Posts » How I use LLMs
Here’s a follow up to my earlier post about LLMs.
Since last time I’ve started using Claude in a fairly regular manner, to the point I even have Claude Android app, which is telling to those who know me well.
So here is how I use LLMs these days:
- First of all, nothing you find on this site is written by LLMs, which you can probably tell by my rather sloppy English ;-).
- I use Claude to brainstorm ideas, research more complex topics I cannot find a Wikipedia article or webpage about. I also use it as “limited trust” information source in software development, when I don’t feel like digging through the documentation. I also find it fairly good at discussing philosophical or gardening topics. I don’t trust it fully as it tried to sell me completely false claims couple of times already.
- I don’t use Claude Code as I like my code written manually. Claude sometimes suggests a specific approach, or gives me a working piece of code, which I take and adapt to my needs. Also I don’t like sharing my entire codebase with a remote resource. If I had a similarly powerful local LLM then I could give it a try. Finally, I don’t believe writing more code faster is the right way to go forward - LLMs will quite happiliy generate tens of thousand lines of code if you ask them, or build features just because they can. And more is rarely better.
- We use some open source models at work, quite successfully in some areas but also fairly miserably in others. It’s been hit or miss really.
That’s it. LLMs are a tool we can leverage to do things more efficiently, but also misused to put even bigger strain on our planet and the society.
As they say: with greater power comes bigger responsibility.
I’m doing my best to use it wisely.