Vrushank Desai @vrushankdes Twitter profile | Pikagi

Pikagi

Vrushank Desai

@vrushankdes

836

Followers

287

Following

23

Media

117

Statuses

don’t take life too seriously, nobody gets out alive anyway

https://t.co/km18V58eoe

Joined March 2024

Don't wanna be here? Send us removal request.

Pinned Tweet

@vrushankdes

Vrushank Desai

13 days

Meet Robo! i'm in the very early stages of building an accessible, general-purpose robotics platform, and looking for a few passionate people to help shape it from ground up. if you'd be excited about moving fast on a small, hands-on team I would love to chat -- my DMs are

6

5

62

Last Seen Profiles

@Trevor143336932

@danny_reardon

@stw_pdg

@JoeMinden

@metalisforgirls

@PBMedias

@DanishCvnt

@Leandro12414157

@yamalfan01

@Knutsdotter82

@theagulbra

@Yamach50

@ChristiMacom

@yu_wa_529

@kjetilhope

@alphacentaur1an

@saraymnn

@uvanlig

@StreamGateway

@BinorRaja

@PSDAthletics

@auspornaddict

@amaso_elvi25930

@XOY1N4xjFgkhv2Z

@chiita1108

@YboMarjor

@sotwecom

@hamster_kombat

@infinity_arena

@4h25m

@stwmaniax

@aiahloveclub

@CoachNSanderson

@ibubohay2

@atreon

@PatrickRiot

@vrushankdes

Vrushank Desai

3 months

>be me 🤡 >impulse pre-order AI hardware wearable (Oct '23) >Get an email later that there won't be an Android release, "reply for refund" >reply for refund .... >months pass, no refund >send three follow-up emails, nada >DM the founder on X, crickets >find out they spent $1.8M

Tweet media one

@AviSchiffmann

Avi

3 months

introducing friend. not imaginary. order now at

4K

1K

13K

78

190

8K

@vrushankdes

Vrushank Desai

3 months

@LucianKnock

Tweet media one

Tweet media two

Tweet media three

17

16

2K

@vrushankdes

Vrushank Desai

3 months

. @AviSchiffmann

Tweet media one

2

2

2K

@vrushankdes

Vrushank Desai

5 months

I spent a couple months at the beginning of this year learning about GPU programming through trying to optimize inference for @chichengcc awesome Diffusion Policy paper. I was able to improve inference time for the denoising U-Net by ~3.4x over Pytorch eager mode and ~2.65x over

31

91

607

@vrushankdes

Vrushank Desai

3 months

@AviSchiffmann thanks! not assuming malintent, but others also mentioned they're missing refunds ( @chrisemmett_ , @ignacioaal , @SoloOrTroll ). given the pivot in pricing/product, you should refund $501 to all 2023 buyers & also offer full refunds with an automated form to make this right

1

2

492

@vrushankdes

Vrushank Desai

5 months

GPUs have a memory hierarchy which scales from large capacity, low bandwidth, high latency (global memory) to low capacity, high bandwidth, low latency (thread registers). The part of the chip that does the actual computation (FP32/INT8 units, tensor cores, etc) can do math

3

14

88

@vrushankdes

Vrushank Desai

4 months

@var_epsilon global swe productivity just took a dump

0

0

76

@vrushankdes

Vrushank Desai

5 months

The full blog post series is a lot more detailed and hopefully helpful for anyone interested in GPU kernel optimization! You can find all relevant code for the blog post here: If you’ve got an Nvidia GPU with CUDA & pytorch

Tweet card media

GitHub - vdesai2014/inference-optimization-blog-post

Contribute to vdesai2014/inference-optimization-blog-post development by creating an account on GitHub.

