Saurabh Bhatnagar @analyticsaurabh Twitter profile

Pinned Tweet

Saurabh Bhatnagar

6 years

I just published “How I scaled Machine Learning to a Billion dollars: Strategy”

How I scaled Machine Learning to a Billion dollars: Strategy

Rent The Runway is valued at $700 million and is well on its way to be a unicorn. I joined in 2012 to bring data science to then, a small…

medium.com

1

4

35

Last Seen Profiles

@Fa70938920

@MartinJaliff

@KatHazelton

@FWmaternity

@cigamons

@RepTonyCabral

@tenmakenz0

@cigamons

@RFFLifestyle

@yyates05

@Nescafe_jp

@praew_wkwk

@bluberry_123

@FerrySoda8274

@Shabazz_PBG

@walterzie

@Bee_2929

@secondpartinwho

@cigamons

@sapalldan

@wonrbin

@nonoo9999999

@Xguavaa

@HauntingGG

@iiereui

@truckcoineth

@ratatouille_hmm

@stw_pdg

@HiranoToshi0323

@runawaypacy

@swapniltk

@khxng_plxy

@NeatSoraya67284

@tamapi__

@kitxy29

Saurabh Bhatnagar

@analyticsaurabh

7 months

@chamath It’s a mistake to think these guys want to advance the cause for the general public. It’s a tribal dance to elevate their status within their tribe.

28

23

1K

Saurabh Bhatnagar

@analyticsaurabh

7 months

@elaifresh This is the new Brahmin class - compete to see who is ‘cleaner’, ‘learned’

6

13

720

Saurabh Bhatnagar

@analyticsaurabh

10 months

@flybottlemist “Whoever eats my flesh and drinks my blood has eternal life” - Jesus “My turn” - Jesus Redux

0

27

669

Saurabh Bhatnagar

@analyticsaurabh

2 months

@zanyfen New trolley problem

1

0

625

Saurabh Bhatnagar

@analyticsaurabh

5 months

@anammostarac The seven body problem

10

2

303

Saurabh Bhatnagar

@analyticsaurabh

1 year

@vv1lder @dustinfreeman This tells us more about the human who was looking for any reason to deny the refugee, rather than a shortcoming of AI. Kafka would like a word.

10

4

198

Saurabh Bhatnagar

@analyticsaurabh

1 year

@jeremyphoward Im glad they can spring into action faster on imagined risk, than they did on a real pandemic 🙄

5

3

164

Saurabh Bhatnagar

@analyticsaurabh

1 year

@arghavan_salles @PeterHotez What a waste of his time to even consider this

Saurabh Bhatnagar

@analyticsaurabh

1 year

I don’t know about you, But I like my researchers doing research and advancing human civilization. Not fighting mudslinging on podcasts.

16

6

75

22

8

145

Saurabh Bhatnagar

@analyticsaurabh

3 months

@cosminnegruseri There is money in the banana stand.

0

97

Saurabh Bhatnagar

@analyticsaurabh

1 year

@paulg He might actually understand git

2

0

87

Saurabh Bhatnagar

@analyticsaurabh

7 months

@NewPlagueNewMe @elaifresh Yes!! That’s the syndrome. Midwit meme strikes again

0

84

Saurabh Bhatnagar

@analyticsaurabh

11 months

@steviemctweets As an ML engineer, “AI” is going to be commodity. All that matters is who the end user trusts. GPT wrappers that deliver a trusted product is a moat.

4

7

76

Saurabh Bhatnagar

@analyticsaurabh

3 months

@nikitabier This was even true for books when I worked at B&N Women accounted for 70%-80% revenue Women make the economy go round

3

1

80

Saurabh Bhatnagar

@analyticsaurabh

1 year

I don’t know about you, But I like my researchers doing research and advancing human civilization. Not fighting mudslinging on podcasts.

16

6

75

Saurabh Bhatnagar

@analyticsaurabh

1 year

@pmddomingos Yes. But have you dropped a list of chatgpt prompts?

