Denis Yarats @denisyarats Twitter profile

Pinned Tweet

Denis Yarats

@denisyarats

3 months

exciting to announce our new funding and launch of enterprise pro

Perplexity - Enterprise Pro

Perplexity Enterprise Pro

www.perplexity.ai

Perplexity

@perplexity_ai

3 months

We’re excited to announce that we’ve raised $62.7 million in a Series B1 funding led by Daniel Gross. The round also includes Stanley Druckenmiller, NVIDIA, Jeff Bezos, Tobi Lutke, Garry Tan, Andrej Karpathy, Dylan Field, Elad Gil, Nat Friedman, IVP, NEA, Jakob Uszkoreit, Naval

100

247

2K

2

5

64

Last Seen Profiles

@seeminglee

@prioshua

@yoshiakira2ndg1

@raiauad

@BNyzt3

@kmfaho

@mahendrudiksha2

@luisesgo

@avanjogiasslut

@wolannosads

@InsideTexas

@Mingyushiiw

@company_te78589

@bntlry715864583

@6x_xn

@Finety_FN

@AldlmMnsqh76359

@zx1234xxx

@arundoanalytics

@9U0N4T40XR4D8

@AgentRayEpps

@AlbthaBnt27300

@mnmdry22

@WICKSHINER

@bntalry16439641

@huntzerr

@muhammetbalta61

@AshtonGillotte

@NikilisRBX

@CoachLoop

@ladygaga

@sunnivaixchel

@RSKAGY

@Tango21IX

@LipscombBisons

@7mskm

Denis Yarats

@denisyarats

3 years

Thrilled to announce the Unsupervised Reinforcement Learning (URL) workshop at ICML 2021, where we bring together researchers to discuss challenges and possible solutions applying un(self-)supervised learning techniques to enhance RL agents. Website: 1/

2

81

459

Denis Yarats

@denisyarats

4 years

Exciting to announce our new work together with @ikostrikov and @rob_fergus : Image Augmentation Is All You Need: Regularizing Deep Reinforcement Learning from Pixels. Paper: Code: Website: [1/N]

LIVETOTOBET situs slot gacor maxwin

Livetotobet adalah situs slot gacor hari ini terbaru gampang menang yang bekerja sama dengan beragam jenis provider slot online termasuk provider paling populer sekalipun. Permainan kami telah...

sites.google.com

6

97

383

Denis Yarats

@denisyarats

3 years

Excited to release DrQ-v2! DrQ-v2 is more sample efficient, runs 3.5X faster than DrQ, and is the first model-free agent that solves humanoid from pixels. co-authors: @rob_fergus , @alelazaric , @LerrelPinto tech report: code: 1/N

9

45

248

Denis Yarats

@denisyarats

3 years

Currently It is challenging to measure progress in Unsupervised RL w/o having common tasks & protocol. To take a step in addressing this issue we release our #NeurIPS2021 paper: (URLB) Unsupervised RL Benchmark! Paper: Code: 1/N

2

55

230

Denis Yarats

@denisyarats

6 months

Extremely excited to announce our Series B! We really appreciate the trust of our investors and users in giving us an opportunity to advance our mission of building the best answer engine!

Perplexity

@perplexity_ai

6 months

We are happy to announce that we've raised $73.6 million in Series B funding led by IVP with participation from NVIDIA, NEA, Bessemer, Elad Gil, Jeff Bezos, Nat Friedman, Databricks, Tobi Lutke, Guillermo Rauch, Naval Ravikant, Balaji Srinivasan.

145

314

3K

16

2

163

Denis Yarats

@denisyarats

2 years

Currently, Offline RL data is collected under the same reward that is used for evaluation, not ideal... @brandfonbrener and I propose an alternative approach – ExORL, that uses Unsupervised RL & relabeling to construct datasets for Offline RL. paper: 1/10

4

31

155

Denis Yarats

@denisyarats

3 years

Happy to share my new work -- Proto-RL, a task-agnostic pre-training scheme that reconciles exploration and representation learning in image-based RL! with: @rob_fergus , Alessandro Lazaric, and @LerrelPinto . paper: code: [1/N]

