stephen balaban @stephenbalaban Twitter profile

Pinned Tweet

stephen balaban

3 months

Lambda has raised a $320M Series C led by Thomas Tull's USIT. This new financing will be used to expand the number of NVIDIA GPUs available in Lambda Cloud and build features that will absolutely delight you.

47

62

506

Last Seen Profiles

@rufusgifford

@ciaomack

@Prestisa_id

@chris

@Okajimahal

@haytaesan

@medienmil

@SetonHall

@DJ_MAMOT

@chhopsky

@yktv37

@enbyoonji

@THEQUEET

@cucarecheto

@KEIENDO_jp

@BS4chan

@Bllury

@ChibunduEunice

@maid_misa1

@PoisonedPixiie

@GRDS0inqHYoch8D

@yuriyuri0803

@HIEE5EUNG

@Ali_morol

@ExploreKulgam

@hocca_official

@stpatricksgaa

@ScarletFindomme

@drjahanpur

@swampbottom____

@KLEELove_0727

@TurnerCael

@AdbaTv

@teletubisolo

@FlavorFlav

@bunnyyycat

stephen balaban

@stephenbalaban

3 months

This is the strongest evidence that I’ve seen so far that we will achieve AGI with just scaling compute. It’s genuinely starting to concern me. I used to think that we would run into roadblocks: end of scaling laws, maybe we don’t have the right model architecture, power density…

Tsarathustra

@tsarnick

3 months

Sora performance scales with compute

102

297

2K

300

661

6K

stephen balaban

@stephenbalaban

11 months

Fuck it, we just dropped our H100 on-demand prices to 1.99. From now on, Lambda will always have the cheapest price in the entire world, if we don't, just tweet at me and we will drop the price. H100s here:

Brian Roemmele

@BrianRoemmele

11 months

“Nvidia H100 and A100 GPUs - comparing available capacity at GPU cloud providers”

4

21

118

67

242

2K

stephen balaban

@stephenbalaban

3 months

Sometimes shader golf makes me think the minimum description length of our universe might not be that long.

yonatan

@zozuar

3 months

for(float e,i,a,w,x,g,h;i++<90.;o+=.01-.02/exp(max(s,e)*3e3)/h){vec3 p=vec3((FC.xy-.5*r)/r.y*g+2.,g);p.zy*=rotate2D(.5);e=p.y;h=e+p.x*.3;p.z+=t;for(a=.6;a>.001;a*=.7)p.xz*=rotate2D(5.),x=(p.x+p.z)/a+t+t,e-=w=exp(sin(x)-3.)*a,h+=abs(dot(sin(p.xz/a*.3)*a,r/r));g+=e=min(e,h*.5-1.);}

179

2K

18K

29

146

2K

stephen balaban

@stephenbalaban

30 days

That’s nice, but try $1.4B in revenue with 24 people or $3.7B with 28 people. Let’s see who can name these two companies.

Bryce Roberts

@bryce

1 month

$100M in revenue with less and 10 employees is this generations' 4 minute mile.

25

30

517

152

175

2K

stephen balaban

@stephenbalaban

1 year

Yale has a perpetual bearer bond from the Dutch government from 1648 that pays 2.5% interest, they go to the Netherlands every few decades to collect the back interest. It’s the oldest security still generating payments.

22

171

1K

stephen balaban

@stephenbalaban

1 year

NVIDIA popped because everyone underestimated the amount of GPU compute the world needs. I’m honored that Lambda was mentioned in NVIDIA’s earnings call today. Soon, every human will be using AGI powered by GPUs. And Lambda is going to be the #1 GPU cloud in the world.

25

59

772

stephen balaban

@stephenbalaban

1 year

This is how Copilot is “programmed”. This is how your future AGI co-workers, friends, and lovers will be “programmed”. The future is here, it’s just not evenly distributed.

