Kevin (@O2_Addict_) header
Kevin (@O2_Addict_) profile picture

optimistic • be well

Boston, MA
Joined October 2022
Photo by Kevin (@O2_Addict_)Photo by Kevin (@O2_Addict_)Photo by Kevin (@O2_Addict_)Photo by Kevin (@O2_Addict_)Photo by Kevin (@O2_Addict_)Photo by Kevin (@O2_Addict_)Photo by Kevin (@O2_Addict_)Photo by Kevin (@O2_Addict_)Photo by Kevin (@O2_Addict_)Photo by Kevin (@O2_Addict_)
Waiting for replication but this is a pretty big deal.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
99
73
10
1,469
76,533
Not sure why people think we’ll be entering a post labor world. Plenty of jobs already don’t require intelligence
2
129
9
168
8
1,664
32,601
yeah i have a gambling addiction. i can’t stop betting on myself
137
6,175
180
44,702
1,110,473
147
773
127
6,982
370,300
This is wild
for the past few months i've been asking our models to paint. opus 5.5 is very skilled at emulating different styles every image here is a python program generated pixel by pixel. there is no image model, and no off-the-shelf art software. instead, it's about 7,500 lines of code using standard libraries to emulate different brush styles. the agents don't use any pictures as reference, instead working only from what they know about each painter
2
76
Kevin retweeted
jev is insane 🤯 here is jev playing smash bros against itself he is controlling all 4 different characters. and literally deciding whats the best move to play against itself all within a fraction of a second i used over 22 million tokens to play this match and it only cost me a couple of cents... jev does not replace gpt6 astra but the possibilities with its instant response time are endless
141
342
104
3,671
316,782
Yo this is CRAZY
I've seen people describe Jev as an "AI if statement". But what if it actually WAS an if statement? Introducing Probably: a programming language powered by Jev: probably-lang.southpolesteve… Jev baked into the language. “feels” asks a question. “match” routes between descriptions. “while” keeps going until something stops feeling true. This is obviously a toy, but it's fun to think about what something like Jev unlocks. Jev makes the decisions, an LLM does the writing, and a little program ties it together.
14
11
2
809
124,276
Kevin retweeted
this and no cell service
43
271
13
4,686
82,804
Kevin retweeted
I started taking dance classes and discovered that a girl’s attitude toward you improves considerably if you spin her really fast. Forget everything you thought was the main driver of attraction. It’s torque.
168
2,113
166
55,012
784,694
Kevin retweeted
engineering is just applied imagination
34
458
31
4,434
108,300
There’s a fruit fly in my apartment
2
61
What a legend
Replying to @Anthony_Bonato
A friend of mine once picked up Conway at the airport when he was visiting his dept. for a few days. He told Conway they needed to go to the baggage claim area to get his luggage. Conway held up a zip-lock bag holding a toothbrush & one pair of undies & said "no need."
94
“I think I’ve been in the top 5% of my age cohort all my life in understanding the power of incentives, and all my life I’ve underestimated it. And never a year passes but I get some surprise that pushes my limit a little farther Never, ever, think about something else when you should be thinking about the power of incentives.” -Charlie Munger
1
1
42
Cool maps 👀
Python can now render a city in rich 3D. This is a 2 km-radius view of Luxembourg, built from LoD2.2 roofs, LiDAR terrain and official orthophoto imagery, and rendered with forge3d. I have been working toward this image for a long time, and I am honestly proud of this one.
5
1,176
Kevin retweeted
"havent seen any ai images in a while"
havent seen any ai images in a while
215
8,768
60
260,661
4,246,651
Turing beer kegs into rockets for science 🚀
Starting to look like a rocket!
1
112
I’ll be thinking about this forever
ngl I've been saying this for years
19
1,516
Since nobody called it out in the comments, this “new way” to show 3D Gaussian splats is a basic 19th century construction, studied by Jakob Steiner and others, called a “sphere inversion” or “Möbius transformation”: en.wikipedia.org/wiki/Invers… It looks good because it is an example of a so-called conformal map, which preserves angles, but distorts scales. However, when you look at something up close, it doesn't really matter what its absolute scale is (whereas angular distortion is far more noticeable). Conformal maps, namely the 16th century Mercator projection (old as it is) remains the projection of choice in Google Maps and other mapping tools, since it is the unique map that (i) preserves angles, and (ii) keeps north pointing "up" everywhere on the globe. The 3DGS scheme shown here is just bumping up the usual approach up one dimension: apply a conformal projection; keep the "up" direction fixed. The specific choice of conformal maps (inversions in the sphere) are in fact the only choice available in 3D, due to Liouville's theorem on conformal mappings. [Orange illustration below is from my PhD thesis on conformal geometry: cs.cmu.edu/~kmcrane/Projects…]
3D Gaussian Splattingの新しい見せ方を作りました。その名も Inversphere. 2Dのリトルプラネットと違い、空間ごと球面反転しています。ガラス玉に入れる従来の見せ方は球で「切り取る」のに対して、これは球の中へ「折り込む」ので360度の景色が無限遠まで立体で入っています。 #3DGS
36
369
15
3,309
231,363