Happy 2019!!My interview with
@ayurellahm
@Forbes
is out.We talked about cognition & how today’s
#AI
systems lack commonsense.Understanding event dynamics & building causal models of world in the context of stories is a major driving force.
I've interviewed many graduating PhD students lately who are considering industry jobs, and almost all of them said they had never thought about AI startups before talking with me,so I thought it might help to publicly share “deep-tech AI startups vs big-tech AI research labs"👇
Incredibly honored to be on the
#ForbesUnder30
! My 15-year old self would have been thrilled to know that after all those failures and ups and downs, you will get some recognition for your work on
#AI
and
#NLPRoc
@forbes
GPT-3 with 175B params,10x any prior LM,is out.Setting aside controversies around implications of training such ever-larger models, the fact that they show high zero-shot performance on challenging reasoning tasks is pretty interesting. Ppl often ask me how I feel about all this:
I'm sick of opening my NLU talks with the same exact classic motivating example for 10 years now!! (and I am sure others are too). Someone at
@stanfordnlp
should just fix this one-off so that I can't use it anymore and we can all assume the state has changed!😄
If only every serious deadline reminder email we receive on a daily basis had some dose of humor and creativity in it, our busy stressful days would have passed easier! Thanks
@aclmeeting
!!
#ACL2020
Many don’t realize the magnitude of crazy US immigrants put up with
Many,like me,have gone 6-8+years not seeing parents due to single-entry visa
Many,like me,are afraid of traveling abroad even after they legally can bc of all the recent uncertainties for re-entries
Pls vote!
So, 7 years ago, at a meeting I had with
@satyanadella
discussing AI, he asked me “why is that your generation does not want to work for
@Microsoft
? Be candid”, I replied “well, Microsoft is not a cool innovative place!” I deeply admire what Satya has done in the past few years
#vscode
is an engineering marvel!!! 🙇♀️
And yes, not just for web dev!
It’s actually pure perfection for *Python* development on a remote host!
why did I waste so many years, AND dollars, on IntelliJ IDEA?!?
My hope is that more people learn about the option of deep-tech AI startups. The startup world needs diverse AI people, women and POC, to join. Startups are literally the foundations of future big tech, & by building it from the ground up, we might get to build a better big-tech!
Excited to publicly unveil GLUCOSE: GeneraLized & COntextualized Story Explanations,appearing at
#emnlp2020
main conf!
Despite amazing recent
#AII
advances,we yet don't have systems with basic human ability of building coherent mental causal models of what they read!Why is that?
So far,we have ZERO women in any stage of our interview pipeline, despite having tried pretty hard to reach out to women who could be potentially a good fit.This is nothing new to me,but this time around I'm 100% responsible & have no idea what else to try to get better results😟
What a great news to end the night with☺️So honored that our latest work on commonsense reasoning got the honorable mention for best paper award at
#emnlp2020
☺️
So glad that the
#NLProc
community is shining more light on commonsense, and more people work on tackling it nowadays.
As amazing & amusing as
#gpt3
is, this OpenAI original really reminded me of the point I used to harp on years ago: some of the credit we give our generation models is due to our amazing ability, as humans,to rationalize almost any random or irrelevant content & deem it creative!
This entire story is a deeper mess than it even appeared initially: they paid MTurk crowd workers $1 or $2 take a survey, requiring them to download an app connecting to Facebook, through which they harvested “friends” data and also demanded PII!
However, there exists this other class of startups,called “deep-tech” that “provides technology based on overcoming substantial scientific/engineering challenges...requires lengthy research, development, and often large capital investment before successful commercialization.”
Clearly, each has its own appeal. I personally truly enjoyed spending more than a year during my PhD at big tech labs. What gradually made me personally less enthusiastic about it was the whole “publishing for the sake of publishing” and “chasing citations” aspect of it all.
To contrast the above,at a deep-tech startup you do research for the sake of ultimately building something that truly works for potentially thousands if not millions of real users(beyond beating inherently narrow benchmarks),which has always felt more rewarding & exciting to me.
I,and so many other Iranians who basically can’t leave US,have missed out on presenting our papers too many times throughout years. The “present anywhere” policy is a much better idea than taking all conferences outside US,or holding them all inside! Consider signing.
