heiner @HeinrichKuttler Twitter profile

Pinned Tweet

heiner

@HeinrichKuttler

9 days

Happy to announce I've joined @elonmusk and @ibab at @xai . Exciting times ahead!

66

38

2K

Last Seen Profiles

@TUKANGCUKURrpw

@bravostonemanor

@pjimiinz

@captude71

@TheSafeAnnaAnon

@melenar

@DEXZIP1

@sakimichanmale

@Mrs_Wilson1979

@smashleybell

@LeratoDrizzy

@jandakembangstw

@MiguelitoBanana

@joekent16jan19

@Soumise01597322

@yukirin2000

@maxWshh

@HamiltionL80853

@northvolt

@100heterotop

@stw_pdg

@knox_ridley

@Latoya331

@luvmdrea

@Chen

@smblawgroup

@flyyjo

@Jespertheend

@k_midori_

@HumbleSignCo

@NautyAssgirl

@thehaalag

@OSU_LA

@sunflewers

@Pepito97_

heiner

@HeinrichKuttler

2 years

New plaything! 📦🎁🚀🐮 We'd like to present 🐄moolib🐄, our distributed ML library.

GitHub - facebookresearch/moolib: A library for distributed ML training with PyTorch

A library for distributed ML training with PyTorch - facebookresearch/moolib

github.com

7

123

783

heiner

@HeinrichKuttler

2 years

@romero John Romero is gonna make you his ... employee!

2

5

198

heiner

@HeinrichKuttler

6 months

Lots of Bellmansplaining on Twitter today.

3

6

142

heiner

@HeinrichKuttler

3 years

Will have to stash up on dad jokes now.

16

0

131

heiner

@HeinrichKuttler

2 months

Today is a bit of an inflection point.

10

1

126

heiner

@HeinrichKuttler

3 years

The real AGI is the friends we made along the way.

0

3

106

heiner

@HeinrichKuttler

18 days

Tomorrow is my last day at @InflectionAI . What a great ride! Some highlights in this THREAD. 1/

4

9

106

heiner

@HeinrichKuttler

2 months

We released an improved MT-Bench at . Some questions and reference answers were nonsensical or wrong before. Here's an overview:

Inflection AI

@inflectionAI

2 months

Evaluation is everything! While testing Inflection-2.5, we found that MT-Bench has a bunch of incorrect answers. Here we share the corrections for everyone to use, and we release a new Physics GRE benchmark for people to try out.

17

48

337

12

9

102

heiner

@HeinrichKuttler

1 year

I wrote a thing about automatic differentiation.

0

16

101

heiner

@HeinrichKuttler

3 years

The difference between RL theory and practise: Discounted visitation frequencies in policy gradient methods. A thread. 🧵

4

11

81

heiner

@HeinrichKuttler

9 days

I'm also in Vienna this week for @iclr_conf . Reach out if you want to chat. And if anyone has advice on moving a family of four to the Bay Area I'm also interested. :)

12

0

89

heiner

@HeinrichKuttler

3 years

Read up on the EM algorithm. (It's all the rage now in RL methods!) This 1998 Neal/Hinton paper is *so clear and readable*, I am amazed. Far more accessible than the Wikipedia article on the topic.

Expectation–maximization algorithm - Wikipedia

en.wikipedia.org

3

76

heiner

@HeinrichKuttler

2 years

Happy to announce I've joined @InflectionAI as a member of the founding team. Let's build something great.

5

0

75

heiner

@HeinrichKuttler

4 years

Spending my lockdown weekends on this *excellent* Physics lecture series by V. Balakrishnan (thanks @j_foerst for the recommendation!) in combination with Sussman and Wisdom's SICM . Scheme is fun! I wish @SymPy was this functional.

