Pushmeet Kohli @pushmeet Twitter profile | Pikagi

Pikagi

Pushmeet Kohli

@pushmeet

11,114

Followers

97

Following

2

Media

233

Statuses

Computer Scientist, VP Research, Leading Science @ Google DeepMind.

London, England

https://t.co/yHHwBdsQBY

Joined May 2009

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

Pinned Tweet

@pushmeet

Pushmeet Kohli

27 days

We love the excitement & results from the community on AlphaFold 3 and are doubling the AF Server daily job limit to 20. Happy to also share that we're working on releasing the AF3 model (incl weights) for academic use, which doesn’t depend on our research infra, within 6 months.

27

202

904

Last Seen Profiles

@pengen_stw

@PUfmg1

@Gofblug

@Valenti64592242

@omechan48

@TamilAmerica

@jandakembangstw

@CurrumeoMD

@kawaii2k

@RichMot777

@HedgeFundTips

@ExHT_nowPhase3

@StoryHealy

@oauggeae

@Oce_tesfaye

@SIkaputry90752

@7zip

@Maccah55

@LalooSumar22492

@mistressrose129

@xblackdanx

@StiutG

@leonlillie

@rhnnth453044620

@JGeorgeDrawz

@Endyman

@Sa6x3exUjS6BbOQ

@jfwildermuth

@PacificDrive

@TheLibraryDr

@bevditsie

@math_sari

@SPLCUnion

@pjmafioso

@pengen_stw

@evan_jones21

@pushmeet

Pushmeet Kohli

6 months

We at @GoogleDeepMind are excited to announce #GNoME - an AI tool that has discovered 2.2 million new materials, and helps to predict material stability. We're releasing 381K stable materials to help scientists pursue materials discovery breakthroughs.

Tweet card media

Millions of new materials discovered with deep learning

We share the discovery of 2.2 million new crystals – equivalent to nearly 800 years’ worth of knowledge. We introduce Graph Networks for Materials Exploration (GNoME), our new deep learning tool...

deepmind.google

44

377

2K

@pushmeet

Pushmeet Kohli

2 years

AlphaCode - @DeepMind 's programming agent that can achieve human level performance on competitive programming challenges appears on the cover of Science this month. Proud of our team's work on this hard but important problem.

@ScienceMagazine

Science Magazine

@ScienceMagazine

2 years

#AlphaCode —a new #AI system for developing computer code developed by @DeepMind —can achieve average human-level performance in solving programming contests. Learn more this week in Science:

Tweet media one

19

393

1K

19

125

944

@pushmeet

Pushmeet Kohli

2 years

Excited to share our team's work on the AlphaTensor agent that is able to discover previously unknown faster algorithms for Matrix Multiplication – shining a light at the potential of AI and especially RL for making progress on Computer Science itself. #AI4Science

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

2 years

Today in @Nature : #AlphaTensor , an AI system for discovering novel, efficient, and exact algorithms for matrix multiplication - a building block of modern computations. AlphaTensor finds faster algorithms for many matrix sizes: & 1/

114

2K

8K

4

57

606

@pushmeet

Pushmeet Kohli

6 months

Can LLMs uncover new knowledge - the scientific equivalent of AlphaGo’s move 37? In a Nature paper today, we @GoogleDeepMind unveil that FunSearch, our new LLM based approach for program search, has uncovered new results in Maths and Computing. See details below.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

6 months

Introducing FunSearch in @Nature : a method using large language models to search for new solutions in mathematics & computer science. 🔍 It pairs the creativity of an LLM with an automated evaluator to guard against hallucinations and incorrect ideas. 🧵

48

522

2K

10

69

497

@pushmeet

Pushmeet Kohli

3 years

Our team @DeepMind has shown how AI and mathematicians can collaborate together to uncover surprising patterns and make significant new discoveries in different areas of pure mathematics. Read more in our Nature paper:

Tweet card media

Advancing mathematics by guiding human intuition with AI

Nature - A framework through which machine learning can guide mathematicians in discovering new conjectures and theorems is presented and shown to yield mathematical insight on important open...

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

3 years

A new paper in @Nature details how machine learning was used to make significant new discoveries in pure mathematics by guiding the intuition of some of the world’s top mathematicians from @Sydney_Uni and @OxUniMaths : Paper: 1/

Tweet media one

17

613

2K

8

49

423

@pushmeet

Pushmeet Kohli

3 years

Our AI for maths project is on the cover of Nature.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

3 years

#ICYMI : On the cover of @Nature , our team details how machine learning was used to make significant new discoveries in pure maths by guiding the intuition of some of the world’s top mathematicians from @Sydney_Science & @OxUniMaths . More via and below ⬇️

15

83

335

3

31

404

@pushmeet

Pushmeet Kohli

7 months

@GoogleDeepMind is excited to announce #GraphCast - our state-of-the-art AI model that can make medium-range weather forecasts with unprecedented accuracy. We’re open sourcing the model to help scientists study and prepare for extreme weather events.