6

3

64

@vrushankdes

Vrushank Desai

5 months

There is a physical reason for why global memory access is so dang slow. Global memory stores bits in ‘DRAM cells’ which consist of a capacitor and a transistor (which gates access to the cap). Each access requires precharging a row-buffer to a neutral voltage, hooking up the row

Tweet media one

1

1

28

@vrushankdes

Vrushank Desai

5 months

There isn’t really any hardware to map to what one might imagine when hearing the term ‘CUDA core’. A core in CPU-land is far more capable than an FP32 ALU, which is what roughly maps to a ‘CUDA core’ in NVIDIA GPU-land. For fun, we can try to guess at how many transistors are

Tweet media one

2

2

19

@vrushankdes

Vrushank Desai

1 month

@ChengleiSi @tatsu_hashimoto @Diyi_Yang @stanfordnlp this is a really cool study! did the LLM idea's get ran past a Google search? i see the novelty scores for LLM idea's are higher but i've personally noticed asking for novel idea's sometimes results in copy-pasta from the LLM from obscure blog posts/research papers

1

1

18

@vrushankdes

Vrushank Desai

5 months

The GPUs thread registers, L1/L2 memory stores in contrast are stored with SRAM. SRAM cells consist of a 6-transistor flip-flop circuit, and because the only capacitance involved here is of the transistor gate, data can be accessed much more quickly. The downside of SRAM is it’s

Tweet media one

1

0

16

@vrushankdes

Vrushank Desai

5 months

oops, wrong screenshot for this post, the right one is attached here.

Tweet media one

1

1

15

@vrushankdes

Vrushank Desai

5 months

@yacineMTB Pybullet has a rlly gentle learning curve and easy to use API

1

0

13

@vrushankdes

Vrushank Desai

3 months

@teortaxesTex @yeswondwerr he's still holding money/ignorning many people in the thread whose emails he just happened to not notice. this dude is next level lmao

@SoloOrTroll

SOT

3 months

@vrushankdes @AviSchiffmann @chrisemmett_ @ignacioaal 4 emails and a Twitter DM and still no refund. Please get this sorted @AviSchiffmann 🙏 Thanks for the bump @vrushankdes

Tweet media one

2

1

282

0

0

13

@vrushankdes

Vrushank Desai

5 months

@karpathy thanks Andrej! totally unrelated but I used to be on Battery Design and work out of Deer Creek. I remember recognizing you in the office on a few weekends but never got around to saying hi. its really cool to interact with you over Twitter many years later :D

0

0

11

@vrushankdes

Vrushank Desai

5 months

@TranBaoChi7 @chichengcc This series is great! I read each section, tried to write the code myself for 15-20 mins, then peeked if I couldn't. After that you can go try to speed up your favorite ML paper. Hopefully my series is helpful with the latter part

Tweet card media

How to Optimize a CUDA Matmul Kernel for cuBLAS-like Performance: a Worklog

In this post, I’ll iteratively optimize an implementation of matrix multiplication written in CUDA.My goal is not to build a cuBLAS replacement, but to deepl...

1

2

9

@vrushankdes

Vrushank Desai

5 months

@fleetwood___ almost thought I had a bug in my visualization script the first time I made this but nope. thats was 300cycles of memory access latency compared to 1 cycle of tensor core latency looks like lol

0

0

7

@vrushankdes

Vrushank Desai

29 days

connectionists in 2010 - 'these symbolic AI people are silly, how do they expect to write down all the rules for reasoning' connectionists in 2024 - 'quick, hire 1000 Phd's to...write down all the rules for reasoning'

Tweet media one

Tweet media two

Tweet media three

0

1

6

@vrushankdes

Vrushank Desai

2 months

@jokrvivek @realGeorgeHotz imo the self-driving car analogy breaks down w.r.t manipulation. robots have to intentionally change physical world state. with self-driving you can abstract away any tire/road interactions and explicitly avoid changing world state with the car body

1

0

5

@vrushankdes

Vrushank Desai

2 months

@chris_j_paxton Tangential to your original comment but I think their scaling efforts have a very different profile than big LLM labs. They have tremendous inference time constraints (100W of 2018 compute) so can't have very many params at all. Wish they published every now and then, I'm sure

1

0

6

@vrushankdes

Vrushank Desai

3 months

paging @AviSchiffmann , looks like there's (atleast) four more weird emails you might have missed

@SoloOrTroll

SOT

3 months

@vrushankdes @AviSchiffmann @chrisemmett_ @ignacioaal 4 emails and a Twitter DM and still no refund. Please get this sorted @AviSchiffmann 🙏 Thanks for the bump @vrushankdes