Structure and Interpretation of Classical Mechanics

The new edition of a classic text that concentrates on developing general methods for studying the behavior of classical systems, with extensive use of compu...

mitpress.mit.edu

1

9

72

heiner

@HeinrichKuttler

10 months

@_rockt meanwhile:

Riley Goodside

@goodside

10 months

this is wild — kNN using a gzip-based distance metric outperforms BERT and other neural methods for OOD sentence classification intuition: 2 texts similar if cat-ing one to the other barely increases gzip size no training, no tuning, no params — this is the entire algorithm:

152

1K

7K

3

71

heiner

@HeinrichKuttler

2 years

A while back I tweeted about discounting in policy gradient methods and how the policy gradient isn’t even a gradient. With the help of @MetaAI colleague Yann Ollivier, I think I understand what’s going on now. A thread 🧵. 1/14

heiner

@HeinrichKuttler

3 years

The difference between RL theory and practise: Discounted visitation frequencies in policy gradient methods. A thread. 🧵

4

11

81

2

14

63

heiner

@HeinrichKuttler

4 years

Very happy our NLE paper ( #NetHack for RL) has been accepted at @NeurIPSConf 2020. We also worked hard to make it even faster than before; it's now 10x faster. Complex and challenging environments needn't be slow or expensive!

0

9

59

heiner

@HeinrichKuttler

6 months

OK fine I'll tap the sign

4

5

55

heiner

@HeinrichKuttler

4 years

Very happy to see this laborious piece of research get good reviews: RL needs more analysis of quantitative results. Often the tricks that make things work are barely mentioned in our publications as they distract from the story. But they are essential!

What Matters for On-Policy Deep Actor-Critic Methods? A Large-Scale...

In recent years, reinforcement learning (RL) has been successfully applied to many different continuous control tasks. While RL algorithms are often conceptually simple, their state-of-the-art...

openreview.net

0

6

48

heiner

@HeinrichKuttler

1 month

SF has billboards with that paper we wrote @PSH_Lewis @_rockt @olapiktus

2

47

heiner

@HeinrichKuttler

11 days

Transfer of skills (e.g., train on coding to help with 'reasoning') is more often asserted than demonstrated.

Lucas Beyer (bl16)

@giffmana

11 days

Another big, counter-intuitive, take-away: there is no "transfer of skills", multi-tasking merely has "a regularizing effect". This is a bit too subtle to explain on X, but we have 4 completely different experiments leading to the same conclusion, see Sections 5.4.x

3

6

86

2

5

48

heiner

@HeinrichKuttler

3 years

Proud to serve.

Next up for Facebook: Ruining one of history's most beloved RPGs

NetHack is one of oldest, hardest, most pure dungeon-crawlers ever made. So of course Facebook is going to taint it.

www.inverse.com

1

3

40

heiner

@HeinrichKuttler

2 years

Mild disagreement. PEP 8 explicitly makes the opposite idiomatic and for some data structures (e.g., trees) checking emptiness can be O(1) while length is O(n).

Greg Brockman

@gdb

2 years

Unpopular opinion: don't rely on implicit truthy constructs in your language, and instead always convert to bool yourself. For example, in Python rather than "if mylist:", do "if len(mylist) > 0:". An example of trading more keystrokes for less cognitive burden for readers.

19

23

419

3

2

40

heiner

@HeinrichKuttler

3 years

Python _can_ be made GIL-free. 🤯

GitHub - colesbury/nogil: Multithreaded Python without the GIL

Multithreaded Python without the GIL. Contribute to colesbury/nogil development by creating an account on GitHub.

github.com

0

6

34

heiner

@HeinrichKuttler

3 years

Pong is fine I guess but can this method get SOTA on Montezuma's Revenge?

Neuralink

@neuralink

3 years

Monkey MindPong

2K

11K

47K

1

0

29

heiner

@HeinrichKuttler

2 years

@0xabad1dea @sundhaug92 I'd recommend letting the recruiter propose their range first.

1

0

29

heiner

@HeinrichKuttler

6 months

We trained a LLM.

Inflection AI

@inflectionAI

6 months

🎉 Introducing Inflection-2, the 2nd best LLM in the world! Get ready to experience the future of AI with us.