Tweet card media

GraphCast: AI model for faster and more accurate global weather forecasting

Our state-of-the-art model delivers 10-day weather predictions at unprecedented accuracy in under one minute

deepmind.google

7

62

395

@pushmeet

Pushmeet Kohli

4 years

These incredible results are a testament to our amazing team that worked tirelessly to understand the complexities of protein structures. We look forward to building on this work & opening up new avenues of exploration in the field of biological research. #AlphaFold #CASP14

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

4 years

In a major scientific breakthrough, the latest version of #AlphaFold has been recognised as a solution to one of biology's grand challenges - the “protein folding problem”. It was validated today at #CASP14 , the biennial Critical Assessment of protein Structure Prediction (1/3)

134

3K

10K

12

32

378

@pushmeet

Pushmeet Kohli

3 years

Expanding on research in chemistry, our team @DeepMind has trained neural networks to advance the frontiers of density functional theory. Another example of AI’s potential as a tool to deepen our scientific understanding. More detail in Science:

Tweet card media

Pushing the frontiers of density functionals by solving the fractional electron problem

Deep learning solves a long-standing critical limitation of density functional theory in describing fractional charges and spins.

www.science.org

2

58

374

@pushmeet

Pushmeet Kohli

2 years

Competitive programming is one of the most sophisticated reasoning tasks that humans can accomplish. Really great to share the progress #Alphacode and @DeepMind has made on this problem.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

2 years

Introducing #AlphaCode : a system that can compete at average human level in competitive coding competitions like @codeforces . An exciting leap in AI problem-solving capabilities, combining many advances in machine learning! Read more: 1/

Tweet media one

175

2K

8K

6

23

337

@pushmeet

Pushmeet Kohli

2 years

We have expanded the Alphafold database by including structures for almost every known protein; over 200 million of them. Shining this light at the whole protein universe will enable new directions of scientific research that will increase our understanding of Nature and Life.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

2 years

Today in partnership with @emblebi , we’re releasing predicted structures for nearly all catalogued proteins known to science, which will expand the #AlphaFold database by over 200x - from nearly 1 million to 200+ million structures: 1/

107

2K

6K

4

39

334

@pushmeet

Pushmeet Kohli

7 months

Really happy to share that as part of the Lyria release, our team @GoogleDeepMind have extended SynthID, our state of the art system for watermarking and detecting AI generated content to detect AI generated audio. For more details on SynthID, see:

Tweet media one

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

7 months

Today with @YouTube , we’re announcing Lyria: our most advanced music generation model to date. 🎶 We’re also releasing 2️⃣ AI experiments in close collaboration with participating artists and creators to bring their ideas to life responsibly. →

46

244

1K

3

33

269

@pushmeet

Pushmeet Kohli

5 years

Over several decades, software engineers have developed a toolkit for debugging - from unit testing to formal verification. Our Robust & Verified AI team works on analogous approaches for ensuring that machine learning systems are robust at deployment:

4

59

300

@pushmeet

Pushmeet Kohli

10 months

AI will be a close ally to scientists in their quest to tackle debilitating diseases, and even help us make nuclear fusion - a limitless energy source - viable. I spoke to @newscientist about how @googledeepmind is using AI to take on these challenges:

Tweet card media

The biggest scientific challenges that AI is already helping to crack

AI isn't just for chatbots – many companies are using it to tackle everything from protein folding and drug development to commercially viable nuclear fusion

www.newscientist.com

15

56

299

@pushmeet

Pushmeet Kohli

1 year

AI could play a leading role in solving some of the biggest challenges facing humanity. At @DeepMind , I’ve been privileged to work on impactful innovations from protein folding to nuclear fusion. My thoughts on what we could see for 2023 in @WiredUK :

AI Is Taking On Ever-Larger Puzzles

Machine learning has already proven useful, but its real-world applications are just beginning.

12

53

287

@pushmeet

Pushmeet Kohli

2 years

Magnetic confinement of plasma for fusion is an extremely challenging scientific problem. Really happy to share the progress our team has made in this area - showing how AI can design policies for controlling different plasma configurations in a fusion reactor.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

2 years

Today in @nature , with @EPFL , the first deep reinforcement learning system that can keep nuclear fusion plasma stable inside its tokamaks, opening new avenues to advance nuclear fusion research. Paper:

59

1K

5K

5

28

272

@pushmeet

Pushmeet Kohli

9 months

Honoured to be recognised in @TIME ’s #TIME100ArtificialIntelligence list alongside my @GoogleDeepMind colleagues @DemisHassabis , @LilaIbrahim , @ShaneLegg , @Shakir_za . AI's greatest potential lies in advancing Science for the benefit of humanity.

Tweet card media

TIME100 AI 2023: Pushmeet Kohli

Find out why Pushmeet Kohli made the TIME100 AI list of the most influential people in artificial intelligence.

15

8

260

@pushmeet

Pushmeet Kohli

5 years

Our new results on training verifiably robust neural networks for image classification.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

Neural networks can be vulnerable to adversarial noise. We demonstrate the remarkable effectiveness of interval bound propagation in training provably robust image classifiers:

5

294

827

1

45

253

@pushmeet

Pushmeet Kohli

7 months

Update on the next gen AlphaFold model we have been developing. This model takes structure prediction beyond proteins to general biomolecules like nucleic acids (RNA/DNA), small molecule ligands, PTMs etc. opening us entirely new possibilities in accelerating biological research.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

7 months

Today in partnership with @IsomorphicLabs , we're sharing progress on the next generation of #AlphaFold . 🧬 Our new model greatly expands coverage of structure prediction beyond proteins to other key biomolecular classes. How could it help scientists? 🧵

21

196

758

6

24

249

@pushmeet

Pushmeet Kohli

3 years

Our goal as a company is to build AI that helps accelerate the pace of scientific discovery and advance humanity. Today is just one step towards that goal and I’m very excited to see our team bring #AlphaFold to the world!

@demishassabis

Demis Hassabis

3 years

Last year we presented #AlphaFold v2 which predicts 3D structures of proteins down to atomic accuracy. Today we’re proud to share the methods in @Nature w/open source code. Excited to see the research this enables. More very soon!

