Another update on my #llm / #ai journey.

While writing my first #luanti mod tonight, I pulled up my local #ollama instance in a browser tab through my fork of `django-ollama`.

I did this because my #vim integration is still misconfigured, I think. More on that, later.

Anyway, llama3 batted about 50% for being useful for my requests for the evening.

First I asked it to protoype the mod, in #lua based on my description.

( If you're curious about the mod specifics, there's an [overview on my site](https://edward.delaporte.us/blog/luanti/garden/). )

I think this kind of request **could** go well later, but llama3 was either hallucinating wildly, or simply writing code for a very outdated version of Luanti.

In fairness, Luanti had a big API update in the last year or so. I will try this again in a few months.

So I read the Luanti core API docs manually to get what I needed.

Later, I asked really basic questions about lua syntax to llama3. [I know a number of languages](https://edward.delaporte.us/code/)
, but lua is not one of them.

This was a great experience. Ollama with llama3 reliably answered my syntax questions, and was probably faster than searching the lua docs, for me, since I am unfamiliar with lua‘s doc pages.

Overall, I can now rescind my complaint from earlier today of no return on my time investment.

My hours of time invested have now saved me a few minutes, and I am well on my way to enjoying the [Automation Curve](https://xkcd.com/1319/).

Joking aside, this really is cool stuff.

@tshirtman

Yes. I think I have something misconfigured.

I have been playing with starcoder2, with `vim-ollama`, this week.

`vim-ollama` is pretty easy to use, but I've seen some evidence that the context may not be delivering to the API properly.

I think my next thing to debug is the `OllamaReview` function in `vim-ollama` pretty clearly skips a groove. I don't think that is the fault of my model, but maybe a misconfiguration.

I will also give qwen-coder a try. Thanks for the tip!

It's about time for another update to my #LinuxMint #microblog

There's not too much to tell. It has been unremarkably stable, in spite of my being...myself (I void warranties).

My latest attempt to break my Linux Mint install has been treating it like a server to deliver the #Ollama API to my work desktop to add spicy autocomplete to #vim.

The experience has been unremarkable, from the "server" side. I set an environment variable for Ollama to allow it to listen for traffic in my local network.

Linux Mint isn't meant to be a server environment, but it works fine as one, for me, anyway.

On the topic of spicy autocomplete (#ai), I have found my experiments to be best summarized as a stupid waste of my time...so far.

But I don't mind. I consider it part of my job to learn this technology stack, in order to advise others.

It isn't paying me back with any of the mythical productivity gains I hear so much about, but there is plenty of room to blame lack of tuning for my abysmal results, so far.

I plan to tune my model choices for my use cases and update here again in a few weeks, most likely with better results - if only because it would be hard to get worse results.

@nixCraft does anyone have a credit for who is in this picture? I can't shake the feeling that I should recognize him, but my memory for names has never been good.

@mkb @Lana @futurebird yes. That feeling gets it me, as well.

I'm a fan of this effect in one weird specific way: I collect what I all "re-retro" devices - like the "NES Mini" - remakes of a nostalgic piece of technology, but replicated with newer technology.

I love all the little trade-off decisions each device has - between convenience and authenticity.

My favorite example is a Commodore 64 clone that shipped with a large metal plate inside - to make it weigh the correct (original) amount, in spite of the electronics taking 1/10 th the space inside.

@mkb @Lana @futurebird @DrHyde

I'm going to have to just agree to disagree on that point.

I would still be rocking my Palm Pilot if I could, and I personally prefer Android and Linux portables.

But as a non-Apple-fan, I won't deny that the particular design and build of the iPhone, itself, had a huge impact on the direction of technology.

@mms no joke, Vim 9 made me (at least temporarily) give up NeoVim.

Vim continues to improve in impressive ways!

VIM 9.2

"Vim now adheres to the XDG Base Directory Specification, using $HOME/.config/vim for user configuration."

There's a lot of cool things here!

https://www.vim.org/vim-9.2-released.php

#vim

@mkb @Lana @futurebird Not to be disagreeable, but I giggled that your two (quite valid) examples both still feel forever young to me.

I still can't look at an Apple II with anything other than awe.

And of course the iPhone was revolutionary.

This study is over 6 months old but it still resonates. AI tools can feel more productive but devs tend to ignore the time spent prompting, waiting for results and refining.

Claude Code took 10 minutes to summarize my unread chats and email today. It would have been faster by hand.

https://www.techradar.com/pro/using-ai-might-actually-slow-down-experienced-devs

@roberto you're spot on about humans writing intent. We see this with infrastructure now. Infrastructure as code describes outcomes, instead of specific steps.

I would feel alarmed by any solution still including AI in the ongoing daily unattended operation.

AI can be a fantastic tool for prototyping, but once we know what we need, a hard-coded reliable, predicable, secure layer is preferred.

For me, AI should only ever sit directly facing me, the human guiding it. Once I'm hands off, I want a predicable steady state solution.

Happy Friday.

In the last week, I’ve seen an uptick in ‘AI is good for boilerplate’ posts. It is 2026. Metaprogramming is over 50 years old. Why are we writing boilerplate at all, much less creating expensive tools that let us write more of it faster?

@nixCraft @radhitya what I saw from it was mostly write-ups about what was learned while building it.

But the NCSA article says it supported MPI, so I'm sure it got used for "small" batches of supercomputer research code, as well.

@nixCraft @tehstu here's NCSA's own article about the PS2 Beowulf cluster.

https://www.ncsa.illinois.edu/2003/05/27/playing-the-supercomputer-game/

#supercomputer

@nixCraft as a staff member shortly after this was built, a hilarious side-effect of this awesome build was that supposedly a bunch of the PS2s came with a branded keyboard and mouse, which of course went straight to University surplus.

So we would see some Playstation branded mouse and keyboard pairs around offices. And the rumor was you could go get an armload more from surplus if you needed them for a lab setup.

Sony's introduction of the PS2 Linux Kit caught the attention of researchers at NCSA. They combined 70 PS2 consoles in 2003 to form a supercomputer, highlighting its ability to perform complex scientific calculations.

credit: https://www.reddit.com/r/ObsoleteSony/comments/1ije8cz/sonys_introduction_of_the_ps2_linux_kit_caught/

#linux #opensource

More than 135,000 OpenClaw instances exposed to internet in latest vibe-coded disaster

https://www.theregister.com/2026/02/09/openclaw_instances_exposed_vibe_code/

> By default, the bot listens on all network interfaces, and many users never change it

#technology #vibecoding #llm #lol #slop

@warrioroflatte I recently explained to my son that "The Mummy Returns" is the best in the series.

Because instead of being about how to deal with the consequences of poor choices with ancient curses, "The Mummy Returns" is about using our past mistakes around ancient curses to mentor and support the next generation through their brand new different mistakes around ancient curses.

I had to stop following the #ai hashtag because so much of what gets shared here about it is, frankly, magical thinking.

Lest I become "someone is wrong on the Internet!" guy, I need to step back from this conversation.

Spoiler for three years from now, when the hype runs out:

Computers are still really neat, *and* computers still a huge pain in ass.