iamkylebalmer
Region: CY
Monday 03 August 2026 14:16:53 GMT
8019
199
32
12
Music
Download
Comments
oceanoinker :
All of Wikipedia is like 20 gb and has way more info than thjs
2026-08-23 01:43:14
2
Ash :
I've definitely met some people whose entire body of knowledge would fit on a floppy disk from the 1990s.
2026-08-03 16:51:25
6
Tux Doing Stuff :
it's more like a shadow.
2026-08-06 00:32:09
8
Eric Orr :
I mean it's an incredibly lossy compression of it, so if we're allowing for lossy compression of otherwise lossless artifacts than we can go even further
2026-08-06 01:28:39
3
Dabski Darbinskn :
if thats how you see it then the same can be said abt any tiny quants as well, where do you draw the line?
2026-08-04 12:48:01
2
photogatito :
a lossy diffuse collection wow
2026-08-24 16:47:40
0
BTCMaison :
If you have a 2tb iPhone can you download and inference the model locally?
2026-08-03 14:28:17
0
Vitali :
Fits, but isn’t accessible. You’d just be hoarding a giant database of vectors, or random numbers. Takes about 20k minimum to actually spin up the model to decode the vectors
2026-08-22 20:41:57
1
user760196677555 :
Not knowledge, training data, in itself it is a predictive model that can predict the correct answer, not actually know it.
2026-08-18 03:36:11
0
Arnav :
Plus add another 2tb for the entire Wikipedia in condense image format.
2026-08-17 14:09:23
0
Mr Obvious :
Never thought of it that way.
2026-08-04 15:06:31
1
d4vi35 :
1.5 TB is around 5000–6000 copies of Encyclopedia Britannica, about 180,000 Britannica-sized volumes (back when they used to print them) or about 2.5 billion ordinary books — the total contents of several national libraries. Part of me is impressed, part of me says that this transformer LLM craze will deflate like a burst balloon as soon as someone figures out how language modelling can be done more efficiently.
2026-08-03 15:13:59
0
ethernal sadness :
ur smoothing out the rough edges
2026-08-13 14:33:30
1
mike_diamonde :
It's gonna get even smaller, where everything fits on the head of a pin on the head of a pin. any research that talks about miniaturization or compression without loss?This is where we're headed
2026-08-03 16:35:44
1
Jjj Nnn :
No
2026-08-06 05:26:44
0
perlindholm956 :
There has to be a way to insert new weights in the model depending on what you want to do.
2026-08-03 20:58:11
0
AI & Automations :
interesting 😎
2026-08-03 21:50:09
0
MeOnLife :
😅
2026-08-18 16:54:44
0
To see more videos from user @iamkylebalmer, please go to the Tikwm
homepage.
