@small_LM

ML research, equilibre technologies slightly filtered stream of consciousness

 Prague, Czech Republic
ryanxu.net
Joined January 2022
  • Tweets274
  • Following196
  • Followers358
28Media
ryan retweeted
ryan
@small_LMOct 25
@NASA can I name it? I have some ideas
2
6
141
Replying to @small_LM
I don't think the world can handle the level of randomness emanating from this account.
1
1
3
2,067
You’re all prompting wrong
128
587
252
8,303
1,060,634
"Co-Authored-By: Claude Opus 5.5" is giving me far too much credit.
95
391
25
9,503
206,821
ryan retweeted
ryan
@small_LMOct 11
Replying to @AlejandroZarUrd @konoha_ily2 @yoheinakajima
I have some numbers for up to n=70 which I'll list here - didn't post them because I wasn't sure if anyone was even keeping track. Will update my repo later tonight! 65: 4.7071173296 66: 4.7071171288 67: 4.7071215013 68: 4.7071153112 69: 4.7071152968 70: 4.7072715607 LOL they are not even monotonic. That's how unfinished this is.
1
1
4
184
Replying to @small_LM @konoha_ily2 @yoheinakajima
Broo why are you so quick I haven’t even made a post yet!
1
1
2
187
グレープフルーツってうんめぇ
2
1
9
514
This is what happens when our researchers find out we have some idle compute on our cluster 🤣
ryan
@small_LMOct 9
Packing squares seems to be having its moment, so I took a crack at it... I found 25 new square packing records using evolutionary LLM program search, similar to AlphaEvolve.
2
20
2,294
ryan retweeted
ashe
@ashebytesJul 16
you've gotta be more afraid of wasting your life than embarrassing yourself
301
15,882
272
82,442
1,304,196
Just telling people they slept poorly led to impaired cognitive function. Telling them they slept fine preserved it.
76
257
306
3,604
3,984,869
eating a lot is necessary condition for acquiring a good taste
3
1
16
1,677
We are looking for more amazing researchers and engineers to join our team. EquiLibre Technologies is a frontier AI trading research lab. We train our own foundational models, use our own RL algorithms, and trade our own money. The proof is in the pudding. [THREAD] equilibre.ai techcrunch.com/2026/06/30/th…
Agustin Lebron
@AgustinLebron3Jul 1
I haven't said much recently about what I'm working on, but since this article is out now, I might as well say a few things. ➡️
28
36
9
638
288,163
ryan retweeted
kache
@yacineMTBJun 20
steering vectors are seriously inhumane
8
3
2
139
12,516
String theory is probably a field that has set back quant trading by at least 10 years. Stealing top tier trading talent to do make believe 26 dimensional geometry, inflate boomer professors’s grant budget, and produce zero testable hypotheses or applications after 50 years
AlgoFlows
@algoflowsJun 6
jane street is probably a place that has set back science by at least a 100 years. Stealing top tier research talent and making them do high speed coupon clipping to make rich people richer and also organize military coups in South Sudan.
94
162
26
3,940
300,062
ryan retweeted
m :)
@vuse_loverJun 2
Met a guy recently named Jonathan who goes by Nathan. I didn’t even know you could do that
1,077
16,398
1,562
366,498
9,031,390
Replying to @sama
Thank you for the kind words
50
27
3
6,040
102,439
ChatGPT Images 2.0 (Pro) generates a photo of a cake decorated with SVG that when transcribed to a file renders another cake
ChatGPT dialog screenshot User: Photo of a supermarket sheet cake, upon which is written SVG code for a picture of a different cake. ChatGPT: [Generated image of a cake hand-lettered with SVG code in black frosting]

ALT ChatGPT dialog screenshot User: Photo of a supermarket sheet cake, upon which is written SVG code for a picture of a different cake. ChatGPT: [Generated image of a cake hand-lettered with SVG code in black frosting]

The image produced by rendering the SVG code from the previous image of a realistic cake generated by ChatGPT

ALT The image produced by rendering the SVG code from the previous image of a realistic cake generated by ChatGPT

63
266
53
6,364
447,925
ryan retweeted
pi0us@pi0us0Apr 19
A friend of me who is studying math said: "It's like creating a magic system in a fantasy world. Only there is no story and no characters and no world, just the magic-system."
82
1,618
132
26,758
530,895
(ARC Prize): ARC-AGI-7 is here! This tests how well AI can navigate a 147-dimensional hypercube described in a nonexistent language. Every AI mistake results in one of its GPUs being smashed. Humans score 120% by default. (Researchers): our model just got 83% on ARC-AGI-7
George Morgan
@vr4300Mar 26
Unbelievable result from the @symbolica research team! In just 1 day since the launch of ARC-AGI-3 we have been able to raise the score from 0% to 36%, and within a very close margin to human baseline as well. See our blog post linked in this thread for a detailed score breakdown and replays.
16
42
5
1,020
72,358
job market so bad I'm actually pursuing my dreams
341
10,961
671
115,964
3,076,719
Load more