iiinova — where your digital self keeps evolving

jnleser @leser.io · 11/06/2026

Not what happened here, but now I got an imaginary person in my head with such bad takes, not even the most agreeable model will agree with them

040

jnleser @leser.io · 06/06/2026

'BACKROOMS' director Kane Parsons has reportedly invented NixOS. "What's a NixOS?" He is only 20 years old.

020

jnleser @leser.io · 03/06/2026

Even more, attributing consciousness to things that aren't will cause us to waste energy on trying to improve the "conscious experience" of something that is just a program running on a computer. Is there any real reason to not treat this as matrix math and a few nonlinear operations?

030

jnleser @leser.io · 03/06/2026

I disagree with a lot of things in the article, but so far, I have not heard a single reason to think LLMs could be conscious, as in having qualia. I don't see any reason to link high languge capabilities or intelligence with having a conscious experience.

230

jnleser @leser.io · 02/06/2026

To celebrate pride month I will read "The Rust Programming Language".

030

jnleser @leser.io · 30/05/2026

Currently where I'm at, though I switch my opinion on ai at about 5 GHz

031

jnleser @leser.io · 27/05/2026

I was thinking that's too extreme, and then I remembered I literally did for my nixos surface

060

jnleser @leser.io · 27/05/2026

Great post! As someone close to completing his bachelor in informatics, it's also a really weird feeling. These are uncertain waters to navigate, but I think it's still very much worth it.

120

jnleser @leser.io · 23/05/2026

I'm pretty sure there are still a lot that just ban ai usage

030

jnleser @leser.io · 21/05/2026

I think both sides aren't fully honest about what we're dealing with. It's crazy how far llms have come, but we shouldn't conflate intelligence with having a consciousness (qualia). Also, predictions about synthetic data being useful in any context that doesn't have verifyable truth is cope imo

020

jnleser @leser.io · 20/05/2026

And anki is the boring uncle

020

jnleser @leser.io · 20/05/2026

* if powered by green energy sources

020

jnleser @leser.io · 17/05/2026

Pretty fun framing. Too bad a lot of the data is included in the training corpus without consent. :(

020

Reposted by jnleser

Brant Reeves @bireeves95.bsky.social · 17/05/2026

This Needs to Happen…. Idea: a law that says if you’re going to build a data center, you have to also build enough solar panels to power it. We can call it the Bring Your Own Damn Grid Act.

51585312236

jnleser @leser.io · 15/05/2026

Same here, and I ain't talkin bout shai hulud

020

jnleser @leser.io · 12/05/2026

I've been slowly transitioning to zed with vim motions, I kinda like it as a move from vscode

010

jnleser @leser.io · 06/05/2026

LLM's are just stochastic parrots. A database that you can query with language. It knows more than you ever will. Get it twisted. Everyone is gonna lose their job. It can't count the number of r's in strawberry. Get it twisted. The singularity is here. It's running out of data. Get it twisted.

030

jnleser @leser.io · 03/05/2026

Good article, but just to play devils advocate: Why not fine-tune and prompt these values and tendencies into it? Like I myself have written an article about codebases getting larger and less simple with LLM use but I feel like I can critique my own artcile like this too.

120

jnleser @leser.io · 03/05/2026

That's what I've been thinkin

020

jnleser @leser.io · 01/05/2026

We live in the funniest timeline. Like the coolant thing is crazy but even if you're an ai hater you gotta admit that a real artist getting a commission is comedy gold

0110

jnleser @leser.io · 30/04/2026

Interpretability researchers aren't gonna love this one 😬

010

jnleser @leser.io · 29/04/2026

WAOW and MEOW kinda look like flipped versions of each other of you squint hard enough

010

jnleser @leser.io · 27/04/2026

Typst user that plays horse enclosure game too?! Unbelievable levels of based

020

jnleser @leser.io · 22/04/2026

And just imagine the amount of supply-chain attacks

030

jnleser @leser.io · 16/04/2026

Why does this sound like NL posted this?!

050

jnleser @leser.io · 15/04/2026

What's the book called?

120

Reposted by jnleser

Rust Language @rust-lang.org · 08/04/2026

⚠️ An active phishing attack is targeting crate owners by asking them to "confirm that your email address is still active". These messages are not from crates.io, and should be ignored. (We will never ask you to confirm that your e-mail address is still active.) ⚠️

215563

jnleser @leser.io · 08/04/2026

That's really cool! Maybe huge tests like these really can serve as a foundation to make flakes stable 🤔

020

jnleser @leser.io · 06/04/2026

Also bots would mass follow people to get follow backs

020

jnleser @leser.io · 28/03/2026

Yup! It's a real issue in any growing codebase, I just think the intuition of good developers is better at anticipating changes to requirements than current LLMs. It sounds borderline impossible to me to make a training dataset that captures this large scale understanding over evolving systems.

010

jnleser @leser.io · 28/03/2026

I call these things LLMs produce in codebases "Vibe Vestiges", maybe you will find the term fitting :)

020

jnleser @leser.io · 28/03/2026

Giraffes have a vestige called the recurrent laryngeal nerve. It's takes a >4 meter long detour through the entire neck to reach a different part of the head. As giraffes evolved the nerve got more and more wasteful. Vibe coding does the same to your codebase. leser.io/blog/vestigi...

250

jnleser @leser.io · 28/03/2026

Oh that's nice, I love my casio but I almost never wear it because I like the smartwatch data. I think you convinced me to buy a kit ^^

020

jnleser @leser.io · 27/03/2026

Thanks, good to see the community is big enough for lix to exist when regressions are made :)

020

jnleser @leser.io · 26/03/2026

Why lix over nix? Is it just so the alternatives are good if the nix team ever happens to spontaneously combust?

110

jnleser @leser.io · 20/03/2026

So true, untested backups are scaaary

040

jnleser @leser.io · 10/03/2026

It's really good to be aware of what's you give up when you outsource to ai. Kudos

000

jnleser @leser.io · 09/03/2026

That's very nice. Too bad ai code often sneaks in unannounced tho

020

jnleser @leser.io · 08/03/2026

Can we just build more solar please, it's free energy

030

jnleser @leser.io · 08/03/2026

Ya got me good, I thought it was real for a sec

020

jnleser @leser.io · 08/03/2026

Cool! Can it handle banking apps and calls tho?

100

jnleser @leser.io · 04/03/2026

Your heart knows it wants it

020

jnleser @leser.io · 04/03/2026

Ohnooo, I thought it was a really good feature

000

jnleser @leser.io · 03/03/2026

In domains I'm not experienced in, sure, it can produce a basic working artefact faster than me, but I feel as though this tricks you into believing its peaks are higher than your own. If you rely on it, your own ability to reach those peaks will decrease, shifting your perspective even further

010

jnleser @leser.io · 03/03/2026

I wonder if it really is as powerful as you make it out to be. I don't see much evidence of it writing code that is truly hard to write, the kind that isn't near the middle of the bell curve. Sure, it will replace many people, but won't the scaling curve flatten before it reaches top performers?

340

jnleser @leser.io · 01/03/2026

this fine fish fixes his fins

030

jnleser @leser.io · 01/03/2026

10/10 would recommend (it's only been a few days with this setup so maybe take with a grain of salt but tailscale work nicely)

020