21

67

566

stephen balaban

@stephenbalaban

1 year

I no longer use Google when programming and default to thinking: "I should ask ChatGPT about X." LLMs are a far superior method vs. the standard (read documentation / blog post / stack overflow) loop. It's like having a thoughtful pair programmer with you at all times. A thread:

19

48

527

stephen balaban

@stephenbalaban

1 month

Lambda has raised a $500M GPU-backed financing vehicle. Designed to fund GPUs for hundreds of thousands of developers with No Contract Required. This is half a billion dollars for the 'gpu poor'. When new technologies appear in the market, they require new systems of finance…

Lambda

@LambdaAPI

1 month

Today, we are announcing $500 million in debt financing to scale Lambda’s On-Demand GPU Cloud. Learn more:

5

16

148

33

54

530

stephen balaban

@stephenbalaban

2 years

Lambda GPU Cloud has NVIDIA A100s for $1.10 / hour. On demand. No commitment. That's the lowest on-demand price in the entire world. It's honestly a crazy low price. Spin up a $1.10/hr A100 here ->

GPU Cloud - VMs for Deep Learning | Lambda

NVIDIA H100, A100, RTX A6000, Tesla V100, and Quadro RTX 6000 GPU instances. Train the most demanding AI, ML, and Deep Learning models.

lambdalabs.com

Lambda

@LambdaAPI

2 years

Lambda is excited to offer the best prices on the market for GPU cloud compute. On-demand A100 GPUs starting at $1.10/hr vs AWS $4.10/hr (73% savings). No commitments or negotiation required. Learn more→

6

17

157

12

54

504

stephen balaban

@stephenbalaban

1 year

I've found that it's very useful to speech-to-text a stream of consciousness style overview of what I want to talk about, then ask chatGPT to summarize it. Then, I ask it to re-expand to three or four paragraphs. Like an autoencoder.

8

50

470

stephen balaban

@stephenbalaban

30 days

@amasad Yup! They’re both literal rent seekers.

3

403

stephen balaban

@stephenbalaban

1 year

Announcing that Lambda has raised $44M to build the best cloud for training AI. (Lots of GPUs.) I'll be doing a Q&A on Twitter for the next hour. Just reply in this thread.

Lambda raises $44M to build the world's best cloud for training AI

Today, we’re excited to announce that Lambda has raised a $44 million Series B to accelerate the growth of our AI cloud. In the coming decades, AI will have ...

www.youtube.com

37

47

392

stephen balaban

@stephenbalaban

1 year

There’s so much alpha in being optimistic.

7

37

347

stephen balaban

@stephenbalaban

1 year

Yea wow. Github copilot is amazing. These kind of tools, when they reach mass adoption, are going to dramatically increase global GDP.

14

20

333

stephen balaban

@stephenbalaban

11 months

Working at AWS, GCP, Apple, Meta, Twitter, Microsoft, Oracle, or IBM? Lambda has H100s with 3,200 Gbps InfiniBand at the world's lowest price: $1.89 / hour (100% upfront) $2.04 / hour (80% upfront) $2.15 / hour (monthly payment) Share this to your manager.

11

41

319

stephen balaban

@stephenbalaban

1 year

Today, Lambda Cloud launched H100 instances. We’re the first and only cloud in the world to offer H100s on demand. Grab yours here: Extremely proud of the whole team for launching these before everyone else.

9

42

318

stephen balaban

@stephenbalaban

1 month

@d_feldman Lambda Hyperplane has entered the chat.

12

4

289

stephen balaban

@stephenbalaban

1 year

Lambda is going to be the #1 GPU cloud in the world.

17

25

251

stephen balaban

@stephenbalaban

30 days

First person in the thread to get it was @sarhaan_g both companies are casino REITs: $VICI and $GLPI The other interesting contenders suggested by the thread were RenTec (thought to be $100M/employee), Tether (thought to be $60M/employee or more), and Vitol ($121M/employee).

6

12

253

stephen balaban

@stephenbalaban

1 year

@karpathy how long until prompt hijacking approaches zen/mindfulness for neural nets?