Tweet media one

2

1

282

0

0

6

@vrushankdes

Vrushank Desai

2 months

@deftech_n @_aidan_clark_ this point is underrated! self-driving car companies have to fight a borderline step-function in the value vs. success rate curve, with 0 value until many 9s of reliability. the curve is gentler for household/serbice robots

0

0

5

@vrushankdes

Vrushank Desai

1 month

+1, the 'China just copies American tech' narrative is absolutely cope. I've found Chinese suppliers to be incredibly competent, offering manufacturing tech/process solutions that are just better than domestic counterparts. America has a lot of catching up to do - acceptance is

@RnaudBertrand

Arnaud Bertrand

1 month

Yet another exceedingly bad China take by Noah Smith. I remain genuinely puzzled how anyone still gives the light of day to someone so consistently wrong 🤷 Here's why he's wrong here on Tesla in China 🧵

Tweet media one

95

442

2K

1

1

5

@vrushankdes

Vrushank Desai

7 months

hello world, it me

0

0

4

@vrushankdes

Vrushank Desai

2 months

is the price gouging in the room w us right now?

Tweet media one

@thehill

The Hill

2 months

Kamala Harris to call for federal ban on price-gouging in Friday speech

2K

371

3K

0

0

4

@vrushankdes

Vrushank Desai

2 months

pip install 'torque dense stator/rotor/geartrain, safe and energy dense li-on battery, power efficient inference compute, high stiffness structures to bring it all together, with a factory to build this reliably at scale'

@keerthanpg

Keerthana Gopalakrishnan

2 months

All serious robotics companies are foundation model companies. Acting intelligently in the physical space is an emergent property of a large audio visual language agent, and intelligence at the most fundamental level is the same whether its expression is digital or physical.

13

22

320

0

0

4

@vrushankdes

Vrushank Desai

29 days

damn, new simulation theory just dropped (we're all stuck in a 1X world model roll-out from 2040 which was deployed to eval how well Neo can stack blocks)

@1x_tech

1X

29 days

hello, world model

29

145

962

0

0

4

@vrushankdes

Vrushank Desai

12 days

@eigenrobot nooo, missed opportunity

1

0

4

@vrushankdes

Vrushank Desai

9 days

@Dan_Jeffries1 it's surprisingly difficult. there was an OSS project called OpenWorm that tried to simulate C. Elegans (worm with 302 biological neurons) and it kinda flopped

1

0

3

@vrushankdes

Vrushank Desai

3 months

@ericjang11 its also here for the impatient - shoutout vrtnis on the K-Scale Discord for the find

Tweet card media

23598 The 5th Annual Embodied AI Workshop

www.youtube.com

2

0

3

@vrushankdes

Vrushank Desai

2 months

@EricSteinb Congrats to you and the team!! You guys have made insane progress in the last year

0

0

3

@vrushankdes

Vrushank Desai

27 days

Hall Effect Is All You Need 🙏

@Raunaqmb

Raunaq Bhirangi

27 days

The sense of touch is fundamental to how we interact with the world. But the most exciting developments in robotics continue to focus primarily on vision. I spent the last four years trying to understand why. And we might have found a pretty good fix. Introducing AnySkin

11

66

443

0

0

3

@vrushankdes

Vrushank Desai

5 months

@bayeslord idk but an interesting clue is that neurons don't stop firing under anesthesia (they just slow down to a dull pattern), and they actually increase in firing frequency during epileptic seizures. both cases result in loss of consciousness/loss of information complexity in the

1

0

2

@vrushankdes

Vrushank Desai

29 days

@xlr8harder yea idt this was just the li-on, people have held Note 7s that went into runaway and been mostly OK

Tweet card media

Galaxy Note 7 explosion caught on camera

Jukin Media Verified (Original)* For licensing / permission to use: Contact - licensing(at)jukinmediadotcom

www.youtube.com

0

0

3

@vrushankdes

Vrushank Desai

27 days

@Raunaqmb this is so neat, awesome work! did you find particles embedded in the skin to be necessary or could a handful of discrete round magnets result in good signal as well? just thinking the latter could be more consistent part-to-part if they were positioned by the mold cavity

0

0

3

@vrushankdes

Vrushank Desai

19 days

vibratory bowl feeder's are so fking elegant and underrated. probably hundreds of billions in material wealth unlocked globally from these things humming along 24/7

@FenixAmmunition

Feni𝕏 Ammunition

@FenixAmmunition

20 days

Sorting brass by height - .380, 9x19, and 9x18 all into different bins, all automated. Nested and flat cases go back into the collator for another pass. Simple and effective.

