As promised, Neurivex drawn with spinning circles from one Claude prompt
with 2 circles it's a flat line, at 64 you can read the name, and at 512 the error is 0.5%. each circle is one term of a Fourier series
spinning circles drawing the Twitter bird. each circle is one piece of a Fourier series, and with 49 of them the error drops to 0.12%
one account posts only this format. 727 followers, and two of its reels hit 997K and 716K in the last 3 weeks
next post I'll draw Neurivex the same way with one Claude prompt
Wednesday: google. thursday night: binance. friday open: about $850b gone from us stocks, intel down ~28%. halloween came early this year. who's it coming for next?
Here's the video I promised. made with one prompt in Claude
a 2.4cm ball drops onto a ring that can stretch up to 50%:
- 0.1cm hole: it needs +2300%, so the ball bounces off
- 1cm: +140%, and the ball gets stuck in the hole
- 2cm: +20%, and it goes through
- 4cm: a 5.3cm ball that's 11x heavier. the hole only needs +32%, but the weight smashes the trampolines on the way down
want to make videos like this? comment "PROMPT" and I'll send it to you
A steel ball falls through a hole. 5.5M people watched it
@softshift_s2y on IG has 8,546 followers. glass plates, a marble base, one ball. the hole grows from 0.1cm to 4cm and you keep watching to see what happens
why it works:
- the whole setup is clear in the first second
- one number on screen that changes every few seconds
- glass, chrome and marble look expensive, so people watch twice
a beginner used to need weeks for this: learn Blender, figure out the physics sim, build the scene, wait for the render
my bet is one prompt in Claude gets you there now. testing it today, my version goes up next post
YOUR SEARCH IS ANSWERED BEFORE YOU FINISH READING THE RESULTS PAGE
one query in, 10 clean links out, 290 ms later. here's every step this visualization walks through.
~/web $ search > "best runing shoes for flat feet"
here's the split:
▸ parse: the typo gets fixed first, "runing" → "running"
▸ lookup: an inverted index turns each word into a list of pages, 4.9M hits
▸ expand: synonyms join in (sneakers, trainers), 38.2M matches
▸ fan-out: the query hits 2,048 shards at once, then prunes to 2,400
the part most explanations get wrong: they make it look like the engine reads the web when you hit enter.
it doesn't. nothing is read live. the index was built ahead of time, so search is a lookup, not a crawl.
then the ranker takes over:
▸ relevance, links, freshness, scored per page
▸ strong signals rank up, weak ones drop
▸ 2,400 → top 100 → 10
last check before you see anything: 2 spam pages found, duplicates removed, 10 links served.
Index finds. Ranker sorts. Filter cleans.
9 steps, 0.3 seconds, and you never saw any of it.
What should I take apart next?
@aishortcut_ this is huge for explaining how complex engineering stuff actually works. no hours spent on 3d modeling, just describe the mechanism and get something you can show
Holy sh*t I found a fucking goldmine. One 3D video. 8.5M views.
https://t.co/9FvAcUF7yL on Instagram has 18 posts. One of them, how a dishwasher actually works, has 8.5M views.
The format is simple:
1) Take an object people use every day.
2) Show the inside in 3D: what moves, what heats, what sprays.
3) End the caption with "Comment DISHWASHER" for the guide.
The comment is the trick. Every reply pushes the reel further.
Same format works for any niche:
cooking → how a pressure cooker works
skincare → how sunscreen works
What household object would you want to see inside?
asked opus 5.5 for a mechanical watch in one html file
dial, gear train, balance wheel, an exploded view and even a display back with a spinning rotor. every gear turns at its real ratio
no 3d models. just code
@amananimates Do these play nice with AI coding tools? If I paste one into Claude or Cursor and ask for a variation, does the scroll logic survive, or is it the kind of code models tend to break?