Elman Mansimov
@elmanmansimov
Followers
2,605
Following
668
Media
115
Statuses
1,357
AI scientist whose work has been acquired by a prestigious American museum
New York, NY
Joined January 2016
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Xóchitl
• 504163 Tweets
Maynez
• 180575 Tweets
Wizkid
• 119555 Tweets
Suns
• 106772 Tweets
Real Life
• 93207 Tweets
DDAY SB19 JAPAN CONCERT
• 91225 Tweets
Davido
• 76140 Tweets
Anthony Edwards
• 74306 Tweets
Beal
• 54431 Tweets
#sbhawks
• 32751 Tweets
為替介入
• 31171 Tweets
Twitter調子
• 26464 Tweets
スレイヤー
• 19331 Tweets
Don Jazzy
• 17114 Tweets
#MondayMotivation
• 12581 Tweets
ホークス
• 11896 Tweets
グリフィン
• 11859 Tweets
#リクステ
• 10161 Tweets
Pinned Tweet
Images from Text:
how it started (2015) how it is going (2021)
12
328
2K
It is the internship application season and I want to share advice on landing summer industry research internships in ML.
Coming from a person who had several great ML internships in tech👇
11
88
514
In the last 7-8 years we went through multiple cycles of generative models of images: VAEs, Normalizing Flows, GANs, AR models, now Diffusion models
In the meantime left-to-right autoregressive models remained state-of-the-art in the NLP all this time
What is the lesson here?
8
17
239
Natural Language Generation has been dominated by left-to-right autoregressive models like GPT-2. However it is not the only way to do it and lots of exciting work been done over last ~2 years on alternative approaches with their own pros&cons. Here is my attempt on grouping them
6
43
217
Need an OpenAI Codex tool for converting my paper from one conference format to another.
4
3
189
Personal News:
I have successfully defended my Ph.D. thesis at the end of last year!
I am enormously thankful to my advisor
@kchonyc
, committee members
@hhexiy
, Yaser Al-Onaizan,
@rob_fergus
,
@dipanjand
and my collaborators at NYU for making this happen! 1/3
13
3
157
Someone decided to submit well known LSTM as a paper to ICML 2020🤦. This trolling attempt is just a waste of everyone's time.
4
12
88
I want to big round of applause to the
@fellowshiptrust
team for executing technical details of the alignDRAW auction amazingly well.
Just look at the details of alingDRAW process prompt image cloud. Every little detail from displaying each image piece to showcasing which ones…
7
9
82
GPT-0 is actually a paper from Bengio from 2003
This paper was released in 2016. It's basically GPT-0.
5
30
300
2
7
77
very happy to see the value of the alignDRAW collection grow!
it is a win for everyone from collectors,
@FellowshipAi
, me, and general AI art community.
Looks like people are finally realising the historical value of alignDRAW by
@elmanmansimov
. Great to see
@FellowshipAi
0
4
29
9
12
79
very proud how far my work on text-to-image generation has come.
i never expected that a research project will be one day considered a significant historical artifact and the generated images be shown at largest photography fair
@ParisPhotoFair
let me give you a backstory:…
7
9
74
Mom my work just made it to the museum 🤯
We are delighted to announce that several pieces from
@elmanmansimov
´s alignDRAW project have been acquired by an American Museum. Our mission is to connect digital art to institutions looking to expand their collections by embracing new technologies like AI. Congratulations to…
12
8
84
10
7
71
Celebrating and digesting the digital release of alignDRAW tonight with my girlfriend
Promise will get back to you all tomorrow
Gn 🫡
11
1
70
I logged in to Facebook for the first time in ages to read this post by Tomas Mikolov who is a pioneer in deep learning and NLP reflecting on word2vec and his other achievements in light of test of time paper award
I got a chance to meet Tomas last summer in Prague (thanks…
2
6
68
@martin_casado
"MAGIC would be the only institution in the world permitted to develop advanced AI"
🚩🚩🚩
2
2
60
Excited to finally talk publicly about museum that acquired alignDRAW generated images back in November
And it is 🥁🥁🥁
Worcester Art Museum in Massachusetts
My work being part of museum's portfolio is not something I have imagined to happen in my life, and it's truly…
alignDRAW acquired by the Worcester Art Museum
@Fellowshiptrust
is very excited to announce the acquisition of three paper prompt grids by
@elmanmansimov
from his ground breaking project alignDRAW. The artworks will be included in an upcoming show entitled New Terrain:…
22
18
127
8
3
57
latest work w
@W4ngatang
@kchonyc
on how to view many sequence generation algorithms under one framework and applying them to single trained undirected seq model (multilingual translation variant of BERT)
1
12
54
Excited about my latest work with
@GaborMelis
& Lei Yu where we make sentence-level NMT model better suited for document translation by *self-training* it on generated sentences. This procedure makes more likely that model stays consistent across document.