37

32

862

0

0

3

@vrushankdes

Vrushank Desai

2 months

@raghavaupp Unfortunately I'm smooth brained and still stuck in XY plane

Tweet media one

1

0

2

@vrushankdes

Vrushank Desai

2 months

@raghavaupp this thing's been sweet for me, fast access + takes advantage of Z-dimension

HORUSDY Wall Mounted Storage Bins Parts Rack 30PC Organizer Garage Plastic Shop Tool for Men's...

HORUSDY Wall Mounted Storage Bins Parts Rack 30PC Organizer Garage Plastic Shop Tool for Men's Gift, Blue,Orange,Red

1

0

2

@vrushankdes

Vrushank Desai

3 months

@asoare159 also this is the Florida in me, but looks a lot like hurricane spaghetti models lol

Tweet media one

0

0

2

@vrushankdes

Vrushank Desai

1 month

we can make machines that crush Leetcode Hard's but don't have ones that can pick up a strawberry in unstructured domains reliably. economic consequences if the gap holds are pretty grim. robotics needs to pick up the pace!

@satyanutella_

Satya Nutella

1 month

i tested o1 in the recent weekly LeetCode contest. white table is o1-preview results which completed the whole contest in 3 mins.

Tweet media one

1

3

160

1

0

2

@vrushankdes

Vrushank Desai

3 months

@asoare159 really cool viz! could it be the training dataset has strong biases towards certain modes? maybe k-NN could be used to visualize trajectories in the training set for a given proprioceptive + T keypoint state

2

0

2

@vrushankdes

Vrushank Desai

5 months

@simonkalouche i think a fair argument for making the hardware as anthromorphic as possible, is that it makes scaling e2e imitation learning easier. if scaled sim2real + RL works, fancier embodiments become feasible. i was pretty bearish on the latter approach but Deepmind's recent work is🤯

1

0

2

@vrushankdes

Vrushank Desai

1 month

@Fennex628 Delta Line makes very torque dense frameless BLDCs, haven't heard back from their sales team on pricing yet but probably steep

1

0

2

@vrushankdes

Vrushank Desai

2 months

@BartronPolygon @ROBOTIS Maybe worth opening the motor up to check out the drivetrain, I had an AX-12A that would randomly overload and it turned out to be a mechanical issue between output shaft and plastic housing. Warning: you may get mad when you see how cheap all the internal components are and how

1

0

2

@vrushankdes

Vrushank Desai

12 days

@samyog_dhital no neural nets here (except for my own). demo is tele-operated by me!

0

0

2

@vrushankdes

Vrushank Desai

19 days

@LalRitesh @corbtt @chamath @perplexity_ai do you have custom instructions on or is that just perplexity's default personality?

1

0

1

@vrushankdes

Vrushank Desai

5 months

@yacineMTB no lie half the reason I switched careers was because I got tired of clicking around shitty UIs. parametric CAD gets pretty hard to iterate with if you're tearing up designs every other day im going back to being a mech e once they get neuralink figured out

1

0

1

@vrushankdes

Vrushank Desai

1 month

@nlevnaut @Xilo_K Heard good things about Langmuir Install is kind of a pain tho, need to pour concrete for the base

0

0

1

@vrushankdes

Vrushank Desai

2 months

@christinexye

Tweet media one

0

0

1

@vrushankdes

Vrushank Desai

1 month

AGI cancelled rip

@arcprize

ARC Prize

1 month

We put OpenAI o1 to the test against ARC Prize. Results: both o1 models beat GPT-4o. And o1-preview is on par with Claude 3.5 Sonnet. Can chain-of-thought scale to AGI? What explains o1's modest scores on ARC-AGI? Our notes:

