6 y/o just spelled wrestle “resl” and honestly I think we have the makings of a new compression algorithm on our hands
6 replies
@zachleat Maybe he can fix Neural Link ridiculous challenge of compressing brain signals 200 fold, in real time, in an embedded low power device https://content.neuralink.com/compression-challenge/README.html
To put this request into context: The current leader board states ZIP as the best algorithm they know. And its compression ratio is at only 2.2 🫠
@vsaw @zachleat man I hadn’t read the page, just heard about it, and I wonder why they believe they need that resolution in time and bits (especially in time, 20k seems silly). I’m hardly a domain expert but the sEEG and ECog data colleagues I worked with used was sampled at like, 1 to 2k roundabout, and that was good enough for speech processes
(I was also going to say “and why not lossy compress optimized for downstream tasks” but I guess optimizing such compression would be pretty complicated)
@zachleat sl would actually be a pretty cool new spelling (consonant cluster?) in English, I think.
Seems like it replaces "-stle": hustle → husl, castle → casl
…although on second thought it still looks kind of cursed.
@kisaragi_hiu fits right in with the husl and busl of life
@zachleat as much as I know about shorthands, this is exactly how they work: only write spoken form of the syllables* (and many even omit vowels).
for example, in Gregg, "wrestle" looks the same as "rest" (which is written as `r` + `s`) + `l` !
*: that's why every language has their own shorthand. usually a popular shorthand is adapted to conform the given language's pronounciation
sources:
- https://greggdict.rliu.dev/ for word lookup
- https://www.reddit.com/r/greggshorthand/comments/yyynjk/gregg_simplified_alphabet_card_courtesy_of_the/ for alphabet reference@pcdevil fascinating!