3

36

149

Denis Yarats

@denisyarats

5 years

Excited to share our new work lead by @rjerryma , where we attempt to demystify RAdam (Liu et al. 2019) and its automatic learning rate warmup schedule. Paper: [1/4]

4

52

146

Denis Yarats

@denisyarats

7 months

it is incredibly humbling to be mentioned alongside such great companies! a testament to hard work and dedication of the @perplexity_ai team

5

17

137

Denis Yarats

@denisyarats

1 year

At @perplexity_ai , we've been working diligently on our in-house fast LLMs inference infrastructure and gearing up for the training and serving of our own LLMs. Now, with @MetaAI paving the way by open sourcing LLaMA-2, many exciting opportunities have suddenly become available.

4

19

129

Denis Yarats

@denisyarats

4 years

We are releasing a well-tuned and miniature @PyTorch implementation of Soft Actor-Critic () together with @ikostrikov : . We test it on many continuous control tasks from the @DeepMind Control Suite and report the following results:

1

28

130

Denis Yarats

@denisyarats

7 months

Excited to release our online LLMs! These models have internet access and perform very well on prompts that require factuality and up-to-date information. read more here:

Perplexity

@perplexity_ai

7 months

We’re thrilled to announce two online LLMs we’ve trained: pplx-7b-online and pplx-70b-online! Built on top of open-source LLMs and fine-tuned to use knowledge from the internet. They are now available via Labs and in a first-of-its-kind live-LLM API.

59

242

2K

4

6

118

Denis Yarats

@denisyarats

4 months

perplexity runs on kubernetes 🙏

Kelsey Hightower

@kelseyhightower

4 months

I finally got a chance to play with @perplexity_ai and I can see how this is going to replace search engines for some people. Instead of a bunch of links, you get answers, and the sources they were derived from. Best part, no ads, but I'm sure that won't last forever.

50

58

680

3

9

108

Denis Yarats

@denisyarats

6 months

Johnny Ho ( @randomjohnnyh ) was actually the main reason I joined Quora in 2013, instead of Facebook or Google. I didn't know him personally, but I had heard about him from Gena Korotkevich ( @que_tourist ), who lost to Johnny in his last IOI after three consecutive years of

Arman Suleimenov

@suleimenov

6 months

1/ @perplexity_ai , world’s first conversational answer engine, just raised $73M in Series B funding @ $520M valuation. Perplexity’s co-founder, @randomjohnnyh , is IOI 2012 Gold (perfect score: 600/600). In the middle (3rd from the left) of the first picture.

6

12

292

1

3

88

Denis Yarats

@denisyarats

5 years

Excited to announce our new paper on Improving Sample Efficiency in Model-Free Reinforcement Learning from Images with @yayitsamyzhang , @ikostrikov , @brandondamos , Joelle Pineau, and Rob Fergus Paper: Code: [1/6]

2

14

73

Denis Yarats

@denisyarats

18 days

nemotron-4-340B-Instruct is now on -- give it a shot!

Bryan Catanzaro

@ctnzr

24 days

Nemotron-4-340B-Base: * Trained for 9T tokens on 6144 H100 GPUs * Using Megatron-Core at 41% MFU * 96 layers, 18432 hidden state * GQA, Squared ReLU

5

7

49

3

12

73

Denis Yarats

@denisyarats

7 months

It's only been 1 year, can't believe the progress we've made, so happy to be a part of this journey 🚀 Congrats @perplexity_ai team!

Aravind Srinivas

@AravSrinivas

7 months

Today marks the first year anniversary of the launch of , launched on Dec 7, 2022. Lot of people ask me why we ended up being the best product in this category. If I had to pick one word: it is conviction. We believed world needed answers rather than links.

74

54

799

3

7

71

Denis Yarats

@denisyarats

3 years

Excited to see that Proro-RL got accepted to #ICML2021 !

Denis Yarats

@denisyarats

3 years

Happy to share my new work -- Proto-RL, a task-agnostic pre-training scheme that reconciles exploration and representation learning in image-based RL! with: @rob_fergus , Alessandro Lazaric, and @LerrelPinto . paper: code: [1/N]

3

36

149

2

4

64

Denis Yarats

@denisyarats

8 months

Google now indexes @perplexity_ai 🚀 we are so back

5

3

63

Denis Yarats

@denisyarats

3 months

@ylecun @AravSrinivas I think it is a bit exaggerated 🤣

1

2

63

Denis Yarats

@denisyarats

3 years

As a cherry on the cake, we will also have @ylecun as an invited speaker at our workshop!

