Now that the plans are laid, it’s time to start coding! I’ve taken it upon myself to build everything I can, until I find someone who can put it together in a reasonable way. The good news is that I’ve got lots of ideas.
The bad news, of course, is that coding is hard. Building websites is harder. So, what’s a boy to do? I mean, obviously vibe code it. But how do I pay for all those tokens on a bootstrapped salary? (“Bootstrapping” here refers to startups with no external funding and has nothing to do with standard errors)
The solution: local LLMs. I’ve downloaded a few models, plus LM Studio, and have that running in the background as I type. So, since the models are trained, there is no marginal cost once they’re downloaded. No data centers, no water supply issues. The only thing left to think about is electricity. How do I ensure that I’m keeping a low carbon footprint?
I happen to know a place (our allotment garden) that has more solar power than it needs. In fact, since my tablet finished charging half an hour ago, the battery has kicked up from 80 to 86% full, in spite of my computer being plugged in. And I know you’re wondering about how effective solar power is in rainy Hamburg, and I’ll tell you: it’s a rainy day, and that increase in charge happened through the clouds.
Really, the only downside is that what Gemini or Claude can do in seconds takes my machine minutes. And we’re dealing with older models without internet, so they’re not producing perfect work. However, making these exercises doesn’t take state-of-the-art programming, and I can fill in a lot of the gaps myself. Plus, if it takes 15 minutes to run, I can just tend to the compost, take pictures of bees, or have a coffee. I mean, I’m in the garden!
So, until further notice, we’re Jim’s Data Gym: Vibe coding, without the ecological footprint.