By generation 20, every agent played the same way. Safe returns, center positioning. They’d found a local maximum and stopped. No one was discovering anything.

    Novelty search fixed it. Instead of rewarding only performance, you reward uniqueness — points for behaviors the population hasn’t tried. The diversity pressure kept agents exploring. Agents with strange positioning, unusual angles, aggressive strategies started appearing — and some of them turned out to be genuinely superior. The “wrong” strategy was actually better. Pure optimization would never have found it.

    Any system without diversity pressure converges on the same answer. It finds the local maximum and calls it done. That lesson applies well beyond neural networks.

    What didn’t work: high mutation rates to accelerate training. The population collapsed — agents changed faster than they could build on what worked. Every generation erased what the previous one had learned. Slowing it down made the evolution meaningful. Some processes can’t be accelerated without destroying the thing that makes them work.

    This was the first project. Everything since has the same shape: variation, selection, emergence you didn’t design. TurboShells encoded the same loop into turtle genetics. rpgCore formalized it into a composable system. VoidDrift runs it as a drone dispatch loop.

    The Pong agent that discovered a non-obvious return angle at generation 47 is the ancestor of all of it. I just didn’t know that yet.

← rfditservices.com