4

1

248

stephen balaban

@stephenbalaban

10 months

What did he mean by this?

39

7

237

stephen balaban

@stephenbalaban

5 months

So excited to see everybody at NeurIPS this year. One of the big launches from Lambda is the new Vector One PC. Buy a Lambda Deep Learning computer with a 4090 GPU w/ EDU discount just $4,899! Buy it here:

20

18

236

stephen balaban

@stephenbalaban

3 months

My goodness, I thought this level of fidelity was going to be a few years out. How many GPU-hours needed to render? How is this here today!?? Dozens of "AI Pixars" are going to emerge in the next few years and the flood of amazing movies and games will blow our minds.

Greg Brockman

@gdb

3 months

Announcing Sora — our model which creates minute-long videos from a text prompt:

1K

4K

21K

14

11

222

stephen balaban

@stephenbalaban

1 year

You can now launch your own stable diffusion web UI on Lambda Demos. Run it on a 24GB A10 for just $0.60/hour. Repository: Lambda Demos:

GitHub - camenduru/stable-diffusion-webui-lambda

Contribute to camenduru/stable-diffusion-webui-lambda development by creating an account on GitHub.

github.com

camenduru

@camenduru

1 year

I am happy 🥰 to announce #StableDiffusion #WebUI with all amazing 🔥 extensions + #ControlNet + #SadTalker in the 🆕 Lambda Demos 🥳 thanks to @LambdaAPI ❤ and @stephenbalaban ❤ I will create more cool 🔥 demos, Please try it 🐣

10

58

214

8

37

213

stephen balaban

@stephenbalaban

3 months

@ram_bala77 If the model of the world it learns is large enough, complex reasoning could fall out after you scale.

15

3

197

stephen balaban

@stephenbalaban

10 months

Lambda has over 4,000 GPUs coming online in Q4 starting at 1.89/hr. $1.89 / hr: 100% upfront payment $2.04 / hr: 33% upfront, 67% monthly payment 3-year contract term 80GB H100 nodes with 3200 Gbps InfiniBand interconnect! DM me!

sarah guo // conviction

@saranormous

10 months

@sampullara do they have 1,000 h100x8 nodes available though? That’s the rub :)

2

0

1

8

15

168

stephen balaban

@stephenbalaban

1 year

Explorations in AI accelerated game development. Prompt: # A pygame game where you can take care of a pet cat. # The pet can has two actions: pet the cat and feed the cat. # The cat responds accordingly. Generated game:

9

21

165

stephen balaban

@stephenbalaban

1 month

“Hey, what if you did a REIT but for GPUs?”

sophie

@netcapgirl

2 months

hiring this guy at the right stage gives your company a 10x better chance of going public or having a successful exit. but if you hire him at the wrong time your company will spiral into a tangled web of bureaucracy

47

28

837

3

18

157

stephen balaban

@stephenbalaban

1 year

Today LLMs are not good at is dealing with large codebases. There’s going to be a whole industry around fine tuning these models to work within large codebases. They’ll learn the internal APIs and architecture. Then they can work on more significant features/refactors.

10

14

146

stephen balaban

@stephenbalaban

6 years

We've benchmarked the 2080 Ti, V100, Titan V, 2080, and 1080 Ti. 2080 Ti destroys V100 / Titan V on performance per dollar. Full blog post here:

4

67

143

stephen balaban

@stephenbalaban

9 months

@var_epsilon This is what happens when you apply yourself. λs.(s s) λs.(s s)

1

3

123

stephen balaban

@stephenbalaban

1 year

I’m starting a new fund. TPOT Capital Partners VII Looking to back the best Poaster-Founders the world has to offer.

7

10

122

stephen balaban

@stephenbalaban

8 months

@DotSledge

4

118

stephen balaban

@stephenbalaban

1 year

Sometimes writing prompts feels like writing positive affirmations for the LLM. You are a world-class writer. You are a smart engineer. You are helpful. You are happy.