#naacl2018
This is huge news! San Francisco is now the 1st major US city to ban the use of facial recognition by *police* and municipal agencies! This is how the hard work of ethical
#AI
researchers and activists can florish into real legislative action, with the vote 8 to 1 in favor...
Happiest
#Nowruz
and first day of
#Spring
! We tried to make these grim quarantine days brighter by making colorful Persian New Year Haft Seen, Sabzi Polow Mahi, and Persian Love Cake!! May the start of this new year wipe out all the craziness that 2020 started for us all.
This free-waving hair becoming
#Iran
's flag is quite a symbol for all women's liberty...
The entire world is moved by the bravery of Iranian women, as they are being killed for wanting the most basic liberties for bodily autonomy 💔
#IranProtests2022
My heart is broken for
#Afghanishtan
💔The clock is turning back 20 years with the entire world watching. If you don’t know about all the outrageous convoluted wars that Afghans have endured since 70s, I highly recommend reading
#TheKiteRunner
, one of my all time favorite novels.
Moreover, such startups don’t necessarily prioritize publishing their work for the purpose of academic publishing itself (something that big tech labs do), but put the result of their research into real-world products that could potentially get used by real users, if successful.
Aargh, why does this keep happening?!? If you are also scratching your head reading this Economist interview with GPT-2 and how come it’s not edited, read this thread on how they actually prompted the model
If you are reading that Economist interview with an AI [
@openAI
's GPT-2], you absolutely need to read the tweet below for context.
Economist's claim that interview was unedited is grossly misleading; coherence was in the eyes of human, not the output of the machine.
Things got even more and more meaningless observing all the leaderboarding craze and arXiv flag-planting games. Moreover, I could actively see how one individual barely matters at a huge company with all its existing hundreds of researchers and hundreds of ongoing projects.
Also, not being dispensable & mattering to the bottom line of an entire company is a huge distinction.When you are 1 of hundreds of thousands of employees, naturally, you are not as impactful as being 1 of 5, 20,or 100. You are more likely to leave a mark at a promising startup.
Now we are in 2020, and a ridiculously gigantic LM has picked up enough background knowledge about the world, that it gets ~83% on SCT right out of the box! Is this how having some level of commonsense looks like? I'd say YES! We've made a small step forward, but of course...
I’m super excited to announce
@verneekai
, our
#deeptech
#AI
#startup
Getting to do this (despite all the odds and roadblocks) has been by far the most rewarding thing I’ve ever done in my career!!
We are now hiring. Apply to join us in this excitingly ambitious journey ahead😊
Hello World 👋
We are in the middle of a once-in-century pandemic, what better time to build a once-in-century company!? 😉
Come and join us and start building!
#ai
#startups
#ArtificialIntelligence
So what does Verneek mean?
‘Verneek’ is a play on word ‘Wernicke’, area of brain responsible for language development & comprehension. Also Verneek = ver [the root word for truth in Latin] + neek [nerd + geek!],which hints at neeks seeking (data-informed) truth!😎
#verneeeeeeeek
This was when I actually started believing that these large pretrained models are potentially a good chunk of our way forward for encoding large amount of world and commonsense knowledge into our AI system. I talk about this more with
@samcharrington
here
As the definition states, deep-tech startups do very much the same kind of research that happens in big tech research labs, but applied laser-focused at one or few commercial applications.
We had hope,so endured 4years of Trump during which:
-I got my PhD
-Moved to NYC
-Established myself as AI scientist
-Got on America’s u30 Forbes list
-Cofounded an AI startup
-Got greencard!
Elated we didn’t give up on this inspiring
#NewAmerica
,despite all its deep wounds 2/2
I've been opening every single one of my AI talks for the past 12 years with this super simple yet unsolved coreference resolution problem. This year, finally, GPT-4 (and not ChatGPT!) could correctly answer this basic commonsense reasoning question, so to celebrate, this was my
I'm sick of opening my NLU talks with the same exact classic motivating example for 10 years now!! (and I am sure others are too). Someone at
@stanfordnlp
should just fix this one-off so that I can't use it anymore and we can all assume the state has changed!😄
So I wake up, open my email, and read the normally-rather-exciting-line of "we are delighted to inform you that ...", and I feel ZERO happiness! What has 2020 done to us? I guess lack of sadness is the new happiness!