0
16
52
And the
@ChristiesInc
alignDRAW auction is closed!
Congrats to the winner of the auction and thanks
@fellowshiptrust
and
@ChristiesInc
for organizing it.
I am genuinely blown away!
12
6
50
this was an interesting tweet that caught my attention and deserved a way longer reply than yeah keep being skeptical time will show you are wrong
ofc everyone is entitled to their opinion but I would argue that AI art is much more of an "art" that we seen before
- First AI:…
@hofagallery
@emilyxxie
@sougwen
@kevinabosch
@elmanmansimov
@niceaunties
@agoriamusic
AI art is not real art and AI artists are not real artists.
3
0
28
6
14
47
Was great to work on a problem outside of commonly used domains such as text, img, speech which in our case is a generative model of molecular geometry with omar, seokho and
@kchonyc
0
7
45
head of photography at Christie’s
@ChristiesInc
@dariushimes
talking about my 2015 work on text to image generative model. that’s something that wasn’t on my 2023 bingo cards 🤯
@halecar2
did a fantastic job explaining the work. give it a listen!
Head of Photography at Christie's
@dariushimes
talks with our co-founder
@halecar2
about one collection exhibited at our booth: "alignDRAW" by computer scientist
@elmanmansimov
.
This collection of 101 pieces of 32 pixels square represents the birth of Al-generated images from…
17
38
173
5
5
47
God give me confidence of those who think ChatGPT will soon replace traditional search
2
0
45
Subject: PhD student with XYZ at ABC very interested in internship with you
Body: Hi ..., Hope all is well
[Brief intro about yourself]
[Saw your papers (paper A and B) and I liked them. I did work in that area as well (paper C). Would be awesome to work together]
1
0
44
I have been browsing old photos from 2015 from my time in Toronto.
And found the photos of my old desk inside Pratt Building at 6 Kings College Road at University of Toronto. This is the birthplace of the alignDRAW project, where the idea was conceived and brought to fruition.…
3
6
44
Turns out iterative refinement can not only give a decoding speed benefit, but perform better (esp much better on low-resource setting) than strong autoregressive semantic parsers on the conversational domain.
One of my first pieces of work done at AWS
2
5
43
Start with identifying people whose work you admire and want to work with.
You should already have a rough list of people in mind based on your interests.
1
1
41
~160k USDC worth of alignDRAW value exchanged on the market during the week is definitely above my initial expectations
Interesting times in digital art! AI art collections are comparable to top generative art collections in weekly volume.
top-4 alignDRAW by
@elmanmansimov
@FellowshipAi
@fellowshiptrust
top-5 DeepBlack by
@DeepBlackAI
2
6
40
4
3
41
In anticipation of alignDRAW work being auctioned with
@fellowshiptrust
next week, I want do a series on behind the scenes of alignDRAW prompts released back in 2015
I will start with "a stop sign is flying in blue skies" that is currently auctioning with
@ChristiesInc
To give…
5
7
41
Got a chance to see alignDRAW in person in CDMX. As breathtaking as I saw online.
But most importantly got to meet
@halecar2
and many artists
@niceaunties
in person!
3
4
41
this is amazing!
"A group of happy elephants in the green grass field
#15
"
Original, then recursively upscaled with the
@krea_ai
"Enhancer" functionality, using the same original prompt as is
@elmanmansimov
you might enjoy this
10
12
106
1
3
37
@ivanskorokhodov
My advice:
TLDR try connecting with people you would want to work with first before applying
It is the internship application season and I want to share advice on landing summer industry research internships in ML.
Coming from a person who had several great ML internships in tech👇
11
88
514
0
0
39
At the very last day of this decade our paper w Omar, Seokho and
@kchonyc
on molecular geometry prediction with generative graph neural nets has been published to
@SciReports
. Check it out here . Happy New Year everyone!