Tweet media one

88

2K

6K

3

33

245

@pushmeet

Pushmeet Kohli

4 years

Honoured to be selected as 'one to watch' by @BW 's #Bloomberg50 - thank you for including me!

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

4 years

Huge congratulations to @Pushmeet for being named in @BW ’s #Bloomberg50 list as “one to watch” in 2021! Read more:

0

9

89

8

7

240

@pushmeet

Pushmeet Kohli

1 month

Happy to announce #AlphaFold 3 from @GoogleDeepMind and @IsomorphicLabs . Our new AlphaFold Server illuminates the molecular universe in cells (from proteins to RNA, DNA, ligands and beyond) transforming our understanding of biology. Find out more here ⬇️

Tweet card media

AlphaFold 3 predicts the structure and interactions of all of life’s molecules

Our new AI model AlphaFold 3 can predict the structure and interactions of all life’s molecules with unprecedented accuracy.

7

27

241

@pushmeet

Pushmeet Kohli

5 months

AlphaGeometry from our team at @GoogleDeepMind is really good at solving Olympiad-level geometry problems, and shows the power of neuro-symbolic systems in accomplishing tasks that require sophisticated reasoning.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 months

Introducing AlphaGeometry: an AI system that solves Olympiad geometry problems at a level approaching a human gold-medalist. 📐 It was trained solely on synthetic data and marks a breakthrough for AI in mathematical reasoning. 🧵

127

1K

4K

6

13

217

@pushmeet

Pushmeet Kohli

3 years

Great work from our team @DeepMind . Understanding and predicting weather has been a topic that has intrigued and challenged humanity. Great to see learning based AI systems making progress in this important area.

@demishassabis

Demis Hassabis

3 years

Predicting the weather accurately is important for climate and environment planning. Really proud to see our SOTA generative model for predicting rainfall 1-2 hours ahead (nowcasting) published in @nature . Great work from @shakir_za & team in collaboration w/the @metoffice

9

87

574

3

11

174

@pushmeet

Pushmeet Kohli

3 years

Happy to see Nature Methods publish our team’s work on using AI to improve prediction of how DNA sequence influences gene expression. Our first step in research on improving understanding of the fundamental building blocks of life.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

3 years

Published today in @naturemethods together with colleagues from @calico : Enformer - a transformer model that has led to greatly increased accuracy in predicting gene expression from DNA sequence. Blog: Paper: 1/

Tweet media one

5

237

791

2

18

164

@pushmeet

Pushmeet Kohli

3 years

Big day for team #AlphaFold ! The human proteome & the 20 organisms we are sharing today have been the subject of countless research papers & breakthroughs over time. By sharing this foundational resource, we hope to aid many more scientists in their work; present & future!

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

3 years

Today with @emblebi , we're launching the #AlphaFold Protein Structure Database, which offers the most complete and accurate picture of the human proteome, doubling humanity’s accumulated knowledge of high-accuracy human protein structures - for free: 1/

100

3K

7K

3

20

165

@pushmeet

Pushmeet Kohli

5 years

Our latest paper shows how Neural Nets can learn to reason about graph structured objects - an ability that has implications for a broad spectrum of applications from software analysis to computer graphics. (with Yujia, Chenjie, Thomas and @OriolVinyalsML )

Tweet card media

Graph Matching Networks for Learning the Similarity of Graph...

This paper addresses the challenging problem of retrieval and matching of graph structured objects, and makes two key contributions. First, we demonstrate how Graph Neural Networks (GNN), which...

1

34

160

@pushmeet

Pushmeet Kohli

6 years

Our paper on scalable verification of deep neural networks has just received the best paper award at UAI @ACMUIST . Ensuring AI systems are safe/correct is extremely important - happy to see the community appreciating the research our team is doing at @DeepMindAI .

1

22

148

@pushmeet

Pushmeet Kohli

6 months

Really excited to see Gemini being launched today! Great work by a very passionate team @GoogleDeepMind . The demonstration below from our team shows the power of Gemini in assisting scientists and accelerating Science!

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

6 months

A common problem faced by researchers is finding and using information from scientific papers. 🔍 Gemini's capabilities mean it can accurately uncover knowledge within thousands of files including dense text & complex images. How could this help deliver breakthroughs faster? ↓

47

300

2K

1

12

147

@pushmeet

Pushmeet Kohli

11 months

AI systems sometimes make mistakes. To enable their use in safety critical domains like healthcare, we introduce ‘CoDoC’ - an approach that enables human-AI collaboration by learning when to rely on predictive AI tools, or defer to a clinician.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

11 months

How can AI work better with human experts in areas like medicine where safety is paramount? 🩺 With @GoogleAI , we propose CoDoC: a model that learns when predictive AI is offering correct information - and when it's better to defer to a clinician.

Tweet media one

16

104

416

6

13

130

@pushmeet

Pushmeet Kohli

2 years

Excited about our new wet lab @TheCrick . This partnership will help us in our mission of accelerating scientific progress on our understanding of biology by allowing us to confirm computational predictions with real-world experiments.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

2 years

We’re joining forces with @TheCrick ! We will be setting up a research laboratory to bring together the world-renowned research institute’s expertise with machine learning and AI techniques to advance our understanding of biology. Learn more:

Tweet media one

5

108

462

5

15

125

@pushmeet

Pushmeet Kohli

4 years

I wouldn't be here without my wonderful colleagues and team members @DeepMind . Their passion and focus on tackling hard impactful problems is inspiring.