#EMNLP2020
And then of course, big tech has name recognition, and it looks good on people's resumes. Although it sounds vain, as an immigrant myself, I know how much you'd rather people like your parents *knowing* the company you joined vs a seemingly random place that no one has heard of.
We basically haven’t slept for the past week with our hearts racing and the mounting stress... If this is how we feel, I can’t even imagine how the families of the 176 deceased passengers are feeling now with all this roller coaster of news... 💔 Speechless
Tomorrow at
#AAAI2020
#RCQA20
I'll be talking about some exciting work in reasoning that we've been up to during the past year:Building AI systems that learn to make explicit mental models of narratives,moving another tiny step towards giving
#AI
the commonsense of a 5-year-old!!
Moreover, at big tech you could have a lot of people and resources supporting you, e.g., a program manager to manage your projects, or a dedicated software engineer, etc. At a startup though, you have to actually love wearing different hats.
More than a year ago when
@omidb_
and I were walking in our favorite park,Madison Sq,
#NYC
was the global epicenter of the pandemic,& we were in utter despair!
Our city persevered, so did we!!!
#NYCisBack
& now we have opened our brand new
@verneekai
HQ right across the park!🎉
Our city was at its peak beauty today! Miss
#NYC
more than ever...
در نا امیدی بسی امید است
پایان شب سیه سپید است
There is so much hope to seek in our despair, since there is always a bright day after even the darkest night...
I’ve spent the past 4years in the deep-tech AI startup world,& I can’t recommend it highly enough.I honestly don’t know what my career would have looked like if I had chosen the straight path of joining big tech. Loving/thriving in one vs the other is personal since they differ:
I keep scrolling in
#emnlp2020
live tweets to verify whether or not we actually got the honorable mention, but have found no proof yet! Can someone please send me the proof? 😃
When you are stuck in the winter wonderland, and no food can get delivered, you have absolutely no choice but to spend the
#SuperBowl
Sunday making
#Fesenjoon
with Saffron rice,
#Tahchin
and
#Tahdig
, to make it all count!!!
Nothing warms my heart more than seeing an exodus of
#AI
research folks from big tech,going into various startups. I truly believe that startups are our only way of making meaningful disruption of various industries, through AI systems that actually work in the messy real world!
After 5+ wonderful years in Google Brain, working at the forefront of ML alongside inspiring colleagues, I'm excited to share my new adventure. We started Adept with the mission to build the future of human-computer collaboration. .
First and foremost, turns out many people think an AI startup is like that new dating app that uses some sort of existing “AI” for its recommendation algorithm. And honestly, that’s not too crazy given that a majority of startups called AI startups are indeed of that kind.
It feels SO AMAZING to have an actual on-site interview, and not ✌️on-site ✌️, being on Zoom!
In-person interaction has made the entire experience much more informative, meaningful, and quite fun!
I’d hate for our lives to go back to the grim times 😞 …
#GetVaccinated
I've been so busy developing
#AI
that can do sth of substance in real world that I seriously haven't had time to respond to hyped-up
#AI
story. My net:
#LaMDA
is obv not "sentient",but let's admit that few years ago none of us would have predicted seeing such a fluent AI dialog 🧵
President Biden signed an executive order today revoking the racist xenophobic
#MuslimBan
!
I recall vividly how shaken we were, as Iranian immigrants being a major affected population,when Trump signed it in his first days in office. Could we survive 4years of this or leave? 1/2
👍
#AI
reality check: We're getting great at recognizing patterns & finding correlations on narrow tasks, often failing at real-world scale. But, we're lacking in abstraction, reasoning, & commonsense. No AI system shows basic commonsense of a little kid & this needs *repeating*.
Before the pandemic, I’d get sick so frequently that I thought something was up with my immune system!!
I haven’t gotten sick for an entire year&3months,and now getting the 2nd dose chills reminds me of how much I don’t miss getting on packed subways and shared coworking spaces!
It was really great kicking off 2021 talks by virtually visiting
@allen_ai
, thanks for having me
@YejinChoinka
!