55

125

940

3

2

28

heiner

@HeinrichKuttler

3 years

HUGE congrats to Prof Dr @_rockt for finally beating the game of #nethack and ascending to demigodhood. I now expect an an AI to achieve the same in no time ;)

2

0

28

heiner

@HeinrichKuttler

2 years

That's all! A fully scalable agent in a few lines of code. To learn more about moolib, check out our repo [1], read our whitepaper [2] or look at our API documentation [3]. [1] [2] [3]

moolib: A Platform for Distributed RL - Meta Research

We present moolib, a library that enables the implementation of distributed reinforcement learning and other machine learning codebases.

research.facebook.com

1

4

26

heiner

@HeinrichKuttler

1 year

"Is it AGI" flow chart. Developed with @_rockt at NeurIPS 2022.

1

5

30

heiner

@HeinrichKuttler

4 years

Recently, you have begun to find yourself unfulfilled and distant in your daily occupation. Strange dreams of training, learning, evaluating, and analysing have haunted you in your sleep for many months, but you aren’t sure of the reason. (1/N)

2

9

27

heiner

@HeinrichKuttler

4 years

@soumithchintala @PyTorch

yobibyte

@y0b1byte

4 years

3

28

253

1

2

27

heiner

@HeinrichKuttler

3 years

Thanks for all the great responses to yesterday's thread on discounted visitation frequencies in RL. Here's another 🧵 with some of the papers I learned about this way.

heiner

@HeinrichKuttler

3 years

The difference between RL theory and practise: Discounted visitation frequencies in policy gradient methods. A thread. 🧵

4

11

81

1

3

25

heiner

@HeinrichKuttler

3 years

Too real, stackoverflow, too real.

0

25

heiner

@HeinrichKuttler

2 years

We (Vegard Mella, @erichammy , @DanielleRotherm ) wrote moolib to help with our RL workloads, but it can do much more (e.g., distributed retrieval for knowledge-intensive NLP tasks, ).

3

2

25

heiner

@HeinrichKuttler

3 years

Bellman equation is all you need

6

0

24

heiner

@HeinrichKuttler

3 years

Cough ... TensorFlow ... cough

Paul Graham

@paulg

3 years

Most tricks work better on the stupid than the smart. But one trick that does work on many smart people is making things complicated. Over-engineered systems and over-written prose give them more (though pointless) distinctions to proudly master.

101

363

3K

0

1

23

heiner

@HeinrichKuttler

3 years

<- moved to Munich

2

0

23

heiner

@HeinrichKuttler

2 years

Oh and of course it has @weights_biases integration. 📊📈

heiner

@HeinrichKuttler

2 years

New plaything! 📦🎁🚀🐮 We'd like to present 🐄moolib🐄, our distributed ML library.

7

123

783

0

1

22

heiner

@HeinrichKuttler

2 years

moolib is also the engine behind @BIT_silence et al's recent (excellent) RLMeta library. Check it out!

GitHub - facebookresearch/rlmeta: RLMeta is a light-weight flexible framework for Distributed...

RLMeta is a light-weight flexible framework for Distributed Reinforcement Learning Research. - facebookresearch/rlmeta

github.com

1

4

22

heiner

@HeinrichKuttler

3 years

Impression after two weeks of being a parent: It's more fun than I imagined. Dad jokes come from within. Vis pacem para pacifier.

4

0

22

heiner

@HeinrichKuttler

18 days

What's next? I'll announce that shortly. /fin

3

0

22

heiner

@HeinrichKuttler

18 days

Which brings me to my final thanks: I'm extremely thankful for the opportunity Karén, @mustafasuleyman , and @reidhoffman gave me by adding me to the founding team in early 2022. Being employee number 2 (after @JoeFenton ) was an incredible experience. 6/

2

0

22

heiner

@HeinrichKuttler

7 months

Some progress on NetHack. You love to see it. For context: The AI is still exploring only a small part of this hard game. Models like GPT4 know a lot about NetHack when asked but haven't yet been able to play anywhere near human level.