2

4

119

@pushmeet

Pushmeet Kohli

2 months

It was a pleasure meeting you @GoogleDeepMind . Keep up your great work in clarifying and clearly communicating important ideas across the AI and machine learning community.

@twominutepapers

Two Minute Papers

@twominutepapers

2 months

I have this AlphaFold paper signed by some of the key authors - the now Sir @demishassabis , @pushmeet , and John Jumper. A huge honor, thanks so much! 🙏 I can't believe that this work appeared in my lifetime - and all fully open source! Such an incredible gift to humanity. I am

Tweet media one

5

9

109

2

2

114

@pushmeet

Pushmeet Kohli

4 years

My op-ed appearing in the @sciam discusses how learning systems can discover hidden patterns in large data repositories being gathered by scientists, thus enabling new scientific advances and improving humanity's understanding of our World. Read it here:

2

23

110

@pushmeet

Pushmeet Kohli

9 months

Really happy to share another milestone in our team's long term mission of deciphering the genetic code. By making accurate predictions about missense variants, AlphaMissense has the potential to accelerate understanding of genetic diseases.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

9 months

Uncovering the causes of disease is one of the greatest challenges in genetics. 🧬 To help advance this, we created AlphaMissense: an AI model classifying missense variants - or genetic changes affecting proteins. Here's how it can help scientists. 🧵

23

211

682

1

6

108

@pushmeet

Pushmeet Kohli

26 days

Happy to share that @GoogleDeepMind will be open sourcing our scalable and quality preserving text watermarking method (SynthID_Text) to enable developers to responsibly deploy text generation AI systems.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

26 days

SynthID will now expand to 2️⃣ new modalities: text and video. While this tool isn’t a silver bullet for identifying AI-generated content, it’s an important building block for helping millions of people understand the provenance of synthetic content. →

10

30

178

2

14

103

@pushmeet

Pushmeet Kohli

4 years

Excited to be part of the GPAI's working group investigating the responsible development and use of AI - a topic that has implications for everyone.

@OfficeforAI

Office for Artificial Intelligence

4 years

With great pleasure, we can announce @pushmeet and @RTaylorOpenData will join #GPAI ’s Responsible AI Working Group. Group Co-Chair Yoshua Bengio explains how they're laying the groundwork for future ambitions on Responsible AI & how you can get involved:

Tweet media one

1

6

23

4

15

100

@pushmeet

Pushmeet Kohli

4 years

Happy to see our work (AlphaFold) on protein structure prediction coming out in Nature. It's an extremely important scientific problem and a great example of the projects we are pursuing in the Science team @DeepMindAI .

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

4 years

We have 2 papers published in @nature today! 🎉 One describes AlphaFold, which uses deep neural networks to predict protein structures with high accuracy. AlphaFold made the most accurate predictions at the 2018 scientific community assessment CASP13. 1/4

14

543

1K

0

9

94

@pushmeet

Pushmeet Kohli

2 years

Really proud to see our team's work on #AlphaFold recognized by @ScienceMagazine as the Science Breakthrough of the Year and People's Choice.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

2 years

We are honoured that #AlphaFold has been named Breakthrough of the Year and People's Choice by @ScienceMagazine . Huge congratulations to the team, our partners, and to those around the world building on this work to accelerate biological research. On to 2022!

18

116

619

1

5

89

@pushmeet

Pushmeet Kohli

1 year

It was a pleasure talking to @Analyticsindiam about @DeepMind and some of the work our team is doing to help advance Science, broaden our knowledge, and benefit humanity.

@Analyticsindiam

AIM

@Analyticsindiam

1 year

In an exclusive interview with AIM, Pushmeet Kohli, head of research (AI for science, robustness and reliability), shares research philosophy and AI breakthroughs alongside his contributions in the field. @DeepMind #reinforcementlearning #alphago #alphatensor #AI #BigData

Tweet media one

1

2

13

3

3

78

@pushmeet

Pushmeet Kohli

2 months

Very well deserved. Many Congratulations Sir Demis Hassabis!

@demishassabis

Demis Hassabis

2 months

Delighted and honoured to receive a Knighthood for services to AI. It’s been an incredible journey so far building @GoogleDeepMind over the past 15 years, helping accelerate the field and grow the UK & global AI ecosystems. Thanks to everyone who helped make this dream possible!

361

313

5K

0

1

79

@pushmeet

Pushmeet Kohli

20 days

I recently joined @vijaypande and the @a16z podcast to discuss how AI is advancing the frontier of Science & how we approach this topic at @GoogleDeepMind -listen here: . TLDR: AI is not a nice to have. It's a necessity for us to make sense of our world.

Tweet card media

In recent years, the AI landscape has seen huge advancements, from the release of Dall-E 2 in April 2022 to the emergence of AI music and video models in early 2024. While creative tools often steal...

a16z.simplecast.com

1

10

76

@pushmeet

Pushmeet Kohli

6 months

Gemini read 200,000 (yes, two hundred thousand!) papers over a lunch break, filtered the most relevant 250 papers, extracted the genomics data, and then wrote the script to replot the figure to account for the new data. AI will revolutionize research!

1

7

68

@pushmeet

Pushmeet Kohli

2 years

Honoured to receive the Koenderink test of time award at ECCV 2022 (with Nathan, Derek and Rob) for our work on using colour and depth images for scene understanding. Glad to see the progress that has been in computer vision in the last 10 years since this work was published.

