OP here! I love your website and didn't realize you also had an inflation adjustment component - I like the idea of providing them inline but my general philosophy here (which seems different from yours in my mind) was to create a single flow very similar to a regular article one would see in any print media but with as many opportunities for proactive additional engagement as possible. I absolutely love your website and the style and voice it has, though it more leans towards the encyclopedic, mind map, interconnected side of things. That's a distinctly different vision than what I was going for here. Thanks for reading!
I have very fond memories of the occasional Monarch I would see growing up (and the lengths I would go to in order to ensure they were not the similarly colored Viceroys). Equally sad is the dramatic decline of songbirds and bird biodiversity in general (though I don't know if it's due to similar destruction of breeding grounds).
Wow! Very pleasantly surprised to see how popular this got! A few short disclaimers and the like answering general comments:
1: The article is a draft but a mostly complete one - I wrote it in two sittings mostly linearly after doing a bunch of research. I intend to revise it to meet my personal standards sometime soon ish
2: About half of the point of this article for me was to present an article in as pleasing, interesting, and useful way as possible, here are the major features of the site to play around with:
- Squiggle line tooltips: inspired from Codrops but then I procedurally generate the squiggles per line and draw them - I see tooltips as a way to answer questions, and the hovercards as a way to inspire curiousity
- Hovercards on the side of the article: Meant to encourage people to enjoy what they're reading - Skim! Pause! Skip! I don't care - reading things like this should be fun and I wanted to encapsulate that, the hover cards are very pleasing in that sense to me
- Background images on hover: There are many times I wanted to include visual aids but said visual aids weren't quite important enough to break the reading flow, that's what those are for! This effect I had created several years ago for my portfolio: ( https://explosion-scratch.github.io/portfolio ) and got to reuse it here
- Money conversions: A fun component idea that I had, converting old monetary values to present ones accounting for inflation (and back again), turned out nicely I think
- Side timelines: I really wish I had these when I would read articles, it helps me skim and brings clarity to what's going on. I put a lot of work into making these work bidirectionally
- The spanner (grasshopper banner across the page): I really like having this banner and can repurpose it with any other images later =) - Try clicking them
- The jumping locusts: This was the most fun part of the site but it seems people either very much like it or very much don't! I intend to make it more subtle in the future, but double clicking spawns a locust, on which you can click to make it jump lol
- All the jokes and cheesy bits of the article: I am not writing a textbook, I'm writing something I personally would want to read
Also some people commenting about AI usage: I didn't use AI for any of the article content itself (the article was entirely written by me, after tons of research, outlining, etc). The article was essentially written linearly after I made a detailed enough outline to go from there. Someone said that the article was created with AI based on the human outline, this is tickling but no– humans write based on prior outlines too! I'm sure the process is different for everyone! The code _was_ aided by AI though I had a very very heavy hand over it and wrote / repurposed (my own work) much code personally. The site itself represents a UI I personally am very proud of, put countless hours of work into, and take responsibility for! This was a case study for me personally to find a productive usage of AI - not for the content of the article, nor the ideas, nor the creativity, but simply as a tool to guide me towards that.
Thanks for reading everyone, and star on GitHub if you get the chance =)
For most of my sources I didn't trace back to the original source but rather just cited the first source, if it ends up being inaccurate let me know and I'll revise it! The main focus of the article is breadth not depth =)
No, it's not AI generated. I wrote the outline then the article. Parts of the code for the site itself (code only) are AI generated but the article I wrote by hand over the course of a day and a half or so! (Referencing my outline)
I know right! It was a super interesting rabbit hole of mine while writing this to discover how boids simulations work, I'm personally very interested in whatever the mathematics of simple rules giving rise to complex and chaotic systems is. It'd be very interesting to do an article about the population dynamics and such of locusts (esp tying into general math).
I love what a gnash equilibrium (I think) it is of their behavior - and really funny, the constant march of canibalism haha
Sorry guys haha!! I intended the insects to come without prompting only on scroll pause but for debug purposes of the article I had set them to once every 15ish seconds[0]. I thought it was really great and love it personally as it really gets the AAA factor in, though I should probably respect use-reduced-motion lol
This is a super cool paper! The only thing I'm a bit skeptical of is the detection of 'slop' to begin with - it would be very exciting to see some sort of way to steer a model more towards human-like output more generally. Here I imagine a bit of a challenge is that you're 1) creating the measurement for 'slop' and then 2) reducing it, so yes, you can have amazing results in the detection of the slop you've found (the methodology may be novel - a slight improvement over DPO), but this is a chicken and egg problem, as each new model would suffer differently based on training, RLHF, etc. It's not a very useful method to improve an LLM's writing quality, but rather just reducing the symptoms of said bad writing.
I'm additionally concerned about removing the LLM's voice to begin with, as though an LLM may far overuse specific words and phrases, so too do individual authors. Reducing an LLM towards the average author would result in an 'average' voice, paradoxically unlike any specific author.
I am currently doing research on something very similar over the summer but more towards the detection as opposed to the generation side - I'd love to discuss this with someone in the field if you had a few minutes sometime!
It would be nice to actually see what the components look like before using them. Currently I can only see the code or examples of how to use them. Also seems like this might be entirely AI generated.
Good idea, thank you. I'll add examples in the documentation. Currently there are example repos that show all the components for each flavor: React, Vue, Svelte, Blazor, Nunjucks, HTML. Angular coming soon.
AI use is mixed: I write everything by hand in the specifications, and the components.tsv file, and in the Svelte version because it's my stack of choice. If you dig in, you'll see lots of hierarchies that are all hand-written to help Svelte caching, for example. I also research other major design systems, especially government-oriented public-sector systems such as GOV.UK and Reuters, and fold them in.
Then Claude Opus tranforms the Svelte version into the other stacks. Claude does a lot of the documentation text because I'm aiming for clear and consistent explanations, suitable for novice developers.
I'm developing Lily Design System because I work with multiple teams that each use a different tech stacks, each with their own ad hoc HTML tag names and semantic names. As a salient example, for a hospital form one team used terminology "health banner area" and another used terminology "medical red box". Lily is my attempt to converge these into something that works better internationally and across multiple stacks.
That said, I'm seeking help doing human proofing and improvements for the stacks, because we all know AI isn't perfect, and needs tuning, guardrails, expert feedback, and the like.