Denis Yarats

@denisyarats

3 years

Thrilled to announce the Unsupervised Reinforcement Learning (URL) workshop at ICML 2021, where we bring together researchers to discuss challenges and possible solutions applying un(self-)supervised learning techniques to enhance RL agents. Website: 1/

2

81

459

0

6

61

Denis Yarats

@denisyarats

1 year

Excited to announce our Series A funding and the launch of our mobile app!

Perplexity

@perplexity_ai

1 year

Announcing Perplexity AI’s iPhone app and series A funding! Perplexity provides instant answers and cited sources on any topic, now available on iPhone. With follow-up questions, voice search, and thread history, learn and explore faster than ever before.📱

66

214

950

5

4

58

Denis Yarats

@denisyarats

18 days

we've been testing sonnet 3.5 for the last several weeks and have been consistently impressed with the quality of the model. with this new model, we were able to meaningfully improve results on our internal benchmarks. excited for the perplexity users to give this model a spin,

Perplexity

@perplexity_ai

18 days

This model outperforms Claude 3 Opus and GPT-4o on our internal benchmarks.

4

26

192

3

5

54

Denis Yarats

@denisyarats

9 months

API for all the models is coming out soon

Perplexity

@perplexity_ai

9 months

💥 Mistral 7B Instruct is available now. Try it free—

18

49

343

3

2

50

Denis Yarats

@denisyarats

9 months

🚀

2

0

49

Denis Yarats

@denisyarats

7 months

Met someone at NeurIPS yesterday who is a power user of @perplexity_ai , aka tiktok for knowledge 🤯

3

0

48

Denis Yarats

@denisyarats

11 months

It's hard to believe that it's been a year already since it all started! I couldn't ask for better co-founders and team 🚀 @AravSrinivas , @randomjohnnyh & @andykonwinski

Perplexity

@perplexity_ai

11 months

🎂 Celebrating Perplexity's first birthday with a 25% discount on Pro for new subscribers! Use code ONEYEAR25 at checkout. 🔗 Unlock the most powerful AI research assistant with Perplexity Pro. Enjoy enhanced Copilot, GPT-4 access, unlimited uploads, and

11

14

80

4

2

47

Denis Yarats

@denisyarats

2 months

Extremely happy to work with @MParakhin again after all these years and to learn from his wisdom! Fun story: Back in 2011, during his first stint at Bing, Mikhail went on a recruiting trip to Moscow and interviewed many people from Yandex. One of those people was me, and that is

Aravind Srinivas

@AravSrinivas

2 months

Excited to share that we have three new advisors joining Perplexity to help us across search, mobile, and distribution: @emilmichael - former CBO of Uber, @richminer - cofounder of Android, advisor to Google, and @MParakhin - the former CEO of Bing, who launched Bing Chat and

31

40

600

0

46

Denis Yarats

@denisyarats

1 year

@AravSrinivas @minimaxir the fundamental issue is that these libraries (langchain, llama index, etc) are trying to do too many things and have to succumb to a very general interface that is just too rigid for any practical use case, especially in production. they are, however, very useful for educational

2

5

44

Denis Yarats

@denisyarats

9 months

it is interesting how we have double standards for LLMs and general search (e.g. Google). I'm sure you can find all kinds of nasty stuff if you google hard enough and people are OK with this, but if an LLM returns something controversial to a *controversial* prompt it is a

Yann LeCun

@ylecun

9 months

An interesting example of how overzealous safety tuning can reduce the usefulness of AI tools.

44

121

977

4

2

44

Denis Yarats

@denisyarats

5 months

both Gemma 7b and 2b are up on , give it a shot

Jeff Dean (@🏡)

@JeffDean

5 months

Introducing Gemma - a family of lightweight, state-of-the-art open models for their class, built from the same research & technology used to create the Gemini models. Blog post: Tech report: This thread explores some of the

106

827

4K

4

3

44

Denis Yarats

@denisyarats

3 years

@MishaLaskin wrote a nice blog post about our recent work on establishing benchmarking for Unsupervised RL, please take a look:

0

9

42

Denis Yarats

@denisyarats

3 months

dbrx-instruct by @DbrxMosaicAI is available on , enjoy!

