Sathvik Redrouthu
@_sathvikr
Followers
4K
Following
4K
Media
161
Statuses
2K
Math @caltech. ARC-AGI @trainloop_ai. Previously @etched.
California, USA
Joined November 2022
Gemini 3 Flash across different test-time compute levels (green line below) represents a new score/cost Pareto frontier on ARC-AGI-2. Congrats to @demishassabis and @sundarpichai on the launch!
30
83
1K
Currently, I am: 1. Part-time at @trainloop_ai, working on a novel system for @arcprize with a close MIT friend (more soon) 2. Learning abstract algebra, probability models, and models of human cognition -- especially interested in Josh Tenenbaum's work 3. Solving competition
8
2
108
Hard launching my silly little pure math research: ⭐️ https://t.co/hAVPDY9TfB ⭐️ tldr; in this paper, I solve the admissibility problem on the two element field of Z_2^n, specifically under the discrete case of the Radon transform, by proposing a cycle-tree decomposition
arxiv.org
The admissibility problem in integral geometry asks for which collections of affine subspaces the Radon transform remains injective. In the discrete setting, this becomes a purely combinatorial...
2
1
16
Do I know anyone who has worked/is working in cognitive science? Specifically, how the mind represents causal relationships? I'm interested in studying this more rigorously -- any leads are much appreciated!
1
1
10
As for goal 4 -- building a system to do well on @arcprize / hand-solving hundreds of problems in a compact DSL, doing proof-based math, and reading lots of papers has put me on a good path to having a better understanding of reasoning behaviorally.
Grading myself on first-term college takeaways: 1. 6/10. Studying history unexpectedly gave a new framework for having takes, and my takes are more rigorous due to quality debate. I also have made progress on figuring out what drives me. However, my opinions haven't changed
0
0
0
Grading myself on first-term college takeaways: 1. 6/10. Studying history unexpectedly gave a new framework for having takes, and my takes are more rigorous due to quality debate. I also have made progress on figuring out what drives me. However, my opinions haven't changed
Sure. I am going to college to: - improve my worldview - build as many abilities as possible - do lots of math and learn basic logic/CS theory - understand basics of intelligence To do that I will: - read and engage with students/profs holding opposite viewpoints - focus on the
1
0
17
I'm spending break learning abstract algebra and probability models, while continuing work on ARC-AGI! Would love to meet if you're in the DMV.
First term at @Caltech is officially over! I really admire people here. So far, the most important thing I learned was the importance of paying attention to details and thinking rigorously in day-to-day conversation, which my friends here naturally do. This enables them to make
1
0
18
First term at @Caltech is officially over! I really admire people here. So far, the most important thing I learned was the importance of paying attention to details and thinking rigorously in day-to-day conversation, which my friends here naturally do. This enables them to make
4
2
181
If you love machine learning and plants come say hi !!
@TrainLoop_ai is one of those companies in our group office hour that has something special. They're hitting an exponential curve and are now hiring more cracked engineer! Had a great time onboarding @jackson_stokes and @mlpierce22 to Superset yesterday with @avimakesrobots
1
1
11
Congratulations @GoogleDeepMind! Jumping to 45% is impressive.
Gemini 3 models from @Google @GoogleDeepMind have made a significant 2X SOTA jump on ARC-AGI-2 (Semi-Private Eval) Gemini 3 Pro: 31.11%, $0.81/task Gemini 3 Deep Think (Preview): 45.14%, $77.16/task
0
0
1
Programming does not have to be rigorous and we will be able to code in pseudocode, with a fuzzy interpreter that just "gets it." If humans can synthesize rigorous code for a pseudocode subroutine based on a few examples, then so can a sufficiently intelligent model.
0
0
6
Great tweet Mihika! I'm proud to be your first follower on X. Looking forward to more bangers.
if you’re interviewing for quant, do yourself a favor. make yourself look more chopped. looking too good is lowk an adverse signal for this type of work.
1
0
15
I will write about this and post it here Reviving my personal website soon
0
0
2
Live (alumni? 🎉) wedding happening at Caltech That's a big deal for a class size of ~200 (half a typical high school) Many people have unknowingly walked past their spouse
1
0
12
Making a trip to SF/Palo Alto next week! Will stay for the weekend. Would love to meet anyone who's around.
0
0
3
In industry, your default state is work experience in X, being surrounded by adults who work on X, and a salary. In college, your default state is an idea of work experience in X, being surrounded by kids who have ideas of what its like to work on X, and a negative salary.
yesterday I was on harvard's campus asking students if they've considered dropping out. to my surprise, the overwhelming answer was YES. so why are they still students and not dropping out? when asked "what would make you drop out?" they'd say either "once I know what I want
0
0
5
Intelligence is the derivative of learning.
Learning is the derivative of knowledge.
0
0
5