Turns out I was talking right when the historic Capitol Hill rampage started, so, it'll be forever memorable!!
Woohoo, this will be an interesting debate to watch! Looks rather similar in topic to
@GaryMarcus
and
@ylecun
debate on Nurture vs Nature, will see how different the content and the directions will be. NLU should be the front and center of this one, looking fw to the arguments!
This Friday I will moderate a debate at
#Stanford
between
@ylecun
and
@chrmanning
on "What innate priors should we build into the architecture of deep learning systems?". More tickets released later today; join the waitlist to be notified!
@stanfordnlp
...have a very long way to go! Back in 2016 I believed that "predicting what happens next" in a short story is a great test for commonsense,& I still do,but all AI benchmarks are inherently narrow,by design,& we have to keep moving the landmark even if we get to 100% on any task!
Many people have reached out to me asking for a list of deep tech AI startups.Turns out there is no such public list to track who is hiring. So let's crowd-source it!😊
Use this form to add a startup:
And here will be the live list:
I've interviewed many graduating PhD students lately who are considering industry jobs, and almost all of them said they had never thought about AI startups before talking with me,so I thought it might help to publicly share “deep-tech AI startups vs big-tech AI research labs"👇
Of course, a big tech company provides a certain level of stability and comfort, which is not guaranteed at a startup. I’ve even heard friends working at big-tech calling it “a place to retire with a cushy job”,and startups are rarely cushy.
Thrilled to welcome Kathy Waller, Fred Nazem, Manuela Veloso, Lisa Schirf, & Regina Barzilay to
@verneekai
advisory board.
@omidb_
and I are so honored to have such stellar business &
#AI
leaders with us!
2021 will be very exciting.Apply at to join us!
Interesting new work which builds universal language-agnostic sentence embeddings, on 90+ languages. This means embeddings that are useful not only across various NLP tasks (as are the likes of BERT, OpenAI GPT models) but also across various languages.
Massively Multilingual Sentence Embeddings for Zero-Shot Cross-Lingual Transfer and Beyond
90 languages, all handled by a single encoder.
New SOTA on XNLI.
From FAIR-Paris.
Today was officially the 1st day of summer for us, with the arrival of our 2021
@verneekai
summer interns!
A 90+ degrees F welcome to our research interns
@_anthonychen
,
@sidvash
, Yubin Ge, and our software engineering interns Shruti Sharma, Mohammad Hossain, and Yuying Su! 🥳
Hey amazing women in tech:We don't have enough women in the entirety of tech pipeline,from founders&core employees at startups to executives&employees in big tech.DEI issues are foundational,& joining early-stage startups could be a v promising step
Join us!Apply at
@verneekai
Are U working on NLU,Dialogue,or Semantic Parsing?Are U doing research in learning with limited data & transfer learning?
We are working on something very cool
@verneekai
,come join us! It'll be much more rewarding than chasing leaderboards & citations 😉
Interesting INLG paper comparing neural vs templatic language generation, concluding that “sometimes the costs of developing complex data-driven models (neural) are not justified and one is better off approaching the problem with simpler (rule-based) techniques”
#NLProc
Many papers claim *without evidence* that building neural NLG systems is faster than handcrafting a rule-based system. Puzikov and Gurevych built both for the E2E challenge, and found neural system took much *longer* to build
On the ominous day of May 17th in
#NYC
, we are planning to come back and work from our brand new office in Flatiron 🎉
Do you know of any companies in the area that have gone fully remote and are getting rid of their office furniture? We'd like to connect!
Please RT🙏
#recycleNYC
Couldn't agree more that if US wants to lead in
#AI
,reforming its immigration policy for attracting top talent is the key. Although my immigration case can be counted as one of the good ones(!),I'm still spending endless time on every single step of the way,seems never ending.1/2
I emphatically agree. My collaborators' visa restrictions have been one of the largest bottlenecks to our collective research productivity over the last few years.
We lost one of our dearest friends, Margie, to
#COVID
ー19 😢 She was truly one of the kindest people ever. She loved flowers & taught me a lot about them. Her favorite color was this very shade of purple that is now in bloom in our own backyard...feels like her purple spirit 💜
I totally agree with
@GaryMarcus
. Building a reliable NLU benchmark that is not prone to exploitation of data intricacies by models is very challenging; and more often than not we are not even lucky enough to uncover such cases (we were lucky in the case of Story Cloze Test v1).