10

0

64

Saurabh Bhatnagar

@analyticsaurabh

9 months

@anothercohen I’m very glad that the AI knew they were talking about a dog

1

0

66

Saurabh Bhatnagar

@analyticsaurabh

5 years

Living in vertical cities is by far the best thing one can do for the environment

Brent Toderian

@BrentToderian

5 years

Quote Tweet this with the thing that you MOST WISH PEOPLE UNDERSTOOD ABOUT CITIES!

54

18

77

2

13

59

Saurabh Bhatnagar

@analyticsaurabh

1 year

@PeterHotez Thank you for keeping the focus on research.

Saurabh Bhatnagar

@analyticsaurabh

1 year

I don’t know about you, But I like my researchers doing research and advancing human civilization. Not fighting mudslinging on podcasts.

16

6

75

0

1

58

Saurabh Bhatnagar

@analyticsaurabh

4 years

@jeremyphoward @angusdav US has 2.6 doctors per 1000 vs SK at 3.7 US has 2.9 hospital beds per 1k vs SK’s 13 So no, unlikely mortality will play out the same here.

4

15

52

Saurabh Bhatnagar

@analyticsaurabh

4 years

@schlaf @thatguyBG @hnshah Deeper issue: Invitee has to fill their name:email like a peasant. Better UX: should be customized links so your info is filled in when you click on it.

5

0

50

Saurabh Bhatnagar

@analyticsaurabh

3 months

@Rainmaker1973 Great. Now the make your own coffee snobs will be even more insufferable

1

0

46

Saurabh Bhatnagar

@analyticsaurabh

2 years

@simonw Here’s the author of tidyverse explaining how to do this using simple models and visualization.

Whole game

Briefly showing the "whole game" of data analysis. See code and data at https://github.com/hadley/building-permits

www.youtube.com

0

2

46

Saurabh Bhatnagar

@analyticsaurabh

10 months

@pfau I don’t know. The work oscillates too much for my taste.

4

0

44

Saurabh Bhatnagar

@analyticsaurabh

2 months

@vboykis Some of you have never upgraded CUDA drivers and it shows

0

2

42

Saurabh Bhatnagar

@analyticsaurabh

6 years

LSTM was more ancient to Alexnet than Alexnet is to us

Andrej Karpathy

@karpathy

6 years

Ancient Egypt was more ancient to Romans than Romans are to us. #funfactoidoftheday

22

170

1K

0

2

43

Saurabh Bhatnagar

@analyticsaurabh

1 year

@soumithchintala So many questions! How did they choose 8? Datasets? Also, how effectively can they prune for specific tasks (assume small domain).

2

0

40

Saurabh Bhatnagar

@analyticsaurabh

9 months

@shaig I don't want to defend TikTok, but short answer, 99% are not.

Saurabh Bhatnagar

@analyticsaurabh

9 months

Remember when twitter was showing people agreeing with Unibomber’s manifesto to every 10 posts. This is the same thing. New Scissor statements have inherent vitality that current algos will boost, because that’s what they are designed to do. And if you see this video in the

3

2

19

1

40

Saurabh Bhatnagar

@analyticsaurabh

1 year

@vkhosla I’m not so certain. You still need creativity that transformers do not currently have. One still needs to imagine the right question to be asked. GPTX will be indispensable like a notepad but need STEM brains for the right questions.

3

1

36

Saurabh Bhatnagar

@analyticsaurabh

1 year

@oldbooksguy

gun

@dril

8 years

issuing correction on a previous post of mine, regarding the terror group ISIL. you do not, under any circumstances, "gotta hand it to them"

142

15K

62K

0

35

Saurabh Bhatnagar

@analyticsaurabh

6 months

@buccocapital The ability to make up stuff with graphs is truly impressive

0

34

Saurabh Bhatnagar

@analyticsaurabh

8 years

New post - dreaming up dresses via AI and game theory #DeepDress

0

9

33

Saurabh Bhatnagar

@analyticsaurabh

5 years

“Far better an approximate answer to the right question, which is often vague, than an exact answer to the wrong question, which can always be made precise” - Tukey

Mara Averick

@dataandme

5 years

🙌 Another great piece by @rdpeng … "Tukey, Design Thinking, and Better Questions" via @simplystats

3

46

179

1

9

35

Saurabh Bhatnagar

@analyticsaurabh

2 years

@moyix Obligatory xkcd

0

2

29

Saurabh Bhatnagar

@analyticsaurabh

1 year

@TaliaRinger You are a veteran. AI Twitter sadly is polluted by hype boys who breathlessly tweet about every paper. Which you know, probably won’t hold up anyway. But it drives news cycles.

