Rejecting LLMs

Sit Down for the History

I am addicted to technology since I can remember. My parents had the care to manage the vices by delaying to give me free access to technology, I had my first own computer with around 13 years old, and my first smartphone only around 16.

With a computer, I sticked into games. The whole day in the computer, stopping just because of hungry, sleepiness or headache. I was aware that it was a idiotic behavior but I didn't had much discipline to handle.

But then around my 15 years old, with the pressure of my father to make something smart, I initiated my journey on Linux and programming through a expensive yet shallow course about pentesting. It made me like how software works, but without understanding the development I would never get rid of the label of "script kiddie".

So after getting a minimal adaption in that Debian distro (Kali Linux), I decided to get rid of my vice of games: Abandoned the Windows 8.1 and the Kali Linux at VMWare, installing a pure Debian as second boot in my notebook.

After some time I removed Windows from the disk, and the ganes weren't a problem anymore. But then… Linux became my new game.

GNU/Linux, programming, reverse engineering and all stuff also might also be seen as entertainment, as a hobby (like I do). But it has a much steeper learning curve and brings perspective of career. You're invited to quit the world of mere consumers!

LLM might be distraction

Warping time back to now.

One of my values is to avoid distractions. Distractions are hard to handle, pleasing to the moment but packed with a high potential for damage in long-term.

What I see from LLMs is convenience, and the problem, it's held by third-parties.

I am writing this because I entered too deep on it, started to get lazy and avoid reading the codebase to understand the changes deeper. It started me an habit of avoid reasoning. I had enough, this is not the right way to use a tool.

Owned by third-parties

This tool requires a detailed context to be useful, and since this technology is extremely resource expensive, you're likely to depend on a third-party provider. Now you have a intentional spyware in your computer.

Either Enough Teraflops or Read the Docs

I know, it's hard to get rid of it. I'm telling this after using it to deliver a full refactor on a legacy React web app while adapting to the framework in a little more than 3 months at work. LLM indeed helped, but in the end, the satisfaction of finishing wasn't even as expected, and the productivity boost wasn't that much.

Conclusion

Because of privacy and ownership, until I get the enough computation power to run a usable conversational LLM locally, I will use the same strategy I use for Google, YouTube, social media and any kind of threat for my time and privacy: Sacrifice the convenience.

Considering the few situations where LLM really help and don't messes the codebase (like code generation instead editing), I still rather spend some extra minutes reading documentation and learning normally as I always did: Facing frustrations, getting stuck and then get relieved and proud when overcome the problem. This builds a solid experience and makes me more efficient to solve future problems.

This is the process I love, and should never depend on online service, so I can continue working normally without internet. And for sure, this is a prerequisite for my airgapped environment.

For now, I rather solve my problems by my own, search at Kiwix snapshots for solutions and run ripgrep on cloned repos for keywords that brings solutions for my problems.

thisago's blog