Amazingly both
#acl2020nlp
best paper & runner-up are on better evaluating
#NLProc
models. As less-rewarding & exhausting as it can sometimes feel to work on metricing, as opposed to developing better (&bigger) models,rethinking how we measure progress is more important than ever
BEST OVERALL PAPER AT
#acl2020nlp
:
🎉🎉🎉
Beyond Accuracy: Behavioral Testing of NLP Models with CheckList
Marco Tulio Ribeiro, Tongshuang Wu, Carlos Guestrin and Sameer Singh
🎉🎉🎉
Excited that our big collaborative effort, "ParsiNLU: A Suite of Language Understanding Challenges for Persian" will appear in TACL'21!
If you're working on multilingual/cross-lingual NLP, give it a look!
Paper:
I can't recommend startup world enough for career& personal growth.I've been strong-willed in spending past 3.5 years in the startup ecosystem. It's hard work &initially risky,but it's incredibly rewarding!
Don't be afraid of taking calculated risks,it can be your best decision.
It was great visiting
@cambridgenlp
this morning to talk about “How far have we come in giving our NLU systems common sense?”.
Thanks for having me virtually travel across the ocean to momentarily forget about our nail-biting elections here!!!
What a wonderful recognition
@mmitchell_ai
and the team, for your years of effort and deep care for pushing
#AI
towards doing good!
Always proud of you & inspired by you Meg!
.
@GoogleAI
's Model Cards recognized as one of three winners in
@Wired
's "The Tech That's Championing the Public Good".
Congrats to
@mmitchell_ai
and the whole Model Cards team!
Last 4years have been crazy for millions,including Iranian immigrants.
All the way from everyday discriminations to Travel Ban to being at the brink of a war with Iran
Many ppl would ask us “why stay?”
It was all because of HOPE & tremendous activism against this bigotry!
🙏💜
The past four years have been characterized by division, injustice, and relentless assaults on all of our rights.
We’ve also seen millions of people come together to defend our rights, preserve our democracy, and fight for our communities.
Together we are powerful.
Of course, I’m painting the distinctions with a board brush, and there are exceptions to the above characterizations. Clearly, at the end of the day, it all boils down to personal leanings and choices.
Years ago, after watching Imitation Game, we decided we should draw Turing's face big on a wall if we ever designed an office. That seeded
@verneekai
Wall of Fame at our office,a satirical bragging contest at the expense of the SOTA in
#AI
"Sorry, I'm not sure I can understand"!
HOW DO THINGS KEEP GETTING MORE INSANE??
I graduated a few weeks after
#MuslimBan
was signed,& decided to stay in US despite all reasons not to.I can't imagine how international students will ever rationalize staying here if
#StudentBan
becomes effective,on top of OPT&H1B orders
Well, it finally happened. This
#ICE
International
#StudentBan
got me back into the “organizing effective lobbying action via twitter” game.
Want to make sure our lovely international students can keep their visas? Here’s how to do it:
THREAD.
1/14
It was so great catching up with
@samcharrington
on
@twimlai
. I shared,for the 1st time, our broad vision for
@verneekai
and the lines of
#AI
research we are pursuing. We talked about starting in the pandemic, my views on how far we’ve come in NLU,& beyond
I also pointed out how the business of likes of
#StabilityAI
is unsustainable & can't build the future big tech, when you need to raise $100M to pay (estimated) $80M of it towards cloud costs only -- which has everything to do with the bottom line reason for the exposé that came
A great point from the always acute
@willoremus
!
@nasrinmmm
made same point back in Feb during our AI Futures roundtable (a version of which will be published Critical AI's February 2024 issue on LLMs).
Her perspective was how start-ups need Big Tech patrons.
All too common for many female
#AI
researchers!!Thanks
@elasri_layla
: “when I’m with male colleagues,researchers shake hands&speak,and I remain this invisible unacknowledged person” “I’ve been told that I was invited to speak at a panel because the organizers needed diversity”...