Pierluca D'Oro

@proceduralia

7 months

Can reinforcement learning from AI feedback unlock new capabilities in AI agents? Introducing Motif, an LLM-powered method for intrinsic motivation from AI feedback. Motif extracts reward functions from Llama 2's preferences and uses them to train agents with reinforcement

15

164

739

2

6

21

heiner

@HeinrichKuttler

18 days

Our latest model Inflection-2.5 () is not bad. In fact, it was the ~4th best publicly "known" models when it was released in early March. And it was created by our pretraining team of < 15 people! 2/

Inflection-2.5: meet the world's best personal AI

We are an AI studio creating a personal AI for everyone. Our first AI is called Pi, for personal intelligence, a supportive and empathetic conversational AI.

inflection.ai

1

21

heiner

@HeinrichKuttler

2 years

moolib is based on async RPCs between peers and supports IMPALA-style dynamic batching. For higher-level usecases, its Accumulator object synchronizes gradients between peers, asynchronously. The accumulator is a state machine with 'wants', 'reduces', and 'has' gradients states.

2

1

19

heiner

@HeinrichKuttler

1 year

Living on the other side of the pond, I never got the full American Thanksgiving experience. Thankfully, these days we have social media to observe and learn.

Gary Marcus

@GaryMarcus

1 year

@ylecun @Grady_Booch @Meta You, my former friend, are burning your reputation to the ground. Everyone is telling you to lie down and go home. Listen to what they are saying. Not for me; for yourself.

6

1

16

2

0

20

heiner

@HeinrichKuttler

3 years

Inspired by @CsabaSzepesvari 's excellent Bandit Algorithms book, here's another _very niche_ blog post: How to show that the Lebesgue measure and integral are well-defined. Many authors make this more complicated than necessary!

0

1

20

heiner

@HeinrichKuttler

11 months

Breaking: DeepMind pivoting to @NetHack_LE .

Edward Grefenstette

@egrefen

11 months

Pleased as punch (the drinky kind, not the hurty kind) to be returning to Google @DeepMind as Director of Research today. It's an exciting time to be helping develop general agents that can adapt to open-ended environments, communicate with us, and help us in novel ways!

51

15

567

2

0

20

heiner

@HeinrichKuttler

18 days

In terms of model quality over time size, Inflection-2.5 is through the roof. How could we train such a good model with such a small team? That's primarily thanks to Jordan Hoffmann. Jordan is amazing and in my opinion one of the world's best AI researchers. 3/

1

0

20

heiner

@HeinrichKuttler

6 months

Teams would have prevented this. I know because it regularly prevents me from meeting people too.

BuccoCapital Bloke

@buccocapital

6 months

Satya invested $10B in OpenAI only for some fake philosophers to use Google Meet to destroy his investment Didn’t even have the decency to use Teams

31

167

3K

0

1

18

heiner

@HeinrichKuttler

2 years

Want to play around with #StarCraft , but 256 colors are just too much and you'd miss @NetHack_LE 's ttyrec replays? And #NetHack is more interesting anyway? I got the solution for you.

0

3

18

heiner

@HeinrichKuttler

3 years

I ❤️ @weights_biases . I also ❤️ that someone puts me in a list together with people like @woj_zaremba and @_rockt 😳

Lavanya 🐝

@lavanyaai

3 years

I could list people doing amazing things using W&B all day. We should probably make this a regular thing! 🙂 Instead I will leave you with some of our users, telling us in their own words what they love about W&B.

1

0

8

1

18

heiner

@HeinrichKuttler

18 days

Forgot to call out one more critical ingredient: @CoreWeave , who are really excellent. Mainly in the form of @sorcer .

heiner

@HeinrichKuttler

18 days

But the rest of the team was also amazing all around. That includes our HPC lead, everyone working on modeling and, dearest to me, the infra folks I had the honor to support and learn from. 4/

1

0

15

2

1

18

heiner

@HeinrichKuttler

2 years

I sometimes complain about unhelpful "pseudocode" in RL papers. So credit where credit is due: The pseudocode hidden in the Supplementary Data of the AlphaStar paper is _excellent_. Kudos to @OriolVinyalsML , @ibab_ml , @trevorycai !