Jonathan Frankle

@jefrankle

3 months

Meet DBRX, a new sota open llm from @databricks . It's a 132B MoE with 36B active params trained from scratch on 12T tokens. It sets a new bar on all the standard benchmarks, and - as an MoE - inference is blazingly fast. Simply put, it's the model your data has been waiting for.

33

267

1K

0

3

43

Denis Yarats

@denisyarats

11 months

Added CodeLLaMa-34B-instruct to , give it a spin! Thanks for the great work @MetaAI , @ylecun , @jnsgehring , @syhw et al.

Perplexity

@perplexity_ai

11 months

Code LLaMA is now on Perplexity’s LLaMa Chat! Try asking it to write a function for you, or explain a code snippet: 🔗 This is the fastest way to try @MetaAI ’s latest code-specialized LLM. With our model deployment expertise, we are able to provide you

28

168

695

1

6

40

Denis Yarats

@denisyarats

4 years

People have been asking us whether our (together with @ikostrikov and @rob_fergus ) data augmentation techniques (DrQ ) can also improve sample efficiency in Atari? The answer is YES! See the tweet chain for further information. [1/N]

1

5

39

Denis Yarats

@denisyarats

5 months

top10 🚀

3

1

38

Denis Yarats

@denisyarats

4 years

We extend our data augmentation techniques from DrQ () to on-policy setting which results to much improved generalization on ProcGen! Now DrQ is proven to be effective for a variety of RL algorithms, both off and on policy!

Image Augmentation Is All You Need: Regularizing Deep...

We propose a simple data augmentation technique that can be applied to standard model-free reinforcement learning algorithms, enabling robust learning directly from pixels without the need for...

arxiv.org

Roberta Raileanu

@robertarail

4 years

Excited to share our new paper “Automatic Data Augmentation for Generalization in Deep Reinforcement Learning” w/ @maxagoldstein8 , @denisyarats , @ikostrikov , and @rob_fergus ! Paper: Code: Website:

2

51

224

1

8

35

Denis Yarats

@denisyarats

6 months

@perplexity_ai is on @CNBC

TechCheck

@CNBCTechCheck

6 months

AI-powered search startup @perplexity_ai is looking to take on $GOOGL, raising a $70M+ Series B at a $520M valuation. But Google has left no shortage of challengers in its wake. @dee_bosa spoke to Perplexity CEO @AravSrinivas on why this time is different.

4

11

54

0

35

Denis Yarats

@denisyarats

21 days

🇯🇵

Aravind Srinivas

@AravSrinivas

21 days

Excited to be partnering with @SoftBank to grow Perplexity in Japan!

102

120

3K

1

2

35

Denis Yarats

@denisyarats

5 months

codellama-70b-instruct is now available on

Yann LeCun

@ylecun

5 months

The new breed of CodeLlamas comes with source, like their predecessor.

9

44

375

2

3

35

Denis Yarats

@denisyarats

7 months

What a year it has been! We want to extend our thanks to everyone who has supported us. As a token of our appreciation, we are offering two months of Perplexity Pro for free. Happy Holidays! We can't wait for 2024!

Perplexity

@perplexity_ai

7 months

We've made incredible progress, thanks to you all who have trusted us and provided feedback and support! To thank you, we're offering: Two Months of Perplexity Pro, for free: or use code HOLIDAYS23 in the next 10 days. Happy Holidays!

57

180

1K

2

1

34

Denis Yarats

@denisyarats

3 years

@ylecun @ylecun @AravSrinivas recent Proto-RL () that uses ideas from both CURL and DrQ gets us closer to the LeCake, as it's contrastively learned representations are **fully detached** from RL, besides they can be learned with just the MaxEnt objective.