10

11

110

stephen balaban

@stephenbalaban

2 years

Plato and Aristotle walking through The School of Athen's data center. Lambda has just added a ton of new 8x A100 capacity to our cloud. You can sign up and launch one here => .

12

21

111

stephen balaban

@stephenbalaban

3 months

@_florianmai I avoid engaging in debate around small probabilities of events with extremely negative utility for the same reason I wouldn’t pay much to play the St. Petersburg paradox game and for the same reason that I might be skeptical of pascal’s wager.

11

1

107

stephen balaban

@stephenbalaban

16 days

@HUN2R Would lose in a fight to Labattman.

0

113

stephen balaban

@stephenbalaban

1 year

Reminds me of the perpetual royalties from Listerine from an 1881 agreement. Any other interesting perpetual securities out there?

Listerine Mouthwash Royalties | Royalty Exchange

auctions.royaltyexchange.com

2

4

103

stephen balaban

@stephenbalaban

1 year

Today, Lambda launched Demos. An easy and inexpensive way to host Gradio-based generative AI apps. Soon, our discovery page will be packed with stable diffusion UIs, balenciaga video generators, rap generators, GPT LLM chatbots, and more gen AI apps.

5

22

100

stephen balaban

@stephenbalaban

1 year

@tszzl @alth0u @Cixelyn @EMostaque Every VC: why are you building a computer company with a manufacturing facility in San Francisco? Lambda: so we can provide infra on-demand in hours when roon pings us on Twitter.

4

9

100

stephen balaban

@stephenbalaban

1 year

I'm looking to hire some absolute killers to work with us on Lambda Cloud. If you want to work on hard engineering problems at the intersection of deep learning and data center scale computation, my DMs are open.

8

18

97

stephen balaban

@stephenbalaban

5 months

Midjourney v6 is a real delight to play with. Text works by just “quoting” the text you want to see. What a year it’s been in AI. Merry Christmas and Happy Hanukkah everyone!

3

4

96

stephen balaban

@stephenbalaban

11 months

Lambda and Coreweave do the same thing. But Lambda charges $1.89/hr list price while Coreweave charges $2.23/hr list price. If you need 1024 H100s with InfiniBand for 3 years, you'd save $9.1M by choosing Lambda. The math and choice is simple: (2.23-1.89)*24*365*1024*3 = $9.1M

3

15

94

stephen balaban

@stephenbalaban

3 months

A good counterpoint from @GaryMarcus was that we will see lots of holes / systemic glitches in Sora in the coming months and that it does not constitute a world model. My thoughts on how generative networks like Sora could be used as a world model:

stephen balaban

@stephenbalaban

3 months

@GaryMarcus OK, if the vast majority of the artifacts (like walking chairs being pulled from the sand) go away, and more complex inter-object dependencies are reliably visually simulated, do you think that constitutes a world model? I would argue that it does, because you could use it for…

5

10

40

2

3

90

stephen balaban

@stephenbalaban

2 years

Lambda was founded 10 years ago today. I was 23 back then and had no idea what I was getting into. I’ve dedicated over a decade of my life to building this company but still feel like we’re just getting started. This is the email I sent to our investors in Jan 2016. (Page 1/4)

stephen balaban

@stephenbalaban

2 years

Wait until this is in full 4K HD, has significantly better spatiotemporal consistency, and also generates audio alongside. Oh yeah, and you’ll be able to talk to the kangaroos too. The future is gonna be crazy.

1

5

35

1

7

83

stephen balaban

@stephenbalaban

2 years

This aged well. In 2011, Stripe was a tiny startup going up against PayPal, a $5 billion dollar revenue incumbent. The better user experience won. Stripe now has more revenue than PayPal did when this tweet was written.

stephen balaban

@stephenbalaban

13 years

met @collision last night and was shown the @stripe HTTP API; PayPal, 100+ lines of code, Stripe, one curl.