1

0

18

heiner

@HeinrichKuttler

18 days

We built our pretraining and inference stack, partially on top of open source solutions (btw thank you @PyTorch ), partially just writing things from scratch. And we were the first team to train LLMs on H100 GPUs, using Inflection's 22k cluster () 5/

1

0

18

heiner

@HeinrichKuttler

3 years

Update: Little one just turned 6 months (well, 6x4 weeks) and it's better than ever. First tooth! On the verge of crawling. Locomotor skills better than any from deep RL but still with cute failures. Sleep almost not an issue. Happy Father's day everyone!

heiner

@HeinrichKuttler

3 years

Update 4 months in: Having a kid is lots of fun, can still recommend. Richard Ferber has a point. Pat leave is a great invention. I have no idea how > 1 is supposed to even work. :D

0

9

0

18

heiner

@HeinrichKuttler

4 years

New blog post: Capital asset pricing & Fama-French factor models as examples of Linear Regression. Thanks to the @RationalRemind podcast ( @benjaminwfelix , @CameronPassmore ) for teaching me this subject and @egrefen for bugging me to finally write this up.

0

3

17

heiner

@HeinrichKuttler

3 years

First up, a follow-up to Thomas (2014): Nota and Thomas: Is the Policy Gradient a Gradient? (2020) I just love the grumpiness of this one! They quote from a number of well-known RL papers and conclude for each one: "[their] claim [...] is erroneous"!

1

2

17

heiner

@HeinrichKuttler

3 years

Come join us virtually at NeurIPS. Because remember: It's not a coronavir-me. It's a coronavir-us. (h/t @EddyElfenbein )

Tim Rocktäschel

@_rockt

3 years

Join us at the @NeurIPSConf 2020 poster session on Thu 5pm GMT if you want to learn about the NetHack Learning Environment and why we believe a terminal-based procedurally generated game from the 80s is pushing the frontier of single-agent RL research.

0

25

115

1

4

17

heiner

@HeinrichKuttler

3 months

Very cool! There's a certain 2d environment I haven't seen in these tweets though

Tim Rocktäschel

@_rockt

3 months

I am really excited to reveal what @GoogleDeepMind 's Open Endedness Team has been up to 🚀. We introduce Genie 🧞, a foundation world model trained exclusively from Internet videos that can generate an endless variety of action-controllable 2D worlds given image prompts.

144

575

3K

2

0

16

heiner

@HeinrichKuttler

4 years

@SimonDeDeo @peterboghossian @Liz_Shepherd @BretWeinstein You might try serving 3B users with a product developed by tens of thousands of engineers and report back on your failure rate. I get how this looks and I understand this all seems so easy. Until you try that is.

6

0

15

heiner

@HeinrichKuttler

1 year

Life is too short to figure out how Python logging is meant to actually be used.

2

0

16

heiner

@HeinrichKuttler

2 years

Congrats @DeepMind ! Prof @_rockt (DGod) is quite the catch!

Tim Rocktäschel

@_rockt

2 years

After seven years, I have returned to @DeepMind today 🔥 Excited about what lies ahead, and catching up with many old friends and new ones over the coming months!

31

11

713

1

0

16

heiner

@HeinrichKuttler

18 days

But the rest of the team was also amazing all around. That includes our HPC lead, everyone working on modeling and, dearest to me, the infra folks I had the honor to support and learn from. 4/

1

0

15

heiner

@HeinrichKuttler

4 years

@deliprao Now do the reverse! ;)

1

0

15

heiner

@HeinrichKuttler

4 years

AMIGo (work by Andres Campero) is out!

Edward Grefenstette

@egrefen

4 years

Got a complicated RL exploration problem? Sparse/no reward? It's dangerous to go alone: bring an AMIGo! This thread introduces work done by Andres Campero, with @robertarail , Josh B. Tenenbaum, @HeinrichKuttler , @_rockt and me during Andres' internship at FAIR London. [1/5]

4

57

286

1

2

15

heiner

@HeinrichKuttler

3 years

@egrefen @DeepMind We used Generalised Matrix Estimates (GME) which made it converge around 4:20:69 this morning.

0

15

heiner

@HeinrichKuttler

3 years

We had lots of fun doing this interview last year. Thanks, @l2k !