1

9

33

Denis Yarats

@denisyarats

3 months

Now available on

Yann LeCun

@ylecun

3 months

🥁 Llama3 is out 🥁 8B and 70B models available today. 8k context length. Trained with 15 trillion tokens on a custom-built 24k GPU cluster. Great performance on various benchmarks, with Llam3-8B doing better than Llama2-70B in some cases. More versions are coming over the next

222

1K

7K

3

2

33

Denis Yarats

@denisyarats

7 years

All translation on Facebook is done via neural networks now, this includes our ConvS2S! More details:

0

11

30

Denis Yarats

@denisyarats

6 months

what a surprise to see @perplexity_ai on the @Nasdaq billboard!

1

30

Denis Yarats

@denisyarats

2 months

high praise

Squawk Box

@SquawkCNBC

2 months

Investor Stanley Druckenmiller talks about his investment in AI platform Perplexity and says, "It's unbelievable. It's an answer machine. Nothing like I've ever seen."

20

93

585

2

1

30

Denis Yarats

@denisyarats

5 months

incredible to see @perplexity_ai being feature on the @nytimes front page.

2

3

30

Denis Yarats

@denisyarats

7 months

Try out our 70-billion-parameter model, which has been fine-tuned to work effectively with our search index and is fully served in-house. Our internal evaluations indicate that this model outperforms GPT-3.5-Turbo in several important aspects, including factuality, conciseness,

Perplexity

@perplexity_ai

7 months

Perplexity Pro users can now pick the recently announced in-house model of Perplexity (pplx-70b-online) as the model of their choice! According to human evals, this model is more factually accurate, helpful, concise, and less moralizing than GPT-3.5-turbo for web searches!

17

40

302

4

1

28

Denis Yarats

@denisyarats

8 months

streaming a 7b model at 220 tokens/sec 🔥 try it out at

3

0

29

Denis Yarats

@denisyarats

5 years

Our work which attempts to use natural language as a communication channel between a hierarchy of control policies in a complicated RTS game was accepted to #NeurIPS2019 . Blog post: Paper: Source code:

GitHub - facebookresearch/minirts: We release dataset collected for our research, code that...

We release dataset collected for our research, code that implement neural network models described in the paper, and scripts to reproduce all of our results, and visualization tool for visualize da...

github.com

0

2

25

Denis Yarats

@denisyarats

1 year

Excited to roll this out! Not the optimal configuration just yet, but should be a good way to start. We will greatly improve latency and throughout over the next week or so. Stay tuned!

Perplexity

@perplexity_ai

1 year

🚨 LLaMa-2-70B-Chat is available now! Try it here: 💬 What will you ask first?

42

190

814

1

27

Denis Yarats

@denisyarats

9 months

Excited to release our LLM API! It's been an exciting journey to build the infra from the ground up and scale it to be able to serve billions tokens daily.

Perplexity

@perplexity_ai

9 months

Introducing pplx-api, our LLM API which serves Mistral and Llama2 models with blazing speed and throughput. pplx-api is in public beta for our Pro subscribers! We partnered with @nvidia and @awscloud to build our proprietary inference. Learn more:

24

74

475

1

26

Denis Yarats

@denisyarats

8 months

@elonmusk how about comparing to @perplexity_ai

what did joe rogan wear in his latest interview?

In his latest interview, Joe Rogan was seen wearing a blonde wig and a Puerto Rico shirt. This was during his podcast with Elon Musk, which was conducted to...

www.perplexity.ai

1

0

26

Denis Yarats

@denisyarats

1 year

Our goal at @perplexity_ai is to provide information that you can trust. It is a difficult problem, but we are working hard to ensure that our UX, search, and LLM all work together to provide accurate and reliable information. Today, we're taking another step in that direction:

Perplexity

@perplexity_ai

1 year

Introducing a new editing tool for : you can now edit answers by either adding sources for more perspectives or deleting irrelevant sources. With just a few clicks, you can add context, remove incorrect information, and curate trustworthy answers:

21

40

298

0

2

26

Denis Yarats

@denisyarats

9 months

Very cool to see that Google itself chose to use @perplexity_ai as a worthy baseline 🚀 More importantly, this paper shows how much alpha there is in combining vanilla search with an LLM

TDM (e/λ)

@cto_junior

9 months

Interesting, a paper by google that evaluated @perplexity_ai on their new eval. It performs really well for factual Q&A and debunking, better than vanilla google search.