0

6

2

9

84

stephen balaban

@stephenbalaban

11 months

@SawyerMerritt They really missed the opportunity to use that new aperiodic monotile.

0

83

stephen balaban

@stephenbalaban

30 days

@sarhaan_g We have a winner!

2

1

78

stephen balaban

@stephenbalaban

11 months

VCs who follow me please take note, if you want to get into Suhail’s next round, you can invest $150M at $1.5B post into playground, then they can get a 3,020 H100 cluster with InfiniBand for $1.89/hr from Lambda. You get allocation, they get GPU compute, win-win.

Suhail

@Suhail

11 months

I just want the world to be aware: I will happily trade 10% of my company for $150M worth of GPU compute. DMs are open.

48

32

651

3

80

stephen balaban

@stephenbalaban

1 year

@Suhail Lambda is working night and day to make the GPU capacity problem go away. 1,000s of A100s, even more, are soon to arrive at our data center door. More GPUs, the world demands, we ratchet up purchase plans. Christmas cheer will spread around with A10 instances soon to be found

4

8

79

stephen balaban

@stephenbalaban

2 years

Everyone: "But you can only do state of the art AI work with millions of dollars of compute and at a big company!!!" @Buntworthy : "text-to-pokemon cost $10 to train on Lambda Cloud." New blog post just dropped on how text-to-pokemon was made:

Lambda

@LambdaAPI

2 years

New post on how @Buntworthy made the text-to-pokemon model by fine tuning stable diffusion. The final model only cost around $10 to train on Lambda GPU Cloud. (Yes, $10.)

1

7

17

0

9

79

stephen balaban

@stephenbalaban

3 months

When I first saw this image on reddit, I knew AI was going to change the world. Thanks @ch402 @mtyka and @zzznah So cool to see where we are today and think about where we will be soon.

kache

@yacineMTB

3 months

>when did you have your "holy shit, AI is actually real and it's going to change hte world in the near future" moment?

91

20

616

3

7

75

stephen balaban

@stephenbalaban

18 days

So lucky to have @twominutepapers stop by Lambda HQ on his Silicon Valley tour. Karoly gave me a named “Hold on to your Papers” card! I’m told that it is a rare artifact!

3

4

78

stephen balaban

@stephenbalaban

11 months

The real question is whether Lambda is more e/acc or more e/λ. Hint:

Beff – e/acc

@BasedBeffJezos

11 months

Top 5 most e/acc companies 1) @Replit 2) @Meta 3) @perplexity_ai 4) @numerai 5) @LambdaAPI

21

24

252

1

3

73

stephen balaban

@stephenbalaban

11 months

California and Nevada should merge into a single state called Calvada with Reno as the capital. It just makes sense. You know what else just makes sense? $1.89/hr H100s with 3,200 Gbps InfiniBand from Lambda Cloud. Lowest price in the world. Get yours:

4

5

74

stephen balaban

@stephenbalaban

1 year

@deliprao And the Monster energy I just drank.

0

69

stephen balaban

@stephenbalaban

8 months

FINALLY! Lambda just launched filesystem storage in the majority of our regions! You can now store datasets, models, and code on shared filesystems that continue to exist after terminating your GPU instance.

Lambda

@LambdaAPI

8 months

Persistent storage is now available in the majority of regions within Lambda Cloud. Read the full announcement:

3

14

3

5

69

stephen balaban

@stephenbalaban

3 months

@obzabor I’m personally extremely optimistic. I think post AGI world is going to be great, basically no chance of doom.

5

0

66

stephen balaban

@stephenbalaban

11 months

Sat down with Pieter Abbeel @pabbeel to talk about the current state of the GPU cloud market, Lambda's founding history, and the deep learning powered future of gaming. @therobotbrains podcast is great.

S3 Ep 13 Stephen Balaban from Lambda on building the most cost-effe...

