pothos elaben auton

california
Joined December 2019
Ayush Tambde retweeted
Generalization in 2026: we trained a chonkers on the energy consumption of London and it is good in some things
A modern perspective on generalization
4
127
8,358
really really cool stuff from @justinglibert and team. godspeed!
In 12 weeks, we built a research facility that is run entirely by AI. AI designs, executes, and observes experiments end-to-end across biology, chemistry, and materials science. We’re introducing SciUniverse: a benchmark that measures AI’s ability to do real-world scientific research.
1
13
1,036
Ayush Tambde retweeted
thank goodness for weekends. whoever designed them really understands that having some days without any meetings is utterly essential for getting some real focused ic work done
2
6
255
8,990
“how would you like to spend your singularity?”
1
7
551
Ayush Tambde retweeted
i don’t mean this negatively. the anthropocene is coming to a close, the world (as we know it) is ending
5
3
5
44
19,251
it’s so cool that classical kinetic energy can be derived from *merely* galilean invariance applied to the lagrangian
1
1
563
fb449b308f66ff98d440f4ad4b12c49dddf359c26216902b2d9c503a652570c3
3
477
this is sick! @apagajewski is cooking
Introducing Pan-1, a highly capable Minecraft model trained with an RL-based pretraining technique that could unlock internet scale video for robotics models. It can achieve diverse goals—fight mobs, build structures, explore—without training on any of them specifically.
5
863
Ayush Tambde retweeted
Paradigm’s 4th fund $1.2B to invest and build in crypto, AI, robotics, and other areas of the technical frontier
361
146
60
2,526
599,009
Ayush Tambde retweeted
68
501
27
5,312
139,143
fable is such a joy to use. haven’t felt this way about a model since o3
1
1
253
Ayush Tambde retweeted
We love puzzles, so we acquired the secret to one of the most notorious puzzles in cryptographic history. Kryptos is a sculpture at CIA headquarters with an encrypted message on it. The final 97 characters of it have never been solved. Here's what we're doing with the solution.
32
37
9
411
142,772
there should be structures that gradient descent disfavors, eg hard boolean circuits, since gradients don’t flow particularly well through them. would be fascinating to map the space of structures “difficult” for gd
2
231
spsa is underrated
184
Ayush Tambde retweeted
Exclusive: Andromeda started out as a project by AI-focused VC firm NFDG. Now it's a standalone startup with a run rate of $100M and a $1.5B valuation after raising $60M from Paradigm. I spoke to CEO @WMoushey about his plans to build a better market around global AI compute.
8
16
7
112
27,495
i'd be curious if models given free reign in synthetic worlds could derive its laws. e.g. can start by successfully navigating mazes in a locally non-euclidean env and then build up to a full mathematical description
1
7
698
might be interesting to rl on this and see if it leads to a meaningful improvement in physics abilities. (the ideal would be, given 1910-level knowledge, discover general relativity, but seems nontrivial to ablate all modern physics)
1
458
If you do everything...
21
67
14
1,103
176,704
Ayush Tambde retweeted
I found this to be true. I had to keep stacking the 100-hour weeks up, because once I could sense freedom it would break me. I would literally get sick or distracted or unmotivated.
overheard in SF: “people who work 100-hour weeks can't take breaks. they're like deep sea creatures, they work well under pressure, but then as soon as they go up to the surface, they combust”
6
3
1
188
15,815