7

25

268

3

1

25

Denis Yarats

@denisyarats

8 months

got quite a bit faster today 🔥

Perplexity

Perplexity is a free AI-powered answer engine that provides accurate, trusted, and real-time answers to any question.

www.perplexity.ai

2

0

24

Denis Yarats

@denisyarats

6 months

@sadlyoddisfying @perplexity_ai we don't control this, this seems like an iOS bug, could you please restart your iphone.

1

24

Denis Yarats

@denisyarats

1 month

now @WSJ runs LLM evals too 🤯

3

5

24

Denis Yarats

@denisyarats

8 months

yikes on the other hand, @OpenAI API has been flawless since all this started...

2

1

22

Denis Yarats

@denisyarats

7 months

@AravSrinivas @ylecun if in doubt remember this

2

0

23

Denis Yarats

@denisyarats

3 years

Today at 12PM PDT together with @ikostrikov we will be giving a spotlight presentation on DrQ () at ICLR 2021 (Oral Session 8). Consider attending if you want to lear more about our work.

0

4

23

Denis Yarats

@denisyarats

1 year

I certainly didn't expect to see our users complaining about our LLaMA-2 inference () being too fast for them. I suppose we have to make it slower. 🤔

4

0

22

Denis Yarats

@denisyarats

1 year

70B is on the way!

Perplexity

@perplexity_ai

1 year

It's live! LLaMa-13B is now available on

16

36

215

3

0

22

Denis Yarats

@denisyarats

1 year

@sherjilozair @perplexity_ai we built our own

1

0

22

Denis Yarats

@denisyarats

3 months

@ylecun @AravSrinivas iirc the demo that did the trick was to ask Perplexity's BirdSQL who are the people who follow Yann but don't follow Gary Marcus on Twitter 🤣

2

1

22

Denis Yarats

@denisyarats

1 year

I'm super excited about @Perplexity Copilot. It offers a completely new way of web searching via an interactive interface filled with rich UI components, not just text. Copilot browses and analyzes tens of web pages on your behalf, ultimately producing a thoroughly researched

Perplexity

@perplexity_ai

1 year

The next iteration of Perplexity has arrived: Copilot, your interactive AI search companion. 🚀🤖 Perplexity Copilot guides your search experience with interactive inputs, leading you to a rich, personalized answer, powered by GPT-4. Try it for free at

94

345

2K

0

3

21

Denis Yarats

@denisyarats

7 months

the mistral-medium, the strongest @MistralAI 's LLM, is now available on

1

0

19

Denis Yarats

@denisyarats

11 months

@AviSchiffmann use instead

1

0

21

Denis Yarats

@denisyarats

4 months

Perplexity "Discover Daily" is top 10 in News on Apple Podcast 🚀

2

0

21

Denis Yarats

@denisyarats

6 months

👀

0

20

Denis Yarats

@denisyarats

1 year

Exciting news! We've just launched a series of updates, including a user sign-up feature. This will allow us to enhance personalization and enable us to deliver advanced features to our users! 🚀

Perplexity

@perplexity_ai

1 year

We are excited to launch the next version of ! Introducing login, threads, focus search, improved formatting, and more. 🎉 Login now to start collecting your own library of threads. Keep them to yourself, or share your latest discovery. You decide.

30

121

636

1

19

Denis Yarats

@denisyarats

3 months

You can try it out on (it is prompted base model, postrained model is in progress)

Mistral AI

@MistralAI

3 months

RELEASE 0535902c85ddbb04d4bebbf4371c6341 lol

29

32

897

0

3

20

Denis Yarats

@denisyarats

3 months

one-click perplexity

Joe Santucci

@ATXsantucci

3 months

I use @perplexity_ai so much it has a dedicated button on my mouse. #SmartActions on Logi Options+ means all I have to do is highlight text, press my button and let Perplexity do the rest.

8

122

3

0

21

Denis Yarats

@denisyarats

3 years

If you are interested to learn more about unsupervised pre-training in image-based RL, consider to attend our #ICML2021 poster session at 9am PST / 12pm PST.