Tweet media one

3

4

68

@pushmeet

Pushmeet Kohli

4 years

Our paper on understanding the dynamics of glassy systems appears on the cover of the April issue of Nature Physics. @DeepMind

@NaturePhysics

Nature Physics

4 years

The physics that underlies the glass transition is both subtle and non-trivial. A machine learning approach developed by @DeepMind is shown to accurately predict the dynamics of glasses over a wider range of temperature, pressure and density

Tweet media one

2

28

68

1

3

68

@pushmeet

Pushmeet Kohli

3 years

Congratulations @broadinstitute on the new Schmidt Center! It’s really exciting to see initiatives that bring together academia and industry, making way for interdisciplinary research that can positively impact the world around us. I look forward to collaborating!

@broadinstitute

Broad Institute

@broadinstitute

3 years

We are thrilled to announce the launch of the Eric and Wendy #SchmidtCenter . Merging biology and machine learning to understand the programs of life, the Center will enable a new field of interdisciplinary research aimed at improving human health.

Tweet media one

5

118

464

2

13

64

@pushmeet

Pushmeet Kohli

4 years

Covid19 is a reminder to our species that we are yet to understand the complexity of Nature, let alone control it. My hope is that this collective experience will spur a new global pursuit of Science and improving our understanding of Nature.

0

9

67

@pushmeet

Pushmeet Kohli

3 years

Our new work on extracting and utilizing symbolic representations of sensory data to enable data-efficient learning - a key requirement for general intelligence.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

3 years

In a new paper, our team uses unsupervised program synthesis to make sense of sensory sequences. This system is able to solve intelligence test problems zero-shot, without prior training on similar tasks:

14

260

1K

1

7

67

@pushmeet

Pushmeet Kohli

3 years

In a year when people lost their lives and livelihood but stock markets rallied; billionaires became richer but children went hungry; you saw your loved ones suffer but couldn’t hug or kiss them. Yes, in this strange year, there is much to be thankful for. As 2020 was also..(1/3)

1

5

65

@pushmeet

Pushmeet Kohli

5 years

Happy to see the release of our work on evaluating mathematical abilities and generalization of neural models (with @dwsaxton , @FelixHill84 and @egrefen ) @DeepMindAI

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

Today we're releasing a large-scale extendable dataset of mathematical questions, for training (and evaluating the abilities of) neural models that can reason algebraically. Paper: Code and data:

Tweet media one

18

792

2K

1

3

59

@pushmeet

Pushmeet Kohli

2 years

Progress on the road to making privacy preserving models as accurate as privacy agnostic ones. Cool work from our team @DeepMind .

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

2 years

Training with differential privacy (DP) prevents models from leaking sensitive training data, but it often incurs a large drop in accuracy. In recent work, our team substantially improved the performance of image classification with DP. Read more: 1/

Tweet media one

18

54

312

2

6

57

@pushmeet

Pushmeet Kohli

3 years

Really excited to see @demishassabis leading @IsomorphicLabs on this ambitious and important mission to accelerate and reimagine drug discovery.

@demishassabis

Demis Hassabis

3 years

Thrilled to announce the launch of a new Alphabet company @IsomorphicLabs . Our mission is to reimagine the drug discovery process from first principles with an AI-first approach, to accelerate biomedical breakthroughs and find cures for diseases. Details:

Tweet media one

69

620

3K

0

1

52

@pushmeet

Pushmeet Kohli

3 years

Happy to see Nature featuring #AlphaFold - our protein structure prediction system that sheds light at the protein universe - the building blocks of life itself.

@demishassabis

Demis Hassabis

3 years

The #AlphaFold 2 papers on the methods and human proteome predictions are out today in hard copy in @Nature ! A really proud moment to see our work featured with a fantastic image on the front cover of the issue:

Tweet media one

35

531

3K

0

7

51

@pushmeet

Pushmeet Kohli

5 years

Enjoyed spending time with @80000Hours discussing robustness & reliability of AI systems and careers in machine learning. Thanks for having me!

@80000Hours

80,000 Hours

5 years

I had a fantastic conversation with @DeepMindAI scientist Pushmeet Kohli about how they're developing new ways to keep AI systems robust & reliable, why it’s a core issue in AI design that everyone has to attend to, and how to succeed as an AI researcher:

0

9

37

1

7

45

@pushmeet

Pushmeet Kohli

5 years

Design (and execution) of build plans for construction tasks is a challenging problem for AI that requires intuitive physics, scene understanding, and planning. Our recent work @DeepMindAI develops AI agents for these problems. (joint work with a very cool team) @jhamrick

@jhamrick

Jess Hamrick is [email protected]

5 years

Excited to share our newest work (& my first final-author paper!) led by Victor Bapst and Alvaro Sanchez-Gonzalez (w/ @CarlDoersch @neuro_kim @pushmeet @PeterWBattaglia ). We investigate structured, object-centric agents that can build things!👷‍♀️🏗

6

35

135

0

6

45

@pushmeet

Pushmeet Kohli

4 years

0

3

44

@pushmeet

Pushmeet Kohli

4 years

Happy to see that the experimentally obtained structure of the ORF3a protein of Sars-cov-2 (recently released by the Brohawn lab) and the structures predicted by our AlphaFold system earlier this year are in agreement. Such results hint at the role AI can/will play in Biology.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

4 years

We’ve posted an updated version of our SARS-CoV-2 structure predictions, and discuss how well our prediction of ORF3a matched experimental data recently released by @kerndm and the Brohawn lab @berkeleyMCB . Read more in the updated post:

4

159

553

0

5

43

@pushmeet

Pushmeet Kohli

6 years

Extremely happy to share the news of the progress we @DeepMindAI are making on leveraging AI for solving one of the most important scientific challenges. Kudos to all the members of our passionate Science team and @demishassabis for his vision and conviction. #AI4Science

@demishassabis

Demis Hassabis

6 years

Proteins are essential to life. Predicting their 3D structure is a major unsolved challenge in biology and could impact disease understanding and drug discovery. I’m excited to announce that we have won the CASP13 protein folding competition! #AlphaFold

53

755

2K

0

1

41

@pushmeet

Pushmeet Kohli

2 years

Congratulations @demishassabis . A very well deserved honour - you are an inspiration leader, exceptional scientist, dear colleague and a wonderful human being.