1

0

28

Saurabh Bhatnagar

@analyticsaurabh

9 months

@hardmaru @fastml_extra @SchmidhuberAI Some of his “non-ML” papers are so poetic

Simple Algorithmic Principles of Discovery, Subjective Beauty,...

I postulate that human or other intelligent agents function or should function as follows. They store all sensory observations as they come - the data is holy. At any time, given some agent's...

arxiv.org

1

2

30

Saurabh Bhatnagar

@analyticsaurabh

2 years

@adad8m Its called, Poincaré conjecture.

1

0

29

Saurabh Bhatnagar

@analyticsaurabh

5 years

@eigenhector Counting is AI now

2

0

27

Saurabh Bhatnagar

@analyticsaurabh

5 years

@paulg Conference talks are advertisements, not reality. This is esp true in ML World where people like to talk about grandiose architectures but what is likely generating revenue at their company, is usually super simple.

0

3

27

Saurabh Bhatnagar

@analyticsaurabh

1 year

@littmath But it’s easy to blame generics :) I thought I was bad at Maths till 8th grade. Then I had one good teacher (retired Oxford) that showed me the beauty. I went on to get another degree in Maths. Persistence, not genetics.

10

0

26

Saurabh Bhatnagar

@analyticsaurabh

1 year

@gunnr @GergelyOrosz This is correct Another reason is sales. Get on the plane when World leaders will meet you. Not when it’s a cabinet of 20.

1

0

25

Saurabh Bhatnagar

@analyticsaurabh

1 year

@nathanbenaich @josephfcox Bingo. Ethics washing

1

2

27

Saurabh Bhatnagar

@analyticsaurabh

1 year

@pmontu33 @SrinivasR1729 I Hope this is a joke.

2

0

21

Saurabh Bhatnagar

@analyticsaurabh

1 year

Actually with @ESYudkowsky Though I disagree with the agent. AI certainly won't kill us all, but an MBA armed with AI will. Complex systems have many 'side effects'. E.g - Solve World Hunger => dead people aren't hungry - Restore wildlife => releases virus targeting humans

4

5

23

Saurabh Bhatnagar

@analyticsaurabh

7 months

@paulg I think it’s good to examine why this might be happening. My guess - reproductive rights affect women disproportionately. So what we are seeing is lack of empathy for the fairer sex. Not necessarily a flight to the right on other issues like DEI, Immigration, etc

5

0

23

Saurabh Bhatnagar

@analyticsaurabh

7 years

Great overview of SOTA for NLP (hint: LSTMs aren't enough) #l Learnt a lot

(((ل()(ل() 'yoav))))👾

@yoavgo

7 years

slides:

2

11

64

0

12

22

Saurabh Bhatnagar

@analyticsaurabh

2 years

@karpathy Looks like this. Combine with Typer

0

22

Saurabh Bhatnagar

@analyticsaurabh

11 months

@nikitabier Human ability is lognormal curve, not normal. In everything - programming, doctors, lawyers.

0

3

24

Saurabh Bhatnagar

@analyticsaurabh

11 months

@mikpanko 1. Data is never clean 2. Anything beyond sums of cubes requires custom code. 3. Performance 4. Performance - yup, really 5. Typically needed for something downstream, like ML

0

23

Saurabh Bhatnagar

@analyticsaurabh

1 year

@ArmandDoma As usual, it’s not AI but the implementation that is wrong

Saurabh Bhatnagar

@analyticsaurabh

1 year

Actually with @ESYudkowsky Though I disagree with the agent. AI certainly won't kill us all, but an MBA armed with AI will. Complex systems have many 'side effects'. E.g - Solve World Hunger => dead people aren't hungry - Restore wildlife => releases virus targeting humans

4

5

23

8

4

21

Saurabh Bhatnagar

@analyticsaurabh

6 years

@jackclarkSF Final point, if displacement extremely large and fast, too much inequality with not enough time to reacquaint, creates a violent counter revolution.

0

21

Saurabh Bhatnagar

@analyticsaurabh

9 months

@WKNDRD @shaig Indeed. They will, just like with unabomber. On a large enough population, there will be some who will ape into the tending topic. They will then be rewarded by the algorithm for adding to momentum. This is a common recommendation problem, not specific to TikTok.

1

0

20

Saurabh Bhatnagar

@analyticsaurabh

9 months

@kchonyc 3. Pivot to RNNs?

1

0

21

Saurabh Bhatnagar

@analyticsaurabh

1 year

@mbateman Don’t use the word spicy when you give them. If they see someone is scared, they won’t like it. Just give them a little bit on their plate like normal. If they don’t like it, just say “thank you for trying”

2

0

21

Saurabh Bhatnagar

@analyticsaurabh

9 months

@broderick 💯

Saurabh Bhatnagar

@analyticsaurabh

9 months

@shaig I don't want to defend TikTok, but short answer, 99% are not.

1

40

1

0

21

Saurabh Bhatnagar

@analyticsaurabh

2 years

@d_feldman Guessing you haven't made a roti or papad either :) Fire is lindy. The main issue is hermetically sealed rooms we live in. Given COVID and CO2, perhaps people should crack open a window?

2

0

21

Saurabh Bhatnagar

@analyticsaurabh

2 years

@tszzl Diving is so strange. You suddenly become aware you have only been moving in two dimensions.

1

2

18

Saurabh Bhatnagar

@analyticsaurabh

6 months

@despair Vim would never do you like this!

0

20

Saurabh Bhatnagar

@analyticsaurabh

2 years

@megankao_ @AdamTutakiewicz I don’t understand what a falsifiable prediction here is. Reads more like a general sceance.

1

20

Saurabh Bhatnagar

@analyticsaurabh

1 year

@roydanroy Deepmind would never do Nature publishing marketing again, right?

0

1

19

Saurabh Bhatnagar

@analyticsaurabh

5 months

@mattyglesias French Dads :)

0

19

Saurabh Bhatnagar

@analyticsaurabh

9 months

Remember when twitter was showing people agreeing with Unibomber’s manifesto to every 10 posts. This is the same thing. New Scissor statements have inherent vitality that current algos will boost, because that’s what they are designed to do. And if you see this video in the

Yashar Ali 🐘

@yashar

9 months

Over the past 24 hours, thousands of TikToks (at least) have been posted where people share how they just read Bin Laden’s infamous "Letter to America," in which he explained why he attacked the United States. The TikToks are from people of all ages, races, ethnicities, and

8K

7K

25K

3

2

19

Saurabh Bhatnagar

@analyticsaurabh

6 years

@jackclarkSF Not all revolutions are created equal. The two factors are: - displacement - speed For example, from buggy -> car, massive displacement of stables. Speed was fast. Lots found themselves without jobs quick. Speed slow for Gutenberg press.

1

5

19

Saurabh Bhatnagar

@analyticsaurabh

1 year

@frantzfries Its been an absolute disaster for ML twitter feed

Saurabh Bhatnagar

@analyticsaurabh

2 years

Crypto spammers are about to pivot to stable diffusion spammers

1

5

0

19

Saurabh Bhatnagar

@analyticsaurabh

5 months

@alexeheath To be fair, Amazon’s ML teams have similar relationship to one another

0

17

Saurabh Bhatnagar

@analyticsaurabh

10 months

@paulg Ironically, programmers will need to be better at English.

1

17

Saurabh Bhatnagar

@analyticsaurabh

7 years

@mrb_bk @jeremyphoward "Ozymandias" is an allegory for every legacy system the interviewee created

1

3

18

Saurabh Bhatnagar

@analyticsaurabh

2 years

@ylecun Louder restaurant -> talk at higher volume -> dry throat -> drink more -> tip more. For the same reason, all snacks are also too salty.

0

18

Saurabh Bhatnagar

@analyticsaurabh

1 year

@yoavgo I tried it. Then found it got in the way. So went to openAI APIs directly. I suppose it would be nice for a level of abstraction if I wanted to switch providers. But I want deeper access (activations for last n layers), not shallow.

1

0

17

Saurabh Bhatnagar

@analyticsaurabh

1 year

@TheStalwart It’s ok. ChatGPT lawyer will defend you.

0

17

Saurabh Bhatnagar

@analyticsaurabh

2 years

@xsteenbrugge @GoogleAI Always has been

1

0

16

Saurabh Bhatnagar

@analyticsaurabh

1 year

@_jasonwei When skill is scarce, demonstrating skill is valuable. With flooded supply, random stamps/verifications are inevitable. Source: someone who never bothered with ML degree because did prod ML before it was cool. Same cycle.

0

17

Saurabh Bhatnagar

@analyticsaurabh

1 year

@jeremiahscholl @Noahpinion I’m not an expert in this particular data, but I’m a data scientist and going to call schenanigans on Chinese life expectancy data just by the looks of it (also COVID)?

2

0

17

Saurabh Bhatnagar

@analyticsaurabh

1 year

@johncutlefish Give weight to the sponsor’s opinion in a crowded exec suite.

0

17

Saurabh Bhatnagar

@analyticsaurabh

3 years

My favorite mnemonic: Boy who cried wolf 🐺 First time, type I Second, type II.

Sean J. Taylor

@seanjtaylor

3 years

Here’s my trick.

61

779

6K

0

4

17

Saurabh Bhatnagar

@analyticsaurabh

5 years

Personal news: just became a US citizen 🇺🇸

2

0

16

Saurabh Bhatnagar

@analyticsaurabh

8 months

@realtimeai @DavidDeutschOxf It’s because every time we have contacted alienated tribes (see Brazil’s various experiments), they have died. Common cold can be deadly to them. It’s complicated by well meaning missionaries.

4

0

15

Saurabh Bhatnagar

@analyticsaurabh

1 year

@zackkanter No one serious back then thought Watson had an actual product. Surprised by The Atlantic.

Saurabh Bhatnagar

@analyticsaurabh

7 years

AFAIK, IBM Watson is Bayesian consulting services + data mart. I don’t think there is any deep learning (what they are implying by AI). A lot of smart people work there, wish more worked on engineering than sales

1

0

4

1

0

15

Saurabh Bhatnagar

@analyticsaurabh

2 years

@amasad Bullying

3

2

16

Saurabh Bhatnagar

@analyticsaurabh

8 months

@jasoncrawford It’s an improvement, but overstating the claims (DeepMind does this often). I think the earliest on this path was learned index It’s just another stepping stone

1

16

Saurabh Bhatnagar

@analyticsaurabh

2 years

It’s remarkable how all crypto experts became AI experts overnight

0

15

Saurabh Bhatnagar

@analyticsaurabh

9 months

@Austen Us Nigerian princes live quiet lives

0

12

Saurabh Bhatnagar

@analyticsaurabh

3 years

@GergelyOrosz Noisy oncall means tech debt, a sign of bad management. Another question: should on-call rotations even exist with global remote teams?

1

0

15

Saurabh Bhatnagar

@analyticsaurabh

2 years

If a food cannot go bad, it’s not good.

2

1

15

Saurabh Bhatnagar

@analyticsaurabh

8 months

@giffmana Wasn’t familiar. Adding reference for others

0

15

Saurabh Bhatnagar

@analyticsaurabh

2 years

@karpathy @ivan_bezdomny So many use cases. Fast prototyping and sharing of ideas with pictures. Here is an upside down house.

Saurabh Bhatnagar

@analyticsaurabh

2 years

My productivity has plummeted today. #stablediffusion

0

4

0

15

Saurabh Bhatnagar

@analyticsaurabh

1 year

@karpathy Thanks! I think I saw this first in this repo. Thanks for a good breakdown.

0

15

Saurabh Bhatnagar

@analyticsaurabh

5 years

Thanks @citnaj Grandma, grandpa, mom and I

0

14

Saurabh Bhatnagar

@analyticsaurabh

7 years

This is a ridiculously cool idea... I encourage you to AMA Stephanie. If you're in a position to help, try holding office hours the same way. PS, will try to do this

0

1

14

Saurabh Bhatnagar

@analyticsaurabh

5 months

@CFDevelop I’d actually caution juniors to use it. I’d recommend 20/20/20 rule - try for 20 minutes, if you’re still stuck, take a walk. Then try again for another 20. If you’re still stuck, ask a senior (now chatGPT/Claude maybe). Without first 60, they don’t have good questions.

5

0

13

Saurabh Bhatnagar

@analyticsaurabh

4 years

@justGLew Markets: Have you tried tuning them off and on again?

0

1

13

Saurabh Bhatnagar

@analyticsaurabh

5 years

@hardmaru @discoverhk Being a human is important. Keep it up.

0

14

Saurabh Bhatnagar

@analyticsaurabh

1 year

@paulg The danger of taking this seriously is that we don’t talk about more plausible risks.

Saurabh Bhatnagar

@analyticsaurabh

1 year

Yeah this is all very shi shi. More realistic: AutoGPT + voice synthesis hooked to get cheap flight tickets, realizes that some well placed, impersonated calls can bring down the price of a ticket. Reference: Cane Toad introduction in Australia.

1

0

3

1

0

14

Saurabh Bhatnagar

@analyticsaurabh

1 year

@ryanjanssen @Grady_Booch Glad you asked. Eg are you sure you know binary search? It took 20 years to find this bug in nearly all implementations. Everyone who implemented the original got an A+, but no one looked deep enough.

Extra, Extra - Read All About It: Nearly All Binary Searches and Mergesorts are

Posted by Joshua Bloch, Software EngineerI remember vividly Jon Bentley's first Algorithms lecture at CMU, where he asked all of us incoming Ph.D. ...

research.google

2

0

13

Saurabh Bhatnagar

@analyticsaurabh

6 years

New recommendation: Don’t use conda/virtualenv, just use docker + pip. You need it for deploy anyway. PS, wishing on a shooting ⭐️ for a better docker

Saurabh Bhatnagar

@analyticsaurabh

6 years

@akelleh We use conda + docker. It’s my fault for starting with conda and then realizing we need docker anyway. I’m wondering now if we need conda/virtualenv in that combo at all. Isn’t just docker + pip enough? Can also install R the regular way. No need to worry about OSes either

1

0

3

0

12

Saurabh Bhatnagar

@analyticsaurabh

6 years

People who forget relational databases are condemned to reinvent them

Milind Bhandarkar

@techmilind

6 years

@acoustik As usual, Stonebraker was right in 2011. NoSQL = Not Yet SQL.

1

0

4

1

5

14

Saurabh Bhatnagar

@analyticsaurabh

1 year

@JoePompliano No good deed goes unpunished

Visakan Veerasamy

@visakanv

1 year

there's an incredible story going on in the mountaineering community rn. there's a Malaysian climber who was at death's door, and was rescued by a sherpa who risked his own life to carry his lifeless body on his back for 6 hours. The climber then blocked the sherpa on Instagram

318

18K

145K

0

1

11

Saurabh Bhatnagar

@analyticsaurabh

1 year

@OfficialLoganK Interactively create a hoodie that reflects your personal style

1

12

Saurabh Bhatnagar

@analyticsaurabh

1 year

@yoavgo I hate to end on a negative note. So I’d recommend guardrails get @ShreyaR I haven’t used it yet (solved my problem before I saw this), but it seems like the correct approach to a real problem I faced.

GitHub - guardrails-ai/guardrails: Adding guardrails to large language models.

Adding guardrails to large language models. Contribute to guardrails-ai/guardrails development by creating an account on GitHub.

github.com

1

13

Saurabh Bhatnagar

@analyticsaurabh

1 year

It happened. Psychology majors pivoted to AI assessment.

Sebastien Bubeck

@SebastienBubeck

1 year

At @MSFTResearch we had early access to the marvelous #GPT4 from @OpenAI for our work on @bing . We took this opportunity to document our experience. We're so excited to share our findings. In short: time to face it, the sparks of #AGI have been ignited.

67

728

3K

1

2

10