S3 E13 Stephen Balaban from Lambda on building the most cost-effective AI cloudWhat's in this episode:00:00:00 Stephen Balaban00:01:20 Sponsors: Index Ventur...

www.youtube.com

2

13

65

stephen balaban

@stephenbalaban

3 months

@rahulchhabra07 Yes and the problem is he’s potentially been right all along and consciousness could just be a time series of neural representations.

5

1

62

stephen balaban

@stephenbalaban

1 year

@andrew_n_carr You know when it really hit? When I was talking with @chuanli11 at NeurIPS and he said “you know, most of these papers were submitted for review before stable diffusion came out.” NeurIPS used to be the place for the greatest reveals and now it’s “months behind”. Which is a lot!

1

6

60

stephen balaban

@stephenbalaban

1 year

1. ChatGPT created a python client based on Lambda Cloud's API spec, and it works! 2. Lambda launched v1 of our API today. (Docs: ) 3. We added lots of new A100 capacity, lots more GPUs! (Use gpu_1x_a100_sxm4 instances in us-east-1.) ChatGPT's client👇

2

7

62

stephen balaban

@stephenbalaban

1 year

The fastest way to get 64 or more H100 GPUs in this environment is to *pay upfront* and sign a reserved H100 contract with Lambda. GPUs are allocated to the companies that pay first, to get in line just fill out this form:

Lambda GPU Cloud | Reserved Cloud Clusters

Affordable, high performance reserved GPU cloud clusters with NVIDIA GH200, NVIDIA H100, or NVIDIA H200.

lambdalabs.com

Suhail

@Suhail

1 year

cry:

30

13

226

8

12

59

stephen balaban

@stephenbalaban

11 months

@tszzl local minimum orthogonal mutually exclusive first/second derivative (referring to business metrics) power law at the limit doubting you’ll “get much of a delta” out of an activity alpha (crypto/fin/poaster crossover) (especially when used incorrectly) law of large numbers

5

0

59

stephen balaban

@stephenbalaban

2 months

I can't believe that I can generate an entire song for maybe a dime or so? Each of these would be tens of thousands of dollars to produce just a few months ago.

4

3

57

stephen balaban

@stephenbalaban

1 year

A friend asked me to make some predictions for AI in 2023. Prediction: nothing changes. My wildest deep learning related predictions from 2013 are only now starting to come true. Little happens in one year, a lot happens in 10 years. So let's start building for 2033 today!

2

4

54

stephen balaban

@stephenbalaban

1 year

Soon, your biggest customer is going to be an AI. Most companies are not ready for this.

Greg Brockman

@gdb

1 year

We’ve added initial support for ChatGPT plugins — a protocol for developers to build tools for ChatGPT, with safety as a core design principle. Deploying iteratively (starting with a small number of users & developers) to learn from contact with reality:

237

2K

8K

3

8

51

stephen balaban

@stephenbalaban

1 year

ChatGPT gave me an example that perfectly demonstrated the superiority of it's solution. Imagine how much googling and documentation reading I would have to do otherwise! Plus, it avoids the horribly formatted blog spam that permeates the internet of code.

3

1

51

stephen balaban

@stephenbalaban

1 year

If you want cloud H100s, we have them. Need H100s + InfiniBand? You can sign a 3 year upfront contract with Lambda. You’ll get the lowest price in the world and you’ll get allocation as fast or faster than other cloud providers. DM me or sign up here:

GPU Cloud - VMs for Deep Learning | Lambda

NVIDIA H100, A100, RTX A6000, Tesla V100, and Quadro RTX 6000 GPU instances. Train the most demanding AI, ML, and Deep Learning models.

lambdalabs.com

2

6

51

stephen balaban

@stephenbalaban

8 months

@aicrumb Hey crumb, we try to delight customers with everything we do. If you don’t like our alien mummy swag item, we can give you a hat instead and send the alien to others. Just email our support team and we can get it squared away for you.