@fpa

Fundación Princesa de Asturias

2 years

Official reception of @demishassabis , 2022 Princess of Asturias Laureate for Technical and Scientific Research, in Oviedo. #PrincessofAsturiasAwards

Tweet media one

0

4

30

0

1

40

@pushmeet

Pushmeet Kohli

1 month

Thanks for sharing this @jankosinski . Great to see the result AlphaFold3 produces for this system.

@jankosinski

Jan Kosinski

1 month

AlphaFold3 is insane. MutS protein (green) recognizes base mismatches in DNA (gray), after which it slides off to find MutL matchmaker protein (blue), which positions and activates MutH nuclease (pink) to nick a hemimethylated GATC sequence (red) to activate removal and repair of

32

346

2K

1

2

40

@pushmeet

Pushmeet Kohli

5 years

Rigorously testing Machine learning models using meta-learning. We show how a Neural Process-based meta-learning formulation allows us to efficiently search for hard examples. Joint work with: @schwarzjn_ , @agalashov , @hyunjik11 , Marta Garnelo, @dwsaxton , @arkitus , @yeewhye

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

Interested in adversarial tests and reinforcement learning? We combine meta-learning in a general probabilistic paradigm to detect failures, helping us build robust algorithms. Includes results on recommender systems and control: @schwarzjn_ @agalashov

2

61

230

0

4

39

@pushmeet

Pushmeet Kohli

5 years

Happy to share our small step towards enabling more robust software systems by leveraging machine learning to address hard exploration problems encountered in software testing.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

Our latest work uses reinforcement learning to learn to explore graph structured spaces, which improves the coverage for programs and Android apps testing with limited budget. Paper: @hanjundai , @liyuajia , Chenglong Wang, @rishabhs , @huangposen , @pushmeet

6

171

488

0

6

38

@pushmeet

Pushmeet Kohli

5 years

Detecting synthesized images is a challenging problem for machine learning with important social implications. Happy to see the effort Google is putting on this space.

@sundarpichai

Sundar Pichai

5 years

Detecting deepfakes is one of the most important challenges ahead of us. Following our release of a synthetic audio dataset in Jan, we're releasing a large dataset of visual deepfakes to support researchers working on synthetic video detection #GoogleAI

75

519

3K

0

4

38

@pushmeet

Pushmeet Kohli

2 years

Had a great conversation with @craigss for the @EyeOn_AI podcast. We covered everything from how DeepMind’s Science team operates, to our ongoing efforts on some of the most challenging scientific problems.

@craigss

Craig S. Smith

2 years

Listen to @Pushmeet Kohli, head of Science @DeepMind talk about AlphaFold and other #AI wonders. #ML @clearmlapp

1

9

39

1

3

36

@pushmeet

Pushmeet Kohli

3 years

our lack of control; governments rolled out income guarantees to protect the livelihood people; the general population/press focused more on Fauci, Whitty and Vallance and less on the Kardashians; and Science took a key place in public discourse. (3/3)

2

1

36

@pushmeet

Pushmeet Kohli

5 years

Our latest work demonstrates that unlabelled data can be used to improve generalization of models under adversarial evaluation.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

"Are labels required for improving adversarial robustness?" TL;DR: no! With only 10% of CIFAR10 labels, UAT has almost no drop in robust accuracy. With additional unlabeled data, UAT obtains SOTA robust accuracy. Paper: Code:

Tweet media one

1

82

212

0

3

37

@pushmeet

Pushmeet Kohli

5 years

Neuro-symbolic methods are a good candidate for tackling the hard challenges of data efficient learning. See the discussion of our latest work on this topic by MIT tech review below. (joint with Jiayuan, Jiajun, Josh and Chuang)

0

4

33

@pushmeet

Pushmeet Kohli

2 years

Perhaps leaders of all countries in the world should be made to travel to space before they take office. Might be an effective strategy to encourage collaboration in tackling climate change, dealing with bio hazards, eliminating violent conflicts.

Tweet card media

William Shatner experienced profound grief in space. It was the 'overview effect'

William Shatner was excited to go to space last year. He didn't realize he'd be overwhelmed with sadness and go through "the strongest feelings of grief" that he'd ever experienced.

1

2

33

@pushmeet

Pushmeet Kohli

19 days

We’ve integrated AlphaMissense into @emblebi ’s open access tools (Uniprot, AFDB, Ensembl etc). This enables overlay of AlphaMissense predictions onto AlphaFold structures and provides researchers easy access to predicted pathogenicity of genetic variants.

