iiinova — where your digital self keeps evolving

DuskyElf @duskyelf.com · 28/08/2026

How are u just so funny T T... I was literally sad and then saw this :3

030

DuskyElf @duskyelf.com · 25/08/2026

how can we rewrite DNS in atproto when atproto depends on DNS ?

131

DuskyElf @duskyelf.com · 23/08/2026

Luna gives it such a vibe xD. Looks really good

111

DuskyElf @duskyelf.com · 23/08/2026

One can write code as a craft and at the same time vibecode experiments that they would have never done otherwise. Like this bluesky real time post tracker that I made, which aggregates hot posts among my followees and their followees...

010

DuskyElf @duskyelf.com · 23/08/2026

Turning on `INTEL_IDLE` in kernel config made a night and day difference in my laptop efficiency. Not all optimizations for the specific hardware are enabled by default in a kernel build.

000

DuskyElf @duskyelf.com · 17/08/2026

Yup, and that's its point. It was intentionally designed to not make it obvious to the user of the Apps built on top of AtProto. Mastodon is just different, it asks the user to choose the server, it's too much for an average user. Just different goals.

030

DuskyElf @duskyelf.com · 17/08/2026

That's Bluesky not AtProto. And they did it intentionally to show an example that the user's side doesn't need to be "compromised" (they made an argument that people didn't care about protocols). There are hundreds of different apps on top of AtProto that showcase cool usage. margin.at, blacksky ...

120

DuskyElf @duskyelf.com · 08/08/2026

Giving a Star is a form of a gratitude as well : D. But I see your point, u want to be able to add a message to the star?

130

DuskyElf @duskyelf.com · 08/08/2026

Technologies have had dark parts to it. We can't call every use of AI evil. I can see power and water exploitation and hardware price rise as a bad thing. It's just sad that we feel so helpless, but striking against using AI is likely not gonna fix anything : ( Poor Hank

010

DuskyElf @duskyelf.com · 08/08/2026

Nice gift for saving maths ><

020

DuskyElf @duskyelf.com · 08/08/2026

that's crazy.. Is that the other side of the milky way ?

010

DuskyElf @duskyelf.com · 16/06/2026

Don't lie, that channel is your brother's

000

DuskyElf @duskyelf.com · 15/06/2026

I have the same as well xDD I love em as well... I think it's called Irish Whistle

110

DuskyElf @duskyelf.com · 11/06/2026

LMAO, for sure we're not ready yet

020

DuskyElf @duskyelf.com · 11/06/2026

The amount of money these AI labs are betting, they'd just keep throwing money at anybody who comes up with a lawsuit against them. Instead of fixing any of it. which is kinda sad as not everybody has the resources for a lawsuit, neither does everybody live in the US

141

DuskyElf @duskyelf.com · 11/06/2026

A mind blowing insight that prediction and compression are mathematically equivalent, and prediction needs some sort of intelligence That means intelligence is somehow deeply related to compression here are more insights from the latest 3b1b video straight from my journal #compsci #maths

claude shannon known as the "father of information theory" as he laid the foundations of information age, interesting thing that, prediction and compression are mathematically equivalent, prefix-free code, is a clever way to compress weighted data based off of frequency, the relation between data size and probabilities is the founding insight of information theory, random noise should be incompressible,, and therefore a perfect compression algorithm's output should be indistinguishable from random noise, `p(any n-bit message) = 1 / 2^n`, => `-log_2(p) = n`, that the number of bits allocated to a message, assuming perfect compression, is `-log_2(p)`, where p is the probability of occurring, `-log_2(p)` is the same as `log_2(1/p)` but more intuitive, entropy h(p), describes the minimum number of bits per symbol necessary to encode a message following this distribution, `h(p) = sigma for i, p_i(-log_2(p_i))`, this is from the noiseless coding theorem by claude shannon's in his paper "a mathematical theory of communication" 1948, which states that no encoding can ever be more efficient than this limit, and it shows that it's always possible to get arbitrarily close to this limit, requires that every symbol follow the same distribution p, but in natural language, the probability of next char is a stochastic process, which works over context, and this is well beyond what exact calculation can give you, that's why claude shannon turned to estimates based on observation, needing to probe at intelligent models of language (humans at that time)

270

DuskyElf @duskyelf.com · 11/06/2026

yup, and there's already hundreds of applications being built on top of ATproto ... like the @tangled.org (github alternative), or @margin.at (bookmarking + public website commenting tool)

190

DuskyElf @duskyelf.com · 11/06/2026

yess, my X account is filled with nothing but bot requests.. btw, did you know that thanks to AT protocol, bluesky's database is all public and can be used to create backend less applications.. here's the last posts you liked (while bluesky's UI doesn't show it, other UIs can)

1120

DuskyElf @duskyelf.com · 11/06/2026

comment down how you feel about this post xD

010

DuskyElf @duskyelf.com · 10/06/2026

Zero JS, Pure CSS ToDo App Walkthrough of the techniques used in a fully functional Todo app built using only HTML and CSS

duskyelf.com

Zero JS, Pure CSS ToDo App

Walkthrough of the techniques used in a fully functional Todo app built using only HTML and CSS

110

DuskyElf @duskyelf.com · 28/05/2026

that's really cool LoL

000

DuskyElf @duskyelf.com · 27/05/2026

I've been doing that, and I tried to use ccache to make it faster, also tried to compile the kernel on github actions. ^ failed in both of that T T

030

DuskyElf @duskyelf.com · 27/05/2026

make sure to run it in an empty directory, and use the /literature-review skill first

000

DuskyElf @duskyelf.com · 27/05/2026

Did you register with ICANN or something ?

110

DuskyElf @duskyelf.com · 26/05/2026

Crazy we can even run these calculators locally even

010

DuskyElf @duskyelf.com · 25/05/2026

I had pretty much the same story and then I found Justin Sung idea is to not write full sentence notes but more visual notes with just keywords and custom syntax for "thinking on paper" And yes, learning only happens when we think about a topic and connect it with other things we already know

000

DuskyElf @duskyelf.com · 25/05/2026

that looks like a local agent connected via discord ? what's the agent?

000

DuskyElf @duskyelf.com · 25/05/2026

who will give us dynamic arrays then :sob:

010

DuskyElf @duskyelf.com · 25/05/2026

at least they be using Nix, <- and users can see all the infra in code. Idk what's gonna happen with global state running random programs

010

DuskyElf @duskyelf.com · 25/05/2026

we need more people using /grill-me and /grill-with-docs skill by Matt

010

DuskyElf @duskyelf.com · 20/05/2026

wha... why? I've no idea what's happening at Github

000

DuskyElf @duskyelf.com · 20/05/2026

I meant, home-manager*

000

DuskyElf @duskyelf.com · 20/05/2026

My issue with Nix is that I couldn't hot-reload my configs without rebuilding home-manage which takes 3 - 5 seconds... That's why I just symlink the hot-reloaded files via Nix

100

DuskyElf @duskyelf.com · 20/05/2026