Denis Yarats

@denisyarats

3 years

Happy to share my new work -- Proto-RL, a task-agnostic pre-training scheme that reconciles exploration and representation learning in image-based RL! with: @rob_fergus , Alessandro Lazaric, and @LerrelPinto . paper: code: [1/N]

3

36

149

0

20

Denis Yarats

@denisyarats

5 months

someone created a Wikipedia page for @perplexity_ai

1

20

Denis Yarats

@denisyarats

7 months

great presentation by @AravSrinivas on how we @perplexity_ai leveraging @awscloud :

AWS re:Invent 2023 - Keynote with Dr. Swami Sivasubramanian

Swami Sivasubramanian, Vice President of Data and AI at AWS explores the powerful relationship between humans, data, and AI, unfolding right before us. Gener...

www.youtube.com

0

1

19

Denis Yarats

@denisyarats

7 months

@j_foerst @sundarpichai @perplexity_ai Thanks for the high praise @j_foerst

3

1

19

Denis Yarats

@denisyarats

5 months

@teja2495 @perplexity_ai haven't shipped all the recent improvements yet, additional 100ms latency reduction is coming. also iOS is about to be an instant answer machine.

1

0

19

Denis Yarats

@denisyarats

1 year

@MichaelRoyzen not true, also not a great strategy to speak negatively about competitors. p.s. running fine-tunned flan-t5 in house with Triton is not that hard, so I wouldn't really brag about it :D

3

0

18

Denis Yarats

@denisyarats

6 months

@ylecun @AravSrinivas thanks @ylecun for your support from the very beginning! I still remember how @AravSrinivas and I were showing your our first twitter search demo back at NYU 🤣

1

0

17

Denis Yarats

@denisyarats

4 years

In the end it is more important that your work is actual being used by others that having it accepted at a ML conference.

Homanga Bharadhwaj

@mangahomanga

4 years

@jachiam0 Another paper that comes to mind is by @denisyarats @yayitsamyzhang Kostrikov @brandondamos Pineau @rob_fergus Although it was rejected from ICLR last year (novelty!), it inspired a host of pixel-based RL algorithms like CURL, RAD, DrQ...this year

1

4

18

0

2

18

Denis Yarats

@denisyarats

4 months

the gmail creator is speaking

Paul Buchheit

@paultoo

4 months

When I joined Google in 1999, I didn't understand how it was possible to win vs much larger and better funded competitors such as Altavista, but it seemed like a good opportunity to learn. What I learned is that if you're in the lead, and you're moving faster than everyone else,

74

170

2K

0

18

Denis Yarats

@denisyarats

3 years

Please join us on our ICML workshop on Unsupervised RL today at 8:45 EST. We have an amazing list of speakers and oral presentations, as well as a poster session! ICML: Website:

Unsupervised Reinforcement Learning (URL)

An ICML 2021 Workshop on Unsupervised Reinforcement Learning

urlworkshop.github.io

Denis Yarats

@denisyarats

3 years

Thrilled to announce the Unsupervised Reinforcement Learning (URL) workshop at ICML 2021, where we bring together researchers to discuss challenges and possible solutions applying un(self-)supervised learning techniques to enhance RL agents. Website: 1/

2

81

459

0

2

17

Denis Yarats

@denisyarats

1 year

I'm especially excited to use this to read articles on ! Try it out!

Perplexity

@perplexity_ai

1 year

👋 Say hello to a more personalized browsing experience with our updated Chrome extension! can now provide answers focused on the page or website you're currently looking at. No more sifting through irrelevant search results:

31

125

703

0

2

17

Denis Yarats

@denisyarats

6 months

thank you for the high praise!

tobi lutke

@tobi

6 months

and its app have definitely replaced my google usage for now. It’s pretty incredible.

136

138

2K

3

0

15

Denis Yarats

@denisyarats

1 year

@soumithchintala rlhf doesn't need a lot of data, discrimination is much more sample efficient than generation...

1

0

16

Denis Yarats

@denisyarats

2 years

Another update from our team on how to use Bird SQL to analyze Twitter data.