0
8
38
Run ChatGPT locally in your terminal with 𝗰𝗵𝗮𝘁𝗴𝗽𝘁 command
~100 lines of shell wrapped around the official ChatGPT API
Source code
0
3
37
In celebration of the full auction of all the alignDRAW pieces by
@fellowshiptrust
on daily xyz I want to share few more stories about alignDRAW prompts
This one is on "A person skiing on sand clad vast desert"
I remember when I first generated those images I was immediately…
1
8
38
During summer 2019, together with Mitchell,
@orf_bnw
,
@MiaXuChen
, Jakob & Puneet at Google, we worked on an ambitious way of tackling in-image translation (translate text in the image and generate the same image with translated text) using the end-to-end neural approach. [1/2]
1
3
38
thank you NFT community 🫶
Congrats
@elmanmansimov
! This level of support from the NFT community is amazing to see. Good luck with the auctions!
6
7
54
3
5
38
I am attending
@naaclmeeting
next week.
First in-person conference since Ph.D. graduation and starting to work at AWS AI.
Felt like a good time to reflect on the work me and the team has done at
@AmazonScience
for the last year and a half 🧵
2
4
38
@kevinabosch
@FellowshipAi
@ParisPhotoFair
i never thought of myself as an artist, i see myself as a creator standing at the intersection of both art and science -- a person who produced a set of alignDRAW works that were initially conceived as a scientific idea making it into the art world
that being said the continuous…
4
5
37
other images from alignDRAW I want to highlight are "A very large commercial plane flying in blue skies" and "A very large commercial plane flying in rainy skies"
unlike the non-sensical prompts, my intention here was to test whether the model is responding sensibly to the small…
2
7
36
We have had GPT3 for 2 years but prompt marketplace and hacking only became mainstream with DALLE, Stable Diffusion and Midjourney text to image models.
Visuals definitely capture our imagination more
4
5
36
With so much work going to testing generalization capabilities of large language models LLM
Has anyone tested how much data can LLMs compress/memorize? Esp relationship between data size / model size / number of train epochs ?
4
0
35
With stable diffusion being so good what’s stopping us from applying it to text 2 speech ?
Similar continuous signal
5
1
34
Has anyone successfully trained transformer language model without any feedforward layers?
5
2
34
cool work that does interesting analysis of representations learned by transformers using various NLP objectives (left2right language model, masked language model and machine translation)
0
8
33
While RL might not be the best way of learning which tokens to place in the sentence,
@wellecks
showed that RL policy can successfully used to select positions to place new tokens in the sentence. Check out our updated preprint
0
5
32
your work being showcased at an art exhibition is a special feeling
you can feel a sense of pride and accomplishment despite being thousands of miles away
and you hope this is not the last time it happens
thanks
@verse_works
for hosting my and
@rainisto
works
4
5
32
These results show that DL architectures don't matter as much as the tools we use to make them work. And the most important ones are Adam, residual connections and layernorm.
Pay Attention to MLPs
pdf:
abs:
based solely on MLPs with gating, show that it can perform as well as Transformers in key language and vision applications
3
85
393
0
0
32
less than 3 hours left before the paper prompt auction closes down
excited to see the new owners of those pieces soon
🚨 Live Auction ⎯ Ending soon today (12/13) at 12:00 PM EST!
alignDRAW (2015)
The first-ever text-to-image artworks
by AI Scientist
@elmanmansimov
Daily website 🔗 Link in bio
1
4
41
3
4
31
Reach out to that person.
Don't be shy!
Many folks in ML I cold-emailed were all friendly and got back to me at some point.
Here is my advice for writing an email:
1
0
30
For the ML part, go over fundamental ML/DL/RL concepts. Bishop's ML book and Goodfellow’s DL book are my favorite guides.
1
0
30
cool paper! embeds a forgotten trick of maintaining exponential moving average of parameters during validation for better stability and results inside the optimization loop.
0
5
29
Peek on behind the scenes from
@niceaunties
and her very unique way of using AI to capture elements of Asian culture and her childhood.
This quote from the article stood out to me:
“Initially, my goal was to use AI to depict events and social phenomena that are difficult to…
GM
As lots of people asked me how
@niceaunties
is creating her iconic AI artworks, I found interesting to share a great article where she says a lot about it:
6
7
36
2
3
30
💯
it is like saying that we wish photos from iphone 1 looked as good as they look in the iphone 15 pro max
that being said, i tried my best with the skills and tools I had at that point in time to make images as "good looking" as possible. my original intention was to make…
On the aesthetics of alignDRAW. A common complaint I hear is, I wish it looked better.
Look, this was not made by human hand. This was created by a machine “learning” what words describe.
The fact that it is recognizable _at all_ is the achievement.
11
7
79
0
4
28
Omg someone just bid 5 eth on alignDRAW “a stop sign is flying in blue skies” work at
@ChristiesInc
I am so so happy to see it
Congrats
@elmanmansimov
on your first bid at
@ChristiesInc
!
