ทวีตที่ปักหมุด
david
1.2K posts
david
@conundrumer
abstract nonsense enjoyer - leave no shape unrotated - maintaining https://t.co/iKkNUm0NvQ
nyc เข้าร่วม Ekim 2009
458 กำลังติดตาม861 ผู้ติดตาม
@conundrumer @loveofdoing It uses the ground truth of the test part of each sample? Or is it using the ground truth of the train part only?
English
316 ARC-AGI tasks solved with zero learning. No neural net, no training, no DSL — just 19th-century projective geometry.
Encode grid cell relationships as Plücker lines in P³, find transversals via Schubert calculus, score candidates by geometric incidence.
95% solve rate on the eval set (of non-timeout tasks).
Single C file, runs in seconds.
Beff (e/acc)@beffjezos
The masculine urge to try to hack a new solution to ARC-AGI benchmarks
English
@loveofdoing for an actual working demo of using "geometry" I highly recommend checking this out
> 10.8% on ARC-AGI-1-Train and 3.0% on ARC-AGI-1-Eval
arxiv.org/abs/2511.08747
English
@loveofdoing its fundamentally flawed: it depends on using the ground truth and doesn't actually predict grids
I fixed that issue and now it's scoring 23% over 15 of the simplest tasks
English
there were two modes: exhaustive (check literally all possible grids using the same dims and colors) and sampling.
exhaustive was only used for ~15 tasks. sampling was used for the rest.
the way sampling works is:
1. start with the ground truth. the correct grid. score it.
2. randomly generate grids with the same dims and colors and see if any of it scores higher
if it fails to find any that scores higher, then, supposedly, this method correctly determines what should be the correct grid
English
@conundrumer @loveofdoing wdym by depends on using the ground truth?
English
@loveofdoing I am 99% certain your code has a bug. For example, in sampling mode you do not compare proposed solution to correct one. Always check predictions and always check counterexamples. This is just AI slop, and everyone falling for it needs better epistemic heuristics.
English
@loveofdoing You didn't say, but is it ARC AGI 1 or ARC AGI 2?
English
@loveofdoing sorry I'm confused, that makes no sense
how could that possibly work?
does it solve this specific task?
arcprize.org/play?task=08ed…
English
Personal AI can only do its best work if it has as much context about you as possible
I've been working at Littlebird developing algorithms to basically help make sense of all this captured data. Try it out!
alexander green@alexframegreen
English
People have no idea of the conceptual density of every single symbol in this theorem (whose statement, full disclosure, I can't comprehend). This is typical of modern algebraic geometry. Never before had humans packaged that much cognitive load in seemingly innocuous letters.
Rogier Brussee@RogierBrussee
This is rather a beautiful example of modern math with very precise and profound statements (not to mention the great use Litt made of it for proving that certain Taylor expansions have rational coefficients) that look like total gibberish to those outside (and many inside) math.
English