Lavanya 🐝

@lavanyaai

3 years

In today's episode, @l2k interviews @_rockt and @HeinrichKuttler , from the @facebookai team, on how they are leveling the playing field for training RL models with the help of NetHack, an archaic rogue-like video game from the late 80s. #deeplearning

1

8

34

0

15

heiner

@HeinrichKuttler

2 years

@kchonyc But I spend *weeks* on my zsh PROMPT='[%F{green}%~%f: %B%(?.%F{green}.%F{red})%?%f%b] %(!.#.$) '

1

0

15

heiner

@HeinrichKuttler

6 months

just for the record I didn't either

Eliezer Yudkowsky ⏹️

@ESYudkowsky

6 months

Lot of new Twitter followers over the last day. I'm a little sad if that correlates to perceived social power. I did not actually give an order to fire Altman, and if you're here for that, you may as well leave now.

96

12

541

1

0

14

heiner

@HeinrichKuttler

4 years

@_rockt I think part of this is due to our field being driven by clickbait titles. Same reason we show hi-res videos although our agents train on 84x84.

1

13

heiner

@HeinrichKuttler

10 months

@pfau It may work for e.g. theorem proving.

2

0

14

heiner

@HeinrichKuttler

3 years

Burying the lead. @SchmidhuberAI was right all along. LSTM > AGI.

yobibyte

@y0b1byte

3 years

Tired of playing with font sizes and other matplotlib parameters every time you start a new project or write a new plotting function? Use this repo to make your own style file interactively in a jupyter notebook!

6

69

434

2

0

14

heiner

@HeinrichKuttler

1 year

@HeleneBismarck @BorisJohnson If only the Ukrainian ambassador had shared his take on the German government's position at that time, that might have enlightened things.

0

2

14

heiner

@HeinrichKuttler

4 years

@higherpytorch @egrefen @PyTorch

1

0

14

heiner

@HeinrichKuttler

3 years

Huge congrats to @samveIyan for having conceived of and developed MiniHack. Many great people contributed, but it would not have happened without Mika.

Mikayel Samvelyan

@_samvelyan

3 years

Creating rich and complex environments for RL has never been easier! I'm excited to introduce MiniHack: A Sandbox for Open-Ended Reinforcement Learning Research. Code: Paper: Blogpost:

3

26

93

1

2

13

heiner

@HeinrichKuttler

2 years

Here's pseudocode of our prototypical example agents in moolob (make👏everything👏async👏). All peers run this code:

1

13

heiner

@HeinrichKuttler

2 years

Who won the 2020 NetHack challenge? Tune in tomorrow to find out!

UCL DARK

@UCL_DARK

2 years

The NetHack Challenge Website: Online Session: - @erichammy , @MeMohanty , @__dipam__ , @egrefen , @MinqiJiang , @y0b1byte , @HeinrichKuttler , Vegard Mella, @nntsn , @jparkerholder , @_rockt , @robertarail , @DanielleRotherm , @samveIyan

1

9

18

2

13

heiner

@HeinrichKuttler

3 years

I couldn't agree more. @ATabarrok has been one of the bright spots during this gloomy pandemic.

1

3

13

heiner

@HeinrichKuttler

3 years

Light reading over the holidays: Started "Bandit Algorithms" by Tor Lattimore and @CsabaSzepesvari . Like it a lot!

2

1

13

heiner

@HeinrichKuttler

3 years

Indeed. Working with @egrefen and @_rockt is the secret killer feature of FAIR London.

Tim Rocktäschel

@_rockt

3 years

It has been a pleasure to work together with @egrefen on many of these projects.

0

2

18

0

1

13

heiner

@HeinrichKuttler

4 years

What is going on here? You just heard about the NetHack Learning Environment, joint work w/ @nntsn @alex_h_miller , @robertarail Marco Selvatici @egrefen and @_rockt . Paper Code (3/N)

1

2

13

heiner

@HeinrichKuttler

6 months

A quick reminder that those actually building are mostly too busy to tweet.* * Superhuman exceptions exist @elonmusk

4

1

13

heiner

@HeinrichKuttler

3 years

An AI/NetHack connection I didn't anticipate. @NetHack_LE

gryps

@autocastratrix

3 years

every job will be automated until 13 remain: archaeologist, healer, samurai, tourist, valkyrie, priest, ranger, barbarian, monk, caveman, knight, rogue, wizard

127

2K

24K

0

2

13

heiner

@HeinrichKuttler

4 years

@yeewhye Hamming distance must be prime.