Tweet card media

AlphaMissense data integrated into Ensembl, UniProt, DECIPHER and AlphaFold DB

The integration enables researchers to easily access AI-generated scores estimating how likely genetic variants are to be pathogenic

0

8

36

@pushmeet

Pushmeet Kohli

4 years

Excited to share recent work from our team @DeepMindAI that tries to make sure that ML systems generalize properly to specific variations encountered in the real world.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

4 years

Training data is often collected through a biased process. Models trained on such data are inherently biased. We demonstrate how adversarial training through disentangled representations can reduce the effect of spurious correlations present in datasets:

Tweet media one

1

142

394

0

4

33

@pushmeet

Pushmeet Kohli

8 months

@m_cocheteux @GoogleDeepMind Really nice to hear that people liked the talk. Was great to attend @ICCVConference and see the interesting work happening in the vision community as well as meet many friends, colleagues and former collaborators.

0

1

31

@pushmeet

Pushmeet Kohli

10 months

It was a pleasure to work on this paper looking at AI’s potential to transform scientific discovery. 🔬 From biology to advanced mathematics, researchers now have a hugely exciting opportunity to apply AI tools and take their work to the next level. Read more in @Nature

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

10 months

What could science at digital speed look like? 🚀 AI is poised to supercharge scientific discovery as we know it, by: 🔮 Exploring theories 🧪 Designing experiments 🔍 Analysing data Find out how in @Nature . ⬇️

13

107

477

1

1

29

@pushmeet

Pushmeet Kohli

5 years

We're excited to see accelerating progress on rigorous specification-testing, robust training, and formal verification, from @DeepMindAI and from the broader ML community. If you’d like to work with us, we are hiring! @uesatoj @sgowal Dj Dvijotham

0

2

29

@pushmeet

Pushmeet Kohli

5 years

Exciting new work from our team @DeepMindAI on development of AI systems for language understanding that are "provably" insensitive to even adversarial changes made to the input text.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

Neural networks in NLP are vulnerable to adversarially crafted inputs. We show that they can be trained to become certifiably robust against input perturbations such as typos and synonym substitution in text classification:

Tweet media one

6

220

634

0

0

27

@pushmeet

Pushmeet Kohli

3 years

Really happy to see this. Congratulations John :)

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

3 years

Huge congratulations to John Jumper for being included in the @TIME 100 Next list for his tremendous contribution to #AlphaFold and its potential to accelerate biological research. Very well deserved! #Time100Next

1

62

468

0

1

25

@pushmeet

Pushmeet Kohli

6 years

The responsible approach to developing & deploying AI systems requires Rigorous Evaluation. We reveal our efficient approach ( ) to search for rare failures in RL trained agents at the security workshop at #NeurIPS2018 today @DeepMindAI @uesatoj @ananyaku

Tweet card media

Rigorous Agent Evaluation: An Adversarial Approach to Uncover...

This paper addresses the problem of evaluating learning systems in safety critical domains such as autonomous driving, where failures can have catastrophic consequences. We focus on two problems:...

0

4

24

@pushmeet

Pushmeet Kohli

4 years

SARS-CoV-2 protease/polymerase enzymes are extremely similar to those of SARS-CoV-1 (cause of the SARS outbreak of 2003). Had we sustained research on SARS inhibitors, we would very likely have active coronavirus medicines today.

Novartis R&D heads Bradner and Tsai see hard lessons and hope in coronavirus research

If industry had responded to the first SARS outbreak the way it is now, we would probably have drugs on hand to treat COVID-19. That sobering conclusion wa...

www.biocentury.com

1

2

23

@pushmeet

Pushmeet Kohli

3 years

the year when people across the world came together to defend against a common adversary; people protested against the inequities and biases that exist in the world; intellect (or the lack) of demagogues around the world was exposed; humans realized the power of Nature and..(2/3)

1

1

22

@pushmeet

Pushmeet Kohli

6 years

While re-reading "the selfish gene" by @RichardDawkins , came across the passage "neuron is slower than the transistor but has gone much further in.. miniaturization..you can pack only a few hundred transistors in the human skull" #TechMovesFast

0

1

19

@pushmeet

Pushmeet Kohli

5 years

Its a pleasure to have Satinder () as a colleague. Every meeting with him is a learning experience.

@demishassabis

Demis Hassabis

5 years

Delighted to welcome reinforcement learning pioneer Satinder Singh to @DeepMindAI . He’ll bring some incredible experience to the team and I'm really looking forward to working with him!

3

61

569

0

1

20

@pushmeet

Pushmeet Kohli

5 years

Happy to share the new @WiredUK write-up describing @demishassabis 's vision for AI and the work our team at @DeepMindAI is doing on leveraging machine learning to tackle the most challenging problems in science.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

“Some of the most interesting areas of science are in the gaps between subjects. When that happens, the magic happens.” @demishassabis gives @WiredUK the lowdown on how AI could help solve tough scientific challenges:

19

160

508

0

3

20

@pushmeet

Pushmeet Kohli

6 months

@OriolVinyalsML Drastic Vision, Execution and Leadership in action :) A great result from a great team!

2

0

19

@pushmeet

Pushmeet Kohli

3 years

The RVAI team @deepmind works on robustness and reliability of AI system and is hiring interns. If you are passionate about developing reliable and trustworthy AI systems (especially in NLP) and would like to work with us, please apply below by Dec 17th:

Tweet card media

Google DeepMind

Artificial intelligence could be one of humanity’s most useful inventions. We research and build safe artificial intelligence systems. We're committed to solving intelligence, to advance science...

deepmind.google

1

8

19

@pushmeet

Pushmeet Kohli

3 months

@EBItraining and @GoogleDeepMind have released this guide for anyone interested in learning more about AlphaFold and the protein structure prediction problem. The guide has a collection of interesting examples of AlphaFold in action.

