@evinism

aspiring chaotic good. mts @ typesafe. prev swe @ discord, youtube, procore. shh, they don't know i'm on twitter. he/him

evin.dev/
Joined August 2009
  • Tweets3K
  • Following632
  • Followers204
196Media
party anima retweeted
Dan Shipper
@danshipperSep 15
we almost never test new foundation models but we've been testing this for ~a week @every and it's pretty wild. the kind of things that will be obviously indispensible in 6-12 months it doesn't produce words as output, it produces probabilities. so it can efficiently act as a judge in cases where you'd need a Fable-level model—but in our testing was 25x faster and 600x lower priced excellent vibe check by @hammer_mt on @every: every.to/also-true-for-human…
Diogo Almeida
@CompleteSkepticSep 15
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution
65
89
27
1,815
222,162
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution
4,029
8,285
6,819
76,640
40,127,892
capybaras can't understand what those people are doing
painting of two capybaras and a frog in the swamp looking from afar at several witches and an alien summoning an interdimensional creature, in the very background there is a spaceship

ALT painting of two capybaras and a frog in the swamp looking from afar at several witches and an alien summoning an interdimensional creature, in the very background there is a spaceship

2
46
456
6,278
party anima retweeted
Yann LeCun
@ylecun4 Jan 2023
LLMs do *not* capture much of human thought, because most of human thought and all of animal thought is entirely non verbal. The factual, logical, and physical reasoning mistakes that current LLMs make clearly show that they have *not* captured much of human thought.
Michael Nielsen
@michael_nielsen4 Jan 2023
I continue to marvel that predict-the-next-word/token, with enough data and parameters, turns out to capture so much of human thought.
76
106
38
833
431,274
🧵: I discovered this woman, who I call Loab, in April. The AI reproduced her more easily than most celebrities. Her presence is persistent, and she haunts every image she touches. CW: Take a seat. This is a true horror story, and veers sharply macabre.
a portrait photo older white woman with long brown hair holding a misconfigured stuffed animal. she has very severe rosacea below her eyes, in the shape of triangles on her cheeks. she looks slightly 3d rendered. there is a lamp in the background.

ALT a portrait photo older white woman with long brown hair holding a misconfigured stuffed animal. she has very severe rosacea below her eyes, in the shape of triangles on her cheeks. she looks slightly 3d rendered. there is a lamp in the background.

a variant of the woman from the first image on what appears to be an album cover with the word "Loab" on it. She appears to be holding flowers, leaves, or feathers of some kind, on a textured, green backdrop.

ALT a variant of the woman from the first image on what appears to be an album cover with the word "Loab" on it. She appears to be holding flowers, leaves, or feathers of some kind, on a textured, green backdrop.

1,303
9,383
5,621
67,535
party anima retweeted
matt sarnoff@txsector10 Aug 2022
Uh… thanks google? I guess all my science teachers and professors were wrong?
A google search for “avogadros number” did not return the value of the scientific constant, but instead returned the phone number (970) 493-555

ALT A google search for “avogadros number” did not return the value of the scientific constant, but instead returned the phone number (970) 493-555

9
101
6
588
[Matiyasevich 2020]: The below Python program halts if and only if the Riemann hypothesis is false:
from math import gcd h=m=p=0 d=0=f3=n=q=1 while p**2*(m-f0)<f3: d=2*n*d-4*(-1)**n*h n=n+1 g=gcd(n,q) q=n*q/g if g==1: p=p+1 m=0; g=q while g>1: g=g//2; m=m+d h=f0 f0=2*n*h f3=(2*n+3)*f3

ALT from math import gcd h=m=p=0 d=0=f3=n=q=1 while p**2*(m-f0)<f3: d=2*n*d-4*(-1)**n*h n=n+1 g=gcd(n,q) q=n*q/g if g==1: p=p+1 m=0; g=q while g>1: g=g//2; m=m+d h=f0 f0=2*n*h f3=(2*n+3)*f3

35
234
52
2,052
party anima retweeted
Hillel
@hillelogram31 Jul 2022
Abraham Wald's aircraft memos are a statistical masterpiece. And it had nothing to do with survivorship bias. The question he was actually assigned: "what's the probability that N hits will down a plane, *given we have no data on the planes that were shot down*?"
Lorin Hochstein@norootcause31 Jul 2022
The thing I don't like about using this story to explain "survivorship bias" is that Abraham Wald didn't have access to the population of planes that were shot down. He, too, was only examining the survivors!
4
43
5
160
throwback to some truly sicko code i wrote a while ago (this is pure haskell)
38
130
26
837
I can’t understand how this can sound so ridiculous and yet be totally reasonable advice Credit: reddit.com/r/economicsmemes/…
54
686
146
4,653
GNU once again proving again that given a choice between a couple reasonable options, they’ll instead choose a 3rd, substantially worse option
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
8
46
328
Sometimes poetry just falls out of my fingertips in the middle of heated email threads – at Mountain View, CA
We don't know any of these things. In some sense, AI is all about discovering what we actually meant by intelligence, by building what we thought was intelligence, and realizing we were missing something. It's useless to talk about this in such general terms. Moving the goalposts is not an annoyance; it's the very point, much like discovering the theorem you actually wish to prove.

ALT We don't know any of these things. In some sense, AI is all about discovering what we actually meant by intelligence, by building what we thought was intelligence, and realizing we were missing something. It's useless to talk about this in such general terms. Moving the goalposts is not an annoyance; it's the very point, much like discovering the theorem you actually wish to prove.

2
3
25
What are the most elegant/beautiful ideas in ML? Feels like mathematicians & physicists often talk about aesthetics, but we very rarely do. Why?
172
157
54
1,775
The Times needs to give an immediate raise to whoever wrote this headline
362
15,097
1,169
157,859
Tag yourselves,, I am “strongly disagree”
8
23
2
339
party anima retweeted
Jason Wei
@_jasonwei20 Apr 2022
progression of prompt engineering research
19
111
14
1,018
party anima retweeted
hardmaru
@hardmaru15 Apr 2022
Trolley problem solved:
271
11,030
1,149
56,682
Interviewer: Write a bubble sort algorithm. Me: tsplay.dev/N7OPDN
124
436
102
3,847
party anima retweeted
critter
@BecomingCritter6 Apr 2022
Thread of DALL-E AI generated images and their prompts
63
918
415
3,530
huh, just found this ultra-rare Lana Del Rey single in the bin at a record store. i guess she wrote it for a movie soundtrack but, uh... i kinda think she was a weird choice... youtu.be/VeL4c_cXSyQ
142
1,901
547
10,757
Load more