2

1

50

stephen balaban

@stephenbalaban

1 year

Lambda has lots of new H100s deployed and coming online soon! Sign up and you’ll get an email the moment it’s launched or DM me to reserve large capacity.

Suhail

@Suhail

1 year

I am gonna forecast this happening a lot to everyone soon. Get your GPU trainings in while the world is still upskilling to AI research.

67

47

743

0

5

49

stephen balaban

@stephenbalaban

5 months

OK, I know you're tired of being asked to sign 3 year contracts with upfront payment from Lambda. So Lambda now has burst compute and flexible contract terms! We're offering 3 month bursts of GH200s with 0% down and 1 year H100 + infiniband contracts: …

1

8

50

stephen balaban

@stephenbalaban

3 years

Lambda has raised $24.5M to grow our GPU cloud and build on-prem deep learning infrastructure software. Excited to be building a world class company with @1517fund @GradientVC @BloombergBeta @Razer and @gharik .

Lambda raises $24.5M to build GPU cloud and deep learning hardware

Lambda secured $24.5M in financing, including a $15M Series A equity round and a $9.5M debt facility that will allow for the growth of Lambda GPU Cloud and the expansion of Lambda's on-prem AI...

lambdalabs.com

2

8

49

stephen balaban

@stephenbalaban

16 days

Such a cool photo behind the scenes of making FF7. One of the finest works of art created by mankind.

Obsolete Sony

@ObsoleteSony

18 days

Kazuyuki Hashimoto working on Final Fantasy VII using a 1995 Sony BVM-1916

22

463

3K

6

2

49

stephen balaban

@stephenbalaban

1 year

This compression to latent space (summarization) / decompression back to input space (expansion), like a real autoencoder, removes verbal 'noise'. Uh/ums, non sequiturs, even disorganized thought. The cool thing is, this time the latent 'latent space' is human editable text!

1

0

49

stephen balaban

@stephenbalaban

1 year

I always loved this ad. Circa 2018. Sort of hurts the eyes, but is memorable.

3

48

stephen balaban

@stephenbalaban

8 months

I just want to say that I have been arena posting for 13 years now — see blog post from this exact day in 2010. You have to skate to where the puck is going people.

3

1

48

stephen balaban

@stephenbalaban

8 months

Best to just keep scrolling. It's too disorienting to stop and think of the implications of this technology...

The Daily Ai

@The_DailyAi

8 months

Created a Star Wars concept film using Pika Labs and Midjourney in just a few hours. It's wild how easy it is to animate the scene using these tools. -Curious_Refuge

64

261

1K

2

3

49

stephen balaban

@stephenbalaban

1 year

Yea, wow, I really think there's something here. @karpathy or @AlecRad any papers on this topic?

2

0

48

stephen balaban

@stephenbalaban

1 year

One of the coolest things about the ChatGPT iOS application is that the dictation can be imperfect but the LLM still understands what you meant. Transcription errors are compensated for by the LLM. This is not the case with Siri dictation + Google.

4

2

46

stephen balaban

@stephenbalaban

30 days

@boborado Nope. Both firms have way higher revenue and also EV / employee Instagram: $76M EV/employee WhatsApp: $345M EV/employee Mystery Co A: $487M EV/employee Mystery Co B: $1.03B EV/employee The “titans” of tech efficiency get rekt.

3

47

stephen balaban

@stephenbalaban

1 year

@Suhail Lots of H100s coming online really soon. 😭 it physically pains me to see Lambda customers not get the GPUs they need.

1

0

47

stephen balaban

@stephenbalaban

1 year

And on top of that, the H100s are $2.40/hour. The lowest price in the world.

1

6

44

stephen balaban

@stephenbalaban

9 months

@karpathy We coined the term at Lambda: GPUnicorn -- you either have $1Bn in GPUs under management or don't.