0

12

heiner

@HeinrichKuttler

2 years

Apropos of nothing, I got myself a some books about UNIX. (Thanks to @segfaulthunter for advice!)

1

0

13

heiner

@HeinrichKuttler

3 years

Come join us for our #NetHack paper at NeurIPS.

Tim Rocktäschel

@_rockt

3 years

We ( @HeinrichKuttler @nntsn @robertarail @egrefen ) are looking forward to meeting you at our poster A1 in room B3 in two hours With NLE and 2 GPUs you can train deep RL agents at 1,200,000,000 steps a day in a challenging stochastic procgen environment 🚀

1

8

39

0

1

12

heiner

@HeinrichKuttler

1 year

it's actually not all bad.

Mustafa Suleyman

@mustafasuleyman

1 year

Today I’m excited to announce the first version of our new personal AI, Pi... Pi is smart, kind and supportive. It’s designed to be better at natural, flowing conversation than lists, plans, or code.

38

51

346

1

0

12

heiner

@HeinrichKuttler

6 months

@b0rk Also running: "HEAD is a symbolic reference pointing to wherever you are in your commit history." "commits are hashes of a tree + parent(s) + author + timestamp + commit message" "ugit is git in Python!!"

0

12

heiner

@HeinrichKuttler

3 years

Life goals: Write software that makes people feel like @theshawwn feels about Jax on TPUs: 😅😂🤣

0

12

heiner

@HeinrichKuttler

3 years

I know everyone loves #Jax anyway but can I point out that the docs are also really good? Like this "autodiff cookbook"

1

0

11

heiner

@HeinrichKuttler

1 year

Huge result from @MetaAI ! Congrats @polynoamial , @anton_bakhtin , @HengyuanH , @adamlerer , @alex_h_miller , @stephenroller , @joespeez , and everyone else involved!

Noam Brown

@polynoamial

1 year

3 years ago my teammates and I set out toward a goal that seemed like science fiction: to build an AI that could strategically outnegotiate humans *in natural language* in Diplomacy. Today, I’m excited to share our Science paper showing we’ve succeeded! 🧵

130

671

4K

0

1

11

heiner

@HeinrichKuttler

2 years

It's really hard to argue Spain needs to reduce its gas consumption if Germany insists on shutting down its remaining nuclear reactors. Why should others suffer for Berlin's idiosyncratic policy choices?

Javier Blas

@JavierBlas

2 years

NORTH vs SOUTH 2.0: Spain, Greece and Portugal reject the EU call for 15% cuts in natural gas consumption to help Germany Spanish Energy Minister (clearly aiming at Berlin): "Contrary to other countries, Spain hasn't been living beyond its means in energy terms" #EnergyCrisis

708

3K

17K

0

11

heiner

@HeinrichKuttler

4 years

We've always had a thing for ASCII art. (Next version of TB is going to be even better btw.)

Edward Grefenstette

@egrefen

4 years

@facebookai @jelennal_ @CompSciOxford To help research using SOTA distributed RL, in work from @facebookai lead by Heinrich Küttler, we released TorchBeast, a @PyTorch platform for distributed RL (11/16)

1

5

13

1

2

11

heiner

@HeinrichKuttler

2 years

First time in NYC since the pandemic. The security at this Wework took my place of birth as my name; I feel like Don Corleone on Ellis island.

0

1

10

heiner

@HeinrichKuttler

2 months

@tszzl also no orgy after water of life? :/

0

10

heiner

@HeinrichKuttler

3 months

That's pretty cool. Doubly impressive considering it's Google.

Demis Hassabis

@demishassabis

3 months

We have a long history of supporting responsible open source & science, which can drive rapid research progress, so we’re proud to release Gemma: a set of lightweight open models, best-in-class for their size, inspired by the same tech used for Gemini

186

360

2K

1

0

10

heiner

@HeinrichKuttler

4 years

"Finally, we would like to pay tribute to the 863,918,816 simulated NetHack heroes who lost their lives in the name of science for this project (thus far)."

0

9