The alignDRAW work is unique in that it marks the start of a new picture-making technology that is revolutionizing the world we live in: text-to-image.
Amazing to see your work with artists like
@yatreda
…
3
5
20
6
2
29
Played a bit with
@AI21Labs
language models to build a conversational semantic parser on fast-food domain with 50 prompts.
It worked well...
And surprisingly generalized to ordering things beyond food such as t-shirt and TV.
Wish I didn't exceed 175B model quota that fast 🙃
1
4
28
A quick and easy way to connect with deep learning person is to mention to them that you worked with restricted botzmann machines and used them for deep belief nets
4
1
28
Very curious to learn more about what drives ppl working on AI alignment?
What in particular a fully aligned AGI with human values look like? With so many human values and cultures on earth hard to image one for me.
If you working on it, pls consider replying!
2
1
27
a second alignDRAW prompt story a want to share is related to the "a herd of elephants flying in the blue skies"
i tried this prompt pretty much right after seeing good results with "a stop sign is flying in blue skies". thought that would be a good idea to try other…
4
6
26
Incredibly humbled to see my alignDRAW work auctioned at
@ChristiesInc
this month alongside the works of other incredibly talented artists
But the big reason why it happened is the work of
@halecar2
and
@fellowshiptrust
distributing and showing my work to collectors, museums and…
I am excited to see this second part of
@elmanmansimov
´s project moving forward. Last month, it was shown at
@ParisPhotoFair
, and it has already been acquired by a Museum in the US and by collectors seeing how these images mark the start of AI text-to-image art.
Thank you…
0
1
5
3
5
26
I am very excited about joining
@awscloud
@AmazonScience
in NYC, where I have started last month. I will be continuing to work on NLP and deep learning. 2/3
3
1
25
2 hours ago: 1000 more pieces on sale I think it will take time
Now: 🤯
1
0
6
6
0
25
24 hours to go for
@ChristiesInc
"a stop sign is flying in blue skies" bids to close
In the meantime, someone put a new 5.5 eth bid on the piece
Exciting!
3
3
25
Excited to share the work our awesome intern
@deng_cai
did with us at AWS AI
@AmazonScience
last summer
Read 👇 to learn more the work
3
5
23
Bonus: We have several NLP internship openings at AWS AI. Feel free to reach out to me.
1
0
22
Before sending that email, check if you have common connections with that person (ideally your advisor).
Ask a common connection abt that person and if you like what you hear, ask to introduce you.
2
0
22
For the coding part, brush your core algorithm/data structure knowledge. Lots of great articles online. Practice some easy/medium leetcode problems.
2
0
22
I heard the same argument few years ago
Why go to PhD if you can go to residency or research engineer role and do similar type of research in the industry
In fact a skill that you gain during PhD that you won’t gain in industry is ability to set right questions, directions…
Agreed. There's so many opportunities in AI now. It's a pretty suboptimal career choice to do a PhD at the moment.
Also, many outstanding AI researchers and hard carry engineers that I know of don't have an AI or CS PhD.
16
23
252
0
2
23
Yixuan (our intern at AWS) did a great job showing that a simple recipe of additional fine-tuning of LMs on different task-oriented dialog (TOD) tasks from various unrelated sources works pretty well when applying it to a new TOD domain.
2
2
23
@DavidSKrueger
@ethanCaballero
AI tech at the current stage is too premature to be worried about human extinction. At the end of the day current AI is just another layer of software controlled by humans.
Once the fully autonomous physical robots start living among us then i would think about getting worried.
6
0
23
Some representative references on each group with more exciting approaches missed and more to come soon!
1
1
23
Important Work ! Makes me realize that we are still early and have lots of room for better out-of-distribution models.
1
1
23
presenting my new art piece at
@lumestudios_
tonight!
can you guess which one from the screens?
GM New York 🍎
Join us today
@lumestudios_
for the Spectra exhibition - curated by
@CoCollectorsNFT
✨
29
18
97
1
2
22
when I first had a call with
@halecar2
he told me how he sees the parallels of AI art happening now with the invention of photography and it becoming an art form centuries ago
I never expected it but I am very pleased that my work is now being compared to first ever photo taken…
The Promise of New Images
In 1827 Nicéphore Niépce created an image with a machine that could fix an image to a piece of metal covered with a silver solution. That experiment yielded something that transformed our world forever: the Photograph. Almost 200 years later, no one…
10
18
75
2
7
22
The images generated by first ever generative text-to-image model (work done by me and collaborators at UofT circa 2015) will be exhibited and sold at largest photography fair
@ParisPhotoFair
next month!
