oneearedrabbit
Joined 17 years ago
Karma
187
Submissions
69
Member since
Jan 28
Recent submissions
Comment · 2 months ago
Comment · 8 months ago
Cold turkey. More than seven years ago, on my way home, I stopped at T-Mobile and bought a random flip phone. I still own that same phone. https://news.ycombinator.com/item?id=21929194
Just a few minor updates to my original post:
- Low-res photos are perfectly fine for my purposes
- I don't need Google Maps when I'm on the go
- Quick fact checking is not important
- One day, I ripped off the band aid and migrated all auth keys to Yubikey
- Tethering is occasionally useful
Comment · 1 year ago
You are not a slow thinker. You are a deliberate thinker. "Slow" suggests laziness; while "deliberate" implies careful consideration.
Comment · 1 year ago
Decades ago, a closure in Forth was especially innovative with DOES>:
: COUNTER
CREATE ,
DOES> DUP 1 SWAP +! @ ;
0 COUNTER PK
PK . \ => 1
PK . \ => 2
const counter = init => {
let x = init;
return () => { x += 1; return x; };
};
const pk = counter(0);
console.log(pk()); // => 1
console.log(pk()); // => 2Comment · 2 years ago
This is curious. It took approximately 19 hours to turn a HN thread into an article. What does this tell about the speed of information?
Comment · 2 years ago
Tangentially related, since you are exploring hexagonal and triangular grids; I once caught a space-filling curve bug and decided to port ideas from https://tixy.land to fractional dimensions: https://oneearedrabbit.github.io/l-sys-playground/curves/ind.... After all, why not?
Comment · 2 years ago
While both MJ and Turnstyle are Turing-complete, Turnstyle shares more similarities with Piet (mentioned by the author), Wireworld, and, of course, Conway's Game of Life.
Comment · 2 years ago
Neat trick. I found that I had to wrestle with Notes more than necessary, but if I were to pick up the project again, I would approach it differently this time:
- Notes function similarly to tweets, and, by definition, cannot by customized much. They are like cute little memos or a river of news where less is more.
- I thought it would also be cool, if within Notes I could subscribe to memos from my friends. Whoa, Apple Notes as a social network! As you are already familiar with Apple Notes structure, it won't be a surprise that it is not difficult to add content to special folders automatically (I built a prototype, which sadly never published.) However, this topic raises a bunch of important questions, and I don't remember whether I found all the answers. It's been a while.
Comment · 2 years ago
Two years ago I got carried away and built a tool with a code name "notespub". It is the very same idea, albeit less polished. I documented my journey of extracting content out of Apple Notes, which naturally has been published with Notes as well [1].
I thought this was a cute idea; however, as I started using it more, I inevitably learned that there are features that are hard to mingle in Notes alone, i.e. custom themes, metadata, automatically generated home page, top/side navigation, cross-references, etc. These features could be implemented with an external system that helps you to manage a blog, yet that ruined the idea that I wanted a tool that puts local & privacy above everything else, a tool that has as little dependencies as possible.
Long story short, I started working on the second iteration that I thought wouldn't sacrifice privacy, but eventually shelved and haven't completed it. I guess it is time to open-source it.
[1] https://devlog.notespub.com/2022/08/site-generator-for-apple...
Comment · 3 years ago
Fascinating. These rules are identical, but I don't have xfractint locally to reproduce/troubleshoot the issue. I think I was using https://dmitrykandalov.com/lsystem/ as a playground, and it appears to be fine there.
Can you try generating 1/8 or 1/4 curves to check the partial generation? Or these rules--should produce the triangular shape:
A=AFFFB-F-FB+F+A
B=B+F+AF-F-AFFFBComment · 3 years ago
That sounds super interesting. I would like to learn more if you're willing to share.
Comment · 3 years ago
Not long ago, I went down the rabbit hole of space-filling curves, and learned about an obscure paper by Rolf Niedermeier, Klaus Reinhardt, and Peter Sanders that introduced a rather peculiar curve with an unfortunate name: H-curve [1]. The paper mentions that H-curve preserves better locality properties compared to Hilbert curve. It fills the space with H-like shapes, hence the name. Also, like the Moore curve, it generates a loop.
Space-filling curves are ridiculously easy to implement with L-system rules, and I spent a few days developing a set of axioms to express it in a rewrite system. This was a fun puzzle [2].
A => BF-F-BFFFC-F-FC+F+BF-F-BFFFC-F-FC
B => BFFFC-F-FC+F+B
C => C+F+BF-F-BFFFC
Comment · 3 years ago
Not directly related to the post, but still feels somewhat relevant.
Back in March I used a bit more elaborate multi-step prompts for GPT3.5 to generate amusing pictures and published a gallery [1]. However, I eventually reached a point where changing prompts did not consistently improve the final results. At the end of the day, the quality of images are only as good as the training dataset, and GPT is a black box.
For something different, to test whether it is possible to "compress" visual content specifically for GPT, I ran another experiment. SVG, being a verbose format, takes time to generate a detailed image, and it also becomes expensive over time. I translated a subset of SVG elements into Forth words [2], which has a nice synergy with GPT tokens--this allowed me to progressively render pictures and produce smaller outputs without sacrificing much in quality.
Finally, I training my own GPT2-like model on the QuickDraw dataset [3]. It's not surprising that a sequence transformer can be trained to produce coherent brush strokes and recognizable images as long as there is a way to translate a graphical content into a sequence of tokens. That said, I found myself with more questions than I started, and trying other ideas now.
[1] https://drawmeasheep.net/pages/about.html
Comment · 3 years ago
This number is tokenized as a list: 43 20 59 83 409 58 340 9 58 340 95 30 95 809 34 850 9 34 850 34 809 58 340 9 58 30 49 850 385 30 49 58 30. If GPT recognizes the context "+ 1 equals" through the attention mechanism, it can predict that the next number in the sequence should be 31: ... 58 30 -> ... 58 31
https://kruzenshtern.org https://twitter.com/oneearedrabbit