3

1

45

stephen balaban

@stephenbalaban

9 months

Lambda has a few thousand more H100s coming online before the end of this year --- if you need 64 H100s or more, DM me or just reserve them here:

Lambda GPU Cloud | Reserved Cloud Clusters

Affordable, high performance reserved GPU cloud clusters with NVIDIA GH200, NVIDIA H100, or NVIDIA H200.

lambdalabs.com

Andrej Karpathy

@karpathy

9 months

Who’s getting how many H100s and when is top gossip of the valley rn

54

137

1K

1

4

42

stephen balaban

@stephenbalaban

2 years

A @tszzl tweet got me thinking: roon says that DL offends theoreticians but empirically works to produce "absurd miracles". Engineers have worked to build the "absurd miracle" of commercial aerospace since the early 1900s. Physicists have yet to figure out how lift works.

3

2

43

stephen balaban

@stephenbalaban

8 months

A lot of people don’t appreciate all the magic (read: hard work) behind great products. Even something as ‘simple’ as dimming an LED.

𝕃𝕚𝕩𝕚𝕖 𝕃𝕒𝕓𝕤 🔌❤️💚💙

@lixielabs

8 months

Both dots are orange and dimming, but only the left-hand dot has temporal dithering to preserve color as the brightness drops and provide extra levels of perceived brightness at the same time. The right-hand dot eventually collapses to red when it runs out of color depth. 🎨

6

11

170

1

2

42

stephen balaban

@stephenbalaban

3 months

@GaryMarcus OK, if the vast majority of the artifacts (like walking chairs being pulled from the sand) go away, and more complex inter-object dependencies are reliably visually simulated, do you think that constitutes a world model? I would argue that it does, because you could use it for…

5

10

40

stephen balaban

@stephenbalaban

1 year

The year is 2038, Lambda is the largest GPU cloud in the world. AI produces 42% of global GDP. Your CEO sends you a calendar invite for your quarterly performance review. This is what you see when you get there. What do you say?

5

1

39

stephen balaban

@stephenbalaban

1 month

If you are ever having a bad day just remember that there are arcs of plasma ten times larger than the earth ejecting from our sun. Any problems you have are small in comparison.

ToughSF

@ToughSf

3 years

2 hour timelapse of sunspots on the Sun. The most impressive part isn't the arcs of plasma larger than Earth. It's that the footage was captured by a 150mm telescope with a Hydrogen Alpha filter by amateur astrophotographer David Dayag.

0

15

74

7

2

39

stephen balaban

@stephenbalaban

2 years

Lambda is very proud to be a Diamond tier sponsor of NeurIPS even when the “big guys” aren’t. See you there!

2

1

39

stephen balaban

@stephenbalaban

30 days

@oFFMetaSweat Yea, there are a bunch of super interesting specialty REITs like American Tower (cell towers), Outdoor Media (billboards), Weyerhaeuser (forests), Gaming and Leisure Properties (casinos), Digital Realty and Equinix (data centers).

2

37

stephen balaban

@stephenbalaban

30 days

@boborado Somebody figured it out in the thread. My lesson learned from today is that people on tech twitter need to look outside of the silicon valley echo chamber more. So many great companies in the US that you’ve never heard of. It’s a fun exercise to read through the S&P 500…

2

4

38

stephen balaban

@stephenbalaban

2 months

@saranormous @gradypb @sequoia My guy forgot Lambda. Fixed that!

4

1

38

stephen balaban

@stephenbalaban

1 year

Is there a sight more beautiful?

anton

@abacaj

1 year

All you need is more GPUs

10

2

136

6

2

38

stephen balaban

@stephenbalaban

2 years

Wait until this is in full 4K HD, has significantly better spatiotemporal consistency, and also generates audio alongside. Oh yeah, and you’ll be able to talk to the kangaroos too. The future is gonna be crazy.

Devi Parikh

@deviparikh

2 years

Nine sets of "two kangaroos busy cooking dinner in a kitchen" 🙂 Generated by Make-A-Video. (Montage courtesy Yaniv; This kangaroo example had become our go-to example in the last few days to the deadline :)) #MetaAIMakes