Thanks
@fellowshiptrust
@halecar2
for immense help distributing it
2
5
22
and big thanks to
@GuillaumeLample
,
@alex_conneau
for their awesome codebase for multilingual masked language models which we build our code on top of!
1
1
21
Very happy to see almost 2 years of our work resulting in the paper in
@NatureComms
. Thankful to my collaborators on this project!
From
@omarmNYU
@elmanmansimov
@RichBonneauNYU
and
@kchonyc
in
@NatureComms
: "Masked graph modeling for molecule generation". Addresses known issues w prev attempts for molecule generation including marginalization, fixed ordering of nodes in a graph:
0
6
27
1
0
21
@Suhail
Chris Bishop’s Pattern Recognition and Machine Learning is my favorite and one I enjoyed reading most when started learning ML in undergrad.
He explains concepts very well, doesn’t overuse math, and focuses on fundamentals only.
0
1
19
"a stop sign is flying in blue skies" by Playground 2.5
1/ We are releasing Playground v2.5, our latest foundation model to create images.
We tested our model across 20K+ users in a rigorous benchmark that went beyond anything we've seen to date.
This model is open weights. More information in the tweets below. 👇
75
162
1K
1
2
19
Be open to those opportunities.
It may not be an exact thing you had in mind before you talking to them.
ML research internship projects usually have a concrete problem in mind but are very open-ended in the execution you take. This is your opportunity!
1
0
18
That person replied to you. Great!
Now schedule a video call to chat.
During that call, you will learn a rough idea of what they have in mind for potential internships.
1
0
17
Also, if you met that person before, remind them in that email.
[We met at the XYZ conference. It was fun chatting with you during the workshop]
1
1
17
My thesis "Neural Structured Prediction
using Iterative Refinement
with Applications to Text and Molecule Generation" is available here 3/3
1
0
18
If you are attending
#AAAI2022
this week make sure to check out our paper on state-of-the-art conversational semantic parsing with recursive insertions
2
4
18
Cool slide with important machine learning tricks! Would make it even more useful to turn it into presentation with short explanations and pointers to references.
Because there is nothing more useful than a mathematical trick, and thanks to your feedback, find the updated ML trick slide -:) Please RT, add even more to the list !!!
#MachineLearningTrick
4
60
240
0
7
17
Wait filmmaker
@paultrillo
whose interview for NYT podcast on his experience using Sora I listened few weeks ago is now part of
@FellowshipAi
artists 😮
Genuinely excited what he does with Sora and other cutting edge video generation models — so far it has been top notch
NEW ARTIST: PAUL TRILLO
We are extremely excited to announce the addition of the amazing
@paultrillo
to the
@fellowshiptrust
@FellowshipAi
Daily Program. Paul has been experimenting for years with storytelling through moving pictures, always with a keen eye to push new…
19
6
102
1
2
17
alignDRAW made it to the
@ARKInvest
newsletter
except that attached images are too good for its time 🤣
4
2
10
0
3
17
this
The importance of
@elmanmansimov
´s work has been lying dormant for 8 years. Sometimes, it takes new light or contextualizing to see how a small discovery can change how we imagine and see the world. Congratulations
@delronde
. This is one of the most important artifacts of the…
4
6
47
0
4
16
@ethanCaballero
@DavidSKrueger
Sure but it is the same thing as a humans deceiving/manipulating others to do things for them. Been happening since humanity started.
Can it be dangerous and harmful. Yes! It is human level extinction? No and very far away from it.
1
0
15
Paper accepted to
@NeurIPSConf
🎉
Big credit goes to
@deng_cai
who led the work during his internship at
@awscloud
and special acknowledgment to my collaborators!
You can check out the arXiv pre-print in the meantime
1
1
15
Being good with words is likely a more important skill than linear algebra for ML at the moment
Back to reading more books
1
1
15
Great thing about doing ML in AWS is seeing what customers care about beyond accuracy
And one of those things is backward compatibility.
A nice paper from my colleagues at
#ACL2021
establishes and analyzes this problem for NLP
0
1
13
I made a short presentation summarizing recent closed-book QA approaches by
@ada_rob
et al. and
@iwontbecreative
et al. at our NYU ML^2 group . Feels to me like a very untapped area of research.
0
4
15
Cool, you both enjoyed the conversation and now the recruiter contacted you to move forward with the interviews.
Time to prepare!
1
0
14