AI Devs are the new "Campfire Guitarist"

2 days ago 1

June 4, 2025

Looking for a solution to run a "simple" llm setup as an economical and local solution, I almost burned my poor Mac Air. Obviously that glorified notebook isn't up to the task. The way it started stuttering made me realize that the doubts I had about even trying this were more than grounded in reality. So what is one to do?

Huggingface to help!

Obviously a cloud solution, as cheap as possible for a proof of concept prototype would make most sense. Knowing the hustle and the SOTA from a year ago, I dreaded the idea. But it couldn't be helped evidently so, off to the marketplace for ideas and great innovation. Right?

"DeepSeek Sites"

I don't know if you heard about the new "groundbreaking innovation" DeepSeek provides, but it's a, and I cite "A way to create any app". Having a few other project ideas, I tried it. Might as well as a prototyping method just work out fine. It did look, let's say promising. Nothing too fancy, just a basic blog and shop setup. Cart worked. Reviews faked. Obviously, as every good WebDev knows nowadays, you have to hardcode -every-single-page- to ensure errors and a fun debugging time. Refactoring is something for those old people. Screw that! Jokes aside, it really was the most anheurysimic way to devlop anything I've ever saw. And I "peer reviewed" code during university. To not ride around on the DeepSeek topic too much, let me summarize my findings:

You can find a ton of code litter. It looks like you'd think a website should LOOK, but isn't anything a website should BE.
Any link you tried to use, any function, button, would not be implemented, leading to effectively more effort and work to be put in.
Don't get me started on the "published web apps" they proudly present (Why would anybody even bother creating a financial webapp in this lackluster and most certainly insecure manner?) Back to the topic at hand: A local llm for 200 token prompts.

Recommended Spaces

Continuing the search you come across the heavily promoted "Huggingface Spaces". Optimistic as I am, I was hoping to find anything worthwhile there. Any new development that could fit my needs maybe? Without bashing anybody personally, I came across one of the top recommended "space" this week. Still being a bit confused about the point of DeepSeek Site I tried to figure out why people put out those, not even half baked products. There has to be a point besides "Cause they can!". Surely. They'll at least promote some hair growth, AI optimized innovation or something. No. The only thing promoted are usually Instagram & Discord. Those lead to (retrospectively not that suprising) ever refreshing and broken webpresences. So basically it's the typical cause of -They can but nobody stops for a second to think about if they should. Not to mention the missing point of "Why?"-

The campfire analogy

Coming to the key reason behind me rambling as the 36th millionth person about "AI BAD": We are at a rather interesting but not unseen devlopment socially in regards to craft. As a guitar player myself, I've put in a few decades of practice to be at least somewhat palatable. At the same time everyone knows the concept of the campfire guitarist (Sorry Ed Sheeran). A person that put in about 1% of the time, effort and blood (You're fingers are pretty roughed up after a few good hours of practice) is, honestly, regarded as a "better" instrumentalist than you. Is there envy speeking on my part? For most guitarists with a good chunk of experience speaking: Yes. It just isn't fair (It's true).

How can it be possible for somebody to barely work as much as one did and have better results? Asking the consumer, they would say that the output from the usual campfire guitarist sounds better than a classical guitarist knowing every nook and cranny of his trusty instrument.

To wrap this up I probably will end this really elegant: