RC @ Stanford @RCS0__ Twitter profile

Pinned Tweet

RC @ Stanford

@RCS0__

1 year

Margin call 2011

0

24

Last Seen Profiles

@daiamondray_

@AlexisP69604221

@teewitemtitties

@sayuri_dev

@fayyworld

@onscarletswalk

@King_Scarlet_1

@galery_basah10

@bokeplokalmalam

@Ayan31130959205

@SquawkCNBC

@eurekax49

@7s_nutrition

@HolyRabbitJonex

@stwmaniax

@AuregaliaAruri

@LaurentMnt

@writeanne

@AmerBIZ

@stw_pdg

@nytopinion

@lucaspipoka

@stwmaniax

@BinorRaja

@Jor_1889

@SirDigALot

@davidfaber

@Drycleaner15

@lechefdelatable

@old_new_history

@antonran11

@Corallianassa

@coopslr

@carry1st

@Monsta_Corp

@Theo_Feel_Us

RC @ Stanford

@RCS0__

1 year

@ropz @ESLCS Brother you wrote the code 😭 why you asking us

3

5

2K

RC @ Stanford

@RCS0__

13 days

@yacineMTB If u read the last section of this book u will have know 90% of all maths used in ML papers. Not kidding

14

58

2K

RC @ Stanford

@RCS0__

7 days

@jam3scampbell This is true, dario amodei worked with richard sutton when he was 4 year old. Sutton saw how baby dario use exploration vs exploitation and got his idea. I was there

6

281

RC @ Stanford

@RCS0__

1 month

@mckaywrigley Bruh you be hyping everything up though. Is there any AI product you tried and didn't like? You are like the fat kid reviewing pizza, of course every thing is tasty

9

5

271

RC @ Stanford

@RCS0__

11 months

@growing_daniel Only read two sentences but I can already see that he cooked

0

176

RC @ Stanford

@RCS0__

9 months

@darrenangle UnsafeLLM : We asked an llm to do something unsafe/unethical 100 times and it finally did it 😯😧

1

180

RC @ Stanford

@RCS0__

1 year

@aizyesque You win pistol & your team already has 25% of rounds needed in the bag. Don't see how 12 rounds will work without big changes to economy

1

0

165

RC @ Stanford

@RCS0__

13 days

@yacineMTB 90% of S tier conference papers in ML just put fancy math for the sake of it. At msft research we made a distillation algorithm and then wrote 1 page full of theory after we had all the numbers just for the sake of it 😭 we thought of the theory 1 week before deadline 💀

2

5

165

RC @ Stanford

@RCS0__

11 months

@khoomeik Some guy named shivam has to become head of anthropic and we will get the whole satyam-shivam-sundaram 🤩😌

4

3

111

RC @ Stanford

@RCS0__

2 months

@LonghornSloth @growing_daniel White supremacists post 100s of disgusting racist memes against indians every day without any reason but if indians answer in kind its suddenly bad?

5

2

102

RC @ Stanford

@RCS0__

9 months

@SchmidhuberAI @DjokerNole Yann been real quite since these pictures dropped 👀

1

0

97

RC @ Stanford

@RCS0__

20 days

@kepe__ @crypt0x_0 They have 4 batches in a year now 😭 turned it into a local college

3

0

96

RC @ Stanford

@RCS0__

19 days

@that_anokha_boy "YC application process is really tough, we look at all applications carefully"

1

0

94

RC @ Stanford

@RCS0__

11 months

@yacineMTB Everything from e/acc vs e/a to the cultural wars is just STEM vs Humanities

7

3

81

RC @ Stanford

@RCS0__

2 months

@karpathy @levelsio Another tweet sponsored by the big micro-entrepreneur complex

2

0

82

RC @ Stanford

@RCS0__

11 months

@tunguz This is what ilya saw

0

80

RC @ Stanford

@RCS0__

10 months

@jxmnop Ilya at 2014 neurips, he knew all along. This is why he is one of the goats

1

7

80

RC @ Stanford

@RCS0__

9 months

@yacineMTB

2

1

74

RC @ Stanford

@RCS0__

1 month

@anammostarac No person named Nick is surviving more than a week in bangalore traffic or metro

3

2

75

RC @ Stanford

@RCS0__

11 months

@tunguz Satya had seen the Indian cricket took lose a final, he wasn't going to take another L

0

57

RC @ Stanford

@RCS0__

11 months

@ylecun But @ylecun you don't hire from outside US. Meta recruiter told me they can't continue the hiring process because I am currently in India with no US work permit 🥺😔

5

1

57

RC @ Stanford

@RCS0__

11 months

@tsarnick I have seen lot of ML guy give interviews but every single ilya interview/podcast has so much alpha. No one explains it better than him

2

55

RC @ Stanford

@RCS0__

20 days

@voooooogel @crypt0x_0

1

0

49

RC @ Stanford

@RCS0__

1 year

@ylecun Ask meta to make it. Call it TensorBook. 24GB VRAM & 32 GB RAM with 24 hour battery. Would sell in minutes.

7

1

43

RC @ Stanford

@RCS0__

1 year

@eigenrobot "All the armies of Europe and Asia...could not by force take a drink from the Ohio River" - Lincoln

2

1

41

RC @ Stanford

@RCS0__

11 months

@MichaelDavLange @SpecialPuppy1 All these other ppl making explanations based on "indians" they met 😂 The real answer is most don't know there are so many indians there, also we are too far away & have never had much to do with South America geopolitics

0

32

RC @ Stanford

@RCS0__

1 year

@yacineMTB "You too blackpilled about AGI bro, just finish linear algebra by axler & you can join oai/anthropic & make agi too"

2

3

30

RC @ Stanford

@RCS0__

2 months

@growing_daniel The replies to the original post where filled with wignat accounts using the most stereotypical indian names and using slurs no indian would know of Basically two jobless demographics fighting and indians getting the short end

0

36

RC @ Stanford

@RCS0__

1 year

@ropz @ohnePixel

1

0

35

RC @ Stanford

@RCS0__

1 year

@mariushelo @aizyesque According to HLTV stats 70% of teams that won the pistol went on a 3-0 run (T & CT) in tier 1 for last one year.

2

0

34

RC @ Stanford

@RCS0__

12 days

@ndt343m @yacineMTB Concentration inequalities section. After I did that i could legit understand so many more papers Every paper that says "our algorithm is guaranteed to converge/bla bla in xyz" ends up being one of these inequalities

0

1

34

RC @ Stanford

@RCS0__

23 days

@__tinygrad__ @AMD Lol I just don't get it. There is billions to be if they could just get competent people on the software team & for same reason they just don't? Sad fumble

3

0

32

RC @ Stanford

@RCS0__

1 year

@MetaAI Just asked my 14 year brother to give up his dreams of becoming a programmer & start playing football or something. Thanks meta

2

0

31

RC @ Stanford

@RCS0__

8 days

@jxmnop For this to be accurate put another car ahead of Geoffrey with Schmidhuber driving it and Geoffrey tailgating

3

0

29

RC @ Stanford

@RCS0__

1 year

@AIatMeta You are going to sell ad space on brain wavelengths, right? Like futurama?

1

0

26

RC @ Stanford

@RCS0__

9 months

@abacaj It usually means proprietary data + field expert post processing

2

0

27

RC @ Stanford

@RCS0__

9 days

@aidan_mclau In the future Aidan going to be like one of those historians u get in the documentary who always knew what was going to happen 😭

2

0

27

RC @ Stanford

@RCS0__

13 days

One secret that changed my mind completely about theory in ML papers is reading the last section of this book. It’s literally just the same 2-3 inequalities used over & over in almost every ML paper 😭

kache

@yacineMTB

13 days

I Can't Believe It's Just Map Reduce

7

20

530

1

4

26

RC @ Stanford

@RCS0__

1 month

@devahaz @varunramg Because they are realizing in real time how dumb the "based tech e/acc shtposting" strategy is

1

0

25

RC @ Stanford

@RCS0__

2 months

@beardedkage @mariusschober @devahaz Freezing point is 30% hot??? 🧐

3

0

25

RC @ Stanford

@RCS0__

11 months

@0xmaddie_ Also gpt4 ended training before 2023. Its amazing they are still running laps around newer models. The diff I think is in their final stack, its not just 1 model. Probably mixture of experts of 180b models + lot of other things. This stack is their sauce.

1

0

24

RC @ Stanford

@RCS0__

1 month

@intervitens "We would have gotten away with the grift if not for the damn llama tokenizer"

0

1

24

RC @ Stanford

@RCS0__

10 months

@yacineMTB Standing on the shoulder of giants 👀

1

0

23

RC @ Stanford

@RCS0__

9 days

@yacineMTB Huggingface library is abstraction hell. Finding where they do their sampling logic feels like i am trying to find the holy grail

1

0

23

RC @ Stanford

@RCS0__

2 months

@FPLFocal Willock finally hitting the transfer button

0

21

RC @ Stanford

@RCS0__

11 months

@yacineMTB This is true story I was the wife

0

21

RC @ Stanford

@RCS0__

10 days

@fchollet Hinton outside the physics department

0

21

RC @ Stanford

@RCS0__

11 months

@IterIntellectus This is what he is building (imo)

2

3

19

RC @ Stanford

@RCS0__

15 days

@bschne dropbox.png

0

18

RC @ Stanford

@RCS0__

1 year

@yacineMTB Legit true story > work in Microsoft Research > Friend working on using LLMs in recommender system > friend hasn't used gpt4 yet & thinks LLMs are only hype > First time using gpt4 & he is blown away Happened like 1 month ago

1

0

18

RC @ Stanford

@RCS0__

2 months

@yacineMTB There are a dozen clones of you in every engineering college in India bruh. I have seen it with my own eyes. Anime pfp + try to talk like roon. Once they all join this side of twitter it's over 😭

1

0

16

RC @ Stanford

@RCS0__

21 days

@squirtle_says @yacineMTB is responsible for this

1

0

16

RC @ Stanford

@RCS0__

11 days

@jxmnop He carrying the ML world on his back like cavs lebron 🐐

0

17

RC @ Stanford

@RCS0__

1 month

@yacineMTB On 4chan people posted for the love of the game, 90% of twitter is now blue ticks posting bait for some elon bucks

1

0

16

RC @ Stanford

@RCS0__

1 month

@shinboson @csahil28 @mattshumer_ The tokenizer test & how it kept changing is enough evidence to know they were lying Also sahil says "we couldn't reproduce the benchmark results" but their private endpoint did work much better than llama 450b (because it was Claude 😂)

0

15

RC @ Stanford

@RCS0__

1 year

@skooookum Atheists after one sip of McDonald's diet coke

0

16

RC @ Stanford

@RCS0__

13 days

@LukeFarritor I guess this is so that new student dont get one shotted by their first heat wave

0

16

RC @ Stanford

@RCS0__

1 month

@joefioti He says that he has a private api endpoint which works correctly but for some reason he also has to retrain lol Why not just upload the private end point weights?

1

0

15

RC @ Stanford

@RCS0__

11 months

@typedfemale Sam not taking your ass to msft then you can stay with ilya and make a safe agi 😭

0

14

RC @ Stanford

@RCS0__

1 year

@nearcyan For some reason the vibes feel just right. Whenever I see this login page my trust on their tokens increases by 5%

1

0

13

RC @ Stanford

@RCS0__

12 days

@nanami0vertime @yacineMTB Most alpha is in "concentration inequalities" section but if u find like a convex something something proof that u dont understand just go over the that section & you will find some inequality or theorem that will clear it up. One of the actually good ML books

0

1

14

RC @ Stanford

@RCS0__

11 months

"Unfortunate leak" Yeah Q* is 100% true and some insane banger alphago type moment for LLMs. Ilya actually did see something crazy

Jimmy Apples 🍎/acc

@apples_jimmy

11 months

From the The Q** persists

0

45

372

0

2

14

RC @ Stanford

@RCS0__

10 months

@eshear It has to be hal finey, right? He had the tech expertise, worked in finance & his neighbor was a software engineer literally named "Satoshi Nakamoto". Hal also first to receive a non anonymous btc transaction. Also hal died 10 years ago which explains why no one has heard from

0

1

12

RC @ Stanford

@RCS0__

22 days

@weilunchao “The paper receives”

0

13

RC @ Stanford

@RCS0__

11 months

@AutismCapital @gdb Ilya just got traded to Anthropic where he can do his safety research for 5 1st round draft picks

0

12

RC @ Stanford

@RCS0__

7 days

@dejavucoder "LLMs are so bad at coding it useless for anything even remotely complex" - Uses the free gpt version and has not touched sonnet yet

3

0

13

RC @ Stanford

@RCS0__

9 months

@yacineMTB The only thing worse than seeing indians talk about caste is seeing non-indians talk about it 😭

1

0

13

RC @ Stanford

@RCS0__

1 year

@yacineMTB Best book ever for people that really understand ML (never read it)

1

11

RC @ Stanford

@RCS0__

1 year

@eigenrobot The full quote is even better "All the armies of Europe, Asia and Africa combined, with all the treasure of the earth (our own excepted) in their military chest; with a Buonaparte for a commander, could not by force, take a drink from the Ohio River"

0

11

RC @ Stanford

@RCS0__

2 months

@fchollet What about price gouging on grocery?

1

0

11

RC @ Stanford

@RCS0__

1 year

@abacaj Can you tell me how you are doing this? 15b in f16 is 24gb, right? So training it full means around 24*3 GB vram. Each gpu only has like 1/8th of the layers?

1

0

10

RC @ Stanford

@RCS0__

1 year

@Teknium1 Most probably an anti AI luddite

1

0

10

RC @ Stanford

@RCS0__

1 year

@ehalm_

0

11

RC @ Stanford

@RCS0__

1 year

@growing_daniel Watching marvel cape slop in 2023 is more embarrassing than using AI art

0

10

RC @ Stanford

@RCS0__

27 days

@MarkovMagnifico Our plans are measured in *centimeters

1

0

11

RC @ Stanford

@RCS0__

9 months

@_akhaliq One normal day of AI research pls 😂 e everyday we have something groundbreaking. Yesterday it was Alphageometry, today this

1

0

10

RC @ Stanford

@RCS0__

11 months

@FirstClassDuck @d78930305 @MichaelDavLange @SpecialPuppy1 Bruh what 😂 US diaspora doesnt command how India acts on geopolitics. 99% of Indians don't even know Carribea has so many Indians/hindus. Also we are too far away to have any geo political impact

0

1

9

RC @ Stanford

@RCS0__

11 months

@thecaptain_nemo Founder of a website no one has used since 2015 teams up with Gordon Levitt's wife and a Fed plant to take down the most valuable AI company. David fincher got a banger on his hands

0

10

RC @ Stanford

@RCS0__

1 year

@Teknium1 me when gpt4 spills out text that will get me 50 years in federal prison just for reading

0

1

9

RC @ Stanford

@RCS0__

1 year

@yacineMTB on my way to fill the form knowing I don't have the GPU compute & will use ggml version anyways

0

9

RC @ Stanford

@RCS0__

1 year

@bryancsk When I laugh at a meme & then realize it was made by AI

0

8

RC @ Stanford

@RCS0__

1 year

@tszzl How long before roon deletes this? 👀

0

6

RC @ Stanford

@RCS0__

11 months

@growing_daniel Fargo s1 and true detective were in same year. Peak Television, it's been downhill since

1

0

9

RC @ Stanford

@RCS0__

1 year

@abhagsain Seeing such detailed debug messages on console makes me feel more secure using their service. It shows they are putting in work 💪😤

0

9

RC @ Stanford

@RCS0__

1 year

@10x_er And its name should always be So that its intuitive & people know which file to run

0

9

RC @ Stanford

@RCS0__

7 days

@HKCorn @SpursOfficial 1:9 ratio 😭 Imagine celebrating getting your own a$s kicked

2

0

8

RC @ Stanford

@RCS0__

1 year

@yacineMTB Even ppl working on LLMs kind of don't understand how good gpt4 & LLMs already are & how good they are going to be

0

8

RC @ Stanford

@RCS0__

11 months

@powerbottomdad1 He going to star in the new movie "500 days of sama"

0

7

RC @ Stanford

@RCS0__

1 year

@abacaj Heaven's mandate is my moat. God selected me to make a saas product that will increase your worker efficiency by 10%. OpenAI cant stop me

0

7

RC @ Stanford

@RCS0__

11 months

@iliekcomputers @cto_junior @sama If sama retweets this with a ♥️ then you are hired

0

8

RC @ Stanford

@RCS0__

1 year

@growing_daniel Top 10 greatest industry plants ever

0

8

RC @ Stanford

@RCS0__

1 year

@kosenjuu He has been in the lab cooking for too long

2

0

8

RC @ Stanford

@RCS0__

11 months

@tsarnick I changed my mind about LLMs being stochastic parrots after seeing that ilya and nvidia ceo interview 💀 he was that convincing 😂

1

0

8

RC @ Stanford

@RCS0__

11 months

@joecappadona @markopolojarvi @KyeGomezB This + ins & outs & tricks needed for full pretraining runs. The CEO of mistral, who also lead the llama team at meta said that there are like a dozen people in the world who actually know how to handle pretraining a LLM correctly

1

0

8

RC @ Stanford

@RCS0__

9 months

@typedfemale Take the moat away from competitors in AI space + attract top talent + more compute is never bad

1

0

8

RC @ Stanford

@RCS0__

1 month

@abacaj It's over

intervitens

@intervitens

1 month

The reflection model on their API is just prefilled Claude 3. (All of the models are limited to 10 tokens, temperature 0, top_k 1)

18

23

435

1

0

8

RC @ Stanford

@RCS0__

11 months

@___frye @deepfates Stanley Kubrick actually wanted it. The directors cut version of the moon landing has this background

0

7

RC @ Stanford

@RCS0__

9 months

@cto_junior Everyone was training llms. Google trained a trillion param llm back in 2021. But no one took it seriously because llms just werent useful enough for their inference costs. Gpt3 was a hit because it was the first llm that was actually useful and fast enough.

0

8

RC @ Stanford

@RCS0__

11 months

@goth600 Me not caring about the e/a vs e/acc thing

1

6