Perplexity

@perplexity_ai

2 years

With Perplexity, you can search over databases. But also do more! Visualize & Summarize results as aggregate stats, plots Everyone gets to analyze data from natural language. No prerequisite knowledge of SQL or plotting libraries necessary. Link:

16

71

452

1

16

Denis Yarats

@denisyarats

3 years

Key insights while working on DrQ-v2: 1. n-step returns with DDPG is much better than SAC. 2. 10x bigger replay buffer is needed for better results. 3. decaying variance of exploration noise allows to gain the last bit of performance. 2/N

1

0

16

Denis Yarats

@denisyarats

11 months

🚀

Phi Hoang

@apostraphi

11 months

. @perplexity_ai made it on the @TODAYshow as one of the AI tools that can help make your life easier. ☀️🤯

0

1

14

1

0

16

Denis Yarats

@denisyarats

2 years

We've built another cool search interface, this time over structured data. Please take a look!

Perplexity

@perplexity_ai

2 years

Introducing Bird SQL, a Twitter search interface that is powered by Perplexity’s structured search engine. It uses OpenAI Codex to translate natural language into SQL, giving everyone the ability to navigate large datasets like Twitter.

226

2K

9K

0

2

15

Denis Yarats

@denisyarats

4 years

Check out our open soursed code for DCEM and some cool examples of its usage:

Brandon Amos

@brandondamos

4 years

With @denisyarats , we've released the @PyTorch code and camera-ready version of our #ICML2020 paper on the differentiable cross-entropy method. Paper: Code: Videos: More details in our original thread:

1

43

134

0

14

Denis Yarats

@denisyarats

3 years

We have an exciting set of speakers ( @pabbeel , @KelseyRAllen , @xkianteb , @chelseabfinn , @hardmaru , @danijarh , @rosemary_ke , @alelazaric ) that will bring a diverse perspective on the challenges of unsupervised exploration and representation learning in RL. 2/

1

0

15

Denis Yarats

@denisyarats

4 months

Thanks Karri & the team for building @linear , we can't live without it!

Karri Saarinen

@karrisaarinen

4 months

Was trying to find a story about a director who always wanted to change one thing. The designer added a duck to the game character so it'd be the most obvious thing to change. Google failed horribly. Tried lot of ways and keywords. @perplexity_ai found it in one single ask

25

10

239

0

15

Denis Yarats

@denisyarats

4 months

a thought provoking point by @adamdangelo

a16z

@a16z

4 months

CEO of @Quora & @Poe_Platform @adamdangelo argues that startups are uniquely positioned to provide a level of fault tolerance unattainable by incumbents. He & @DavidGeorge83 discuss getting AI infrastructure to the masses & its multi-model, multi-modal future.

11

16

95

0

3

15

Denis Yarats

@denisyarats

5 months

Now, you can set @perplexity_ai as the default search engine on the amazing @arc browser!

The Browser Company

@browsercompany

5 months

Your search engine, your internet @perplexity_ai is now a default search engine option 🌐 Live in Arc 🌐

73

98

2K

1

0

13

Denis Yarats

@denisyarats

4 years

We identify that overfitting is a major problem in RL from pixels, as increased model capacity negatively correlates with performance. This makes sense, as an off-policy RL initially trains on a very small replay buffer and can easily overfit on it: [3/N]

1

14

Denis Yarats

@denisyarats

3 years

Pre-training strikes again!

Lerrel Pinto

@LerrelPinto

3 years

Excited to release some of our latest work on visual imitation! We surprisingly find that if you have the right visual representation, even simple k-NN regression works well for both offline imitation and real robotic tasks. paper+code+data: (1/5)

2

46

237

0

4

13

Denis Yarats

@denisyarats

7 months

the OpenAi board meeting featuring shoggoth is in progress

0

13

Denis Yarats

@denisyarats

2 years

An awesome new algorithm for unsupervised RL from @MishaLaskin et al, a great addition to our URLB library of agents ()!

GitHub - rll-research/url_benchmark

Contribute to rll-research/url_benchmark development by creating an account on GitHub.

github.com

Misha Laskin

@MishaLaskin

2 years

New paper on unsupervised skill discovery - Contrastive Intrinsic Control. Tl;dr exploration with contrastive skill learning substantially improves prior skill discovery methods (by 1.8x)! Achieves leading unsupervised RL results. Learn more 👇 1/N

2

20

106

0

2

14