Tweet media one

46

155

877

0

0

1

@vrushankdes

Vrushank Desai

1 month

@ericjang11 well put! lots of companies are going to find the Universal Robots API to be an extremely leaky abstraction :P

1

0

1

@vrushankdes

Vrushank Desai

3 months

@Liu_Zeyi_ awesome work! the taping task with randomized hook/loop velcro placement is really clever. re: robot noise causing distribution shift, is this because of motor vibration transferring to gripper? could be interesting to try mechanically isolating the gripper from the robot with

1

0

1

@vrushankdes

Vrushank Desai

5 months

@MechovaVila @bayeslord i know my personal experience with general anesthesia was kind of reality bending. its not like sleep at all, its closer to 'nothingness'. one second youre in the OR, the next in the recovery room. i imagine that middle period is what death is like, a whole bunch of nothingness

1

0

0

@vrushankdes

Vrushank Desai

1 month

@benjamin_bolte positive-sum ftw!

0

0

1

@vrushankdes

Vrushank Desai

7 days

Lol hurricane's expend Hiroshima nuke levels of energy every second. We could throw the entire global thermonuclear arsenal at Milton and it wouldn't even blink. Conspiracy theorists vastly overestimate our ape powers

@BGatesIsaPyscho

Concerned Citizen

@BGatesIsaPyscho

7 days

“Congress is now confirming that Hurricane Helene, did have US Government Weather Weapons Planes in it” Insane.

2K

24K

56K

0

0

1

@vrushankdes

Vrushank Desai

9 days

@___Harald___ why would you do that when you can go asset-light and outsource the hard stuff to other countries?

1

0

1

@vrushankdes

Vrushank Desai

5 months

/s

0

0

1

@vrushankdes

Vrushank Desai

1 month

1960s colorized LA almost looks more futuristic than LA in 2024... progress in the physical world has been glacial, wen ubiquitious flying cars and robots?

Tweet card media

Los Angeles 1960s in color [60fps, Remastered] w/sound design added

I colorized, restored and created a sound design for this video of Los Angeles 1960s, have population: 2,479,015, surpassing Philadelphia as third in the nat...

www.youtube.com

@sighswoon

sigh swoon

1 month

if you stay off your phone, you'll realize it's still 2005 outside

1K

18K

245K

0

0

1

@vrushankdes

Vrushank Desai

5 months

@jowyang suspiciously lucky

0

0

1

@vrushankdes

Vrushank Desai

9 days

@atakanaltok @___Harald___ haha i was being sarcastic. i agree it's important to build the machine that builds the machine, VCs disagree

1

0

1

@vrushankdes

Vrushank Desai

1 month

@ericjang11 (no shade against UR, great engineering on their part, just not the right hardware for truly unstructured deployment)

0

0

1

@vrushankdes

Vrushank Desai

5 months

@zaph0id @chichengcc Yea for sure!

0

0

1

@vrushankdes

Vrushank Desai

2 months

@tszzl we've been in FOOM for a while now

Tweet media one

0

0

1

@vrushankdes

Vrushank Desai

5 months

~10e9 USD of investor money to speed-run ~10e9 years of evolution, math checks out to me

1

0

1

@vrushankdes

Vrushank Desai

4 months

the little flap that could

0

0

1

@vrushankdes

Vrushank Desai

10 days

And just like with airplanes, scaling bros will extrapolate straight lines on a graph only to find the real world is complex and progress never quite holds pace in the ways the straight line would have you believe

Tweet media one

@jam3scampbell

James Campbell

10 days

this is how it feels working with LLM agents lots of scrap work and tinkering and arbitrary failures, but with occasional, awe-inspiring glimpses of promise just like the airline industry, the technology will not only eventually succeed, but succeed to such an overwhelming

Tweet media one

23

37

441

0

0

1

@vrushankdes

Vrushank Desai

5 months

@JozsefSzalma @rickasaurus why did it stop at two? would have loved to have 4 arms tbh

0

0

1