@EBItraining

EMBL-EBI Training

3 months

In our newest practical guide, developed with @GoogleDeepMind and @emblebi , delve into the fundamentals of #AlphaFold , explore its strengths and limitations, and gain practical skills through hands-on exercises. Start the free, self-paced tutorial today:

Tweet media one

2

172

468

0

3

19

@pushmeet

Pushmeet Kohli

6 months

It was great to work on this project with the amazing team of @ber24 , @BarekatainAmin , @Mpawankumar123 , @SashaVNovikov , @matejbalog , @emidup , @franciscuto , @JSEllenberg , @PengmingWang , Omar Fawzi, and @AlhusseinFawzi .

2

0

19

@pushmeet

Pushmeet Kohli

3 years

Great to see this clinical coordination in action - literally saving lives. It is imperative for scientists and clinicians to work together to make progress on this defining health and scientific challenge of our times.

0

0

18

@pushmeet

Pushmeet Kohli

3 years

@shozeb_haider This was a large team effort, Shozeb. I played a part as the lead of DeepMind's Science Program but John Jumper and Demis Hassabis are the ones who should be credited for leading this project.

0

0

17

@pushmeet

Pushmeet Kohli

5 years

The Robust & Verified AI team @DeepMindAI will present its work at @iclr2019 today on extending formal verification of neural models beyond adversarial robustness of image classifiers to a general class of non-linear specifications such as consistency with physics.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

For neural networks to be deployed in the real world, we need guarantees that our network satisfy desired specifications. We introduce a method to verify complex non-linear specifications: (Presented at #ICLR2019 tomorrow, 4:30-6:30pm, Great Hall BC)

4

165

568

0

3

17

@pushmeet

Pushmeet Kohli

5 years

Scientists battling with natural evolution. Machine Learning and AI have a big role to play in the area of drug discovery.

0

1

16

@pushmeet

Pushmeet Kohli

5 years

Ever wondered whether your seq2seq model will halt? Very excited about the work our team has done on answering this key question.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

The latest work of our Robust AI team, towards bringing formal reasoning & termination analysis to ML models like seq2seq that can leverage variable compute and produce variable sized outputs: By Chenglong, @BunelR , DJ, @huangposen , @egrefen , & @pushmeet

1

36

142

0

1

17

@pushmeet

Pushmeet Kohli

6 months

A lot of credit for our work is deserved by @demishassabis for his vision, leadership and unwavering support for building AI that unlocks new scientific discoveries.

0

1

16

@pushmeet

Pushmeet Kohli

4 years

Looking forward to interacting with the computer aided verification community on the challenges we need to overcome to make AI systems Safe, Reliable and Trustworthy.

@confCAV

CAV

4 years

Day 5 (Thursday) of #cav20 . We'll start the day with a keynote by Pushmeet Kohli, DeepMind at Google ( @pushmeet ) titled "Towards Robust Artificial Intelligence via Spec-consistent Machine Learning".

1

1

7

0

0

14

@pushmeet

Pushmeet Kohli

6 years

Happy to see our work on training of verifiably robust models getting recognized at at Security for Machine Learning workshop at #NeurIPS2018

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

6 years

Also, congratulations to the authors of "On the Effectiveness of Interval Bound Propagation for Training Verifiably Robust Models” for receiving Best Paper Award at the SecML workshop!

0

13

55

0

0

14

@pushmeet

Pushmeet Kohli

10 months

Really happy to see SynthID - our invisible and persistent Embedded Watermarking method launching in @googlecloud today. It is powered by our latest research on reliable tagging and detection of AI generated content - critical for enabling responsible use of generative AI.

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

10 months

We’re excited to launch 𝗦𝘆𝗻𝘁𝗵𝗜𝗗 today with @GoogleCloud : a digital tool to watermark and identify AI-generated images. 🖼️ It will be available on Imagen, one of @Google ’s latest text-to-image models. Here’s how it works. 🧵 #GoogleCloudNext

49

211

727

2

0

14

@pushmeet

Pushmeet Kohli

2 years

This work wouldn't have been possible without our wonderful collaborators at @EPFL_en

0

0

13

@pushmeet

Pushmeet Kohli

3 years

Really happy to see AI continuing to contribute to research in Science and Mathematics. Kudos to Alex Davies who led this project and our collaborators Geordie Williamson ( @Sydney_Uni ), and Marc Lackenby & Andras Juhasz ( @OxUniMaths ).

0

0

13

@pushmeet

Pushmeet Kohli

5 years

Can we share control over AI/ML systems? Here we describe our analysis of the security of mechanisms for achieving distributed control over AI/ML models. @DeepMindAI @ShaneLegg @MiljanMartic @janleike @iamtrask

@GoogleDeepMind

Google DeepMind

@GoogleDeepMind

5 years

Scaling shared model governance via model splitting: @MiljanMartic @janleike @iamtrask @ShaneLegg @pushmeet

1

47

170

0

3

12

@pushmeet

Pushmeet Kohli

4 years

@demishassabis Couldn't agree more. The foundation series is a marvel - conveys so many different concepts with such clarity. Asimov was a genius and the foundation series was his best work!

0

0

10

@pushmeet

Pushmeet Kohli

2 years

Was a pleasure chatting with @Sparkes from @newscientist about the work our team @DeepMind is doing on understanding proteins.

@newscientist

New Scientist

2 years

In July, DeepMind announced that its AlphaFold model had worked out how most of the proteins in our bodies fold. @pushmeet tells New Scientist that there is more to come #2021InReview

0

21

62

0

1

9