Spending the summer hanging
out with
@maria_antoniak
at
@SemanticScholar
,
@allen_ai
thinking about culture, scientific communities, whales, and coffee. If you are into any of these things, send me recs and let's go to a cute cafe together :)
Seattle summer, please stay gorgeous!
I recently found out that in the US, if a holiday falls on the weekend, they get the subsequent Monday off by default. Please someone bring this policy here, we all could use more long weekends. 🥲🥲🥲
Okay, finally making it a little more official: I am starting my PhD at
@LTIatCMU
. I will work with
@841io
and others at the intersection of society and language technologies. Excited and grateful.
I sometimes still can't comprehend how you can do video calls into another continent and we actually see each other in the same moment.
I imagine bits rushing through underground pipes in oceans or through space.
AI seems nothing in front of this genius of a technology.
How do you all fight the urge to over-monitor experiments?
When I run an experiment at night, I open it first thing in the morning. If the results arent as expected, it basically ruins my day.
Similarly, if an experiment is running, I find it hard to focus on parallel tasks.
Finally got around to curating my long list of bookmarks into this repo:
Contains (VERY) helpful articles for grad school applications and research.
Next goal: curating my twitter bookmarks!
It's that time of the year again, so re-surfacing this:
(It is not advice from me, it is just a dump of my bookmarks to helpful articles from when I was applying to grad school)
Excited to have joined
@GoogleIndia
this week as a Predoctoral Researcher in the NLU team. I will be working on multilinguality and fairness alongside
@partha_p_t
,
@vinodkpg
and other amazing folks. Couldn't have been happier :))
Folks graduating in 2022: If you have the slightest bit of interest in research, consider applying to the
#predoc
program at
#google
#research
India. You can send me your CV. DM to know more about the lab/role/culture.
About the lab:
The thing google should be worried about as people download edge to try chatgpt integrations, is that they will discover its pdf reader that has inbuilt annotation and tts capabilities!
It is that day.
Couldn't have been more grateful for the mentors and the people I got here.
Huge huge thanks to
@vinodkpg
@shachi_dave
@partha_p_t
Goodbye, GRI, you were a fun rollercoaster ❤️🙏🏻
FWIW: PhD recruitment tweets from Profs are really helpful. My timeline is flooded with them;
But: a little too late.
I wouldn't decide to apply to place 5 days before the deadline bcz a lot went in shortlisting places to apply early in the process, including asking for letters.
Too many people around me have heard of ChatGPT (great, they know what I work on), but they now ask what I think. And tbh, I don't know how to give a crisp answer that doesn't drown their excitement (and positive experience) but also does justice to my view.
As someone who asks questions in exactly this league, this tweet is a reminder to myself that dumb questions are not a reflection of my competence but simply a knowledge gap that can be easily fixed.
It sucks that there are people in the world who will screenshot private conversations to complain about their colleagues' mistakes (presumably to feel superior?). Don't do this.
Today I would like to coin the word "doom-working". It is when I am staring at my screen and slack and terminal, without making any progress, knowing I am too tired, but not wanting to stop because I don't think I have done enough for the day.
Things like this actually happening at
#BITSPilani
. A prof giving 0 marks for an assignment submitted 6 mins late by a student who was in a covid ward and didnt get good enough internet. And there are many more and worse stories.
Like, seriously? What are we even prioritising??
Tw death
My bestfriend's uncle died and his entire family was crying in the living room while he sat there in his room writing an end sem paper. The level of apathy in colleges is unreal
No matter the conference location, the problem is always US visas. Those who don't have it can't get there, and those who have it can't leave there for other locations. ** Sigh **
Those == mostly students.
For the past few days, I have been asked multiple times about the 'Predoctoral Researcher' role and its hiring process at
#Google
#Research
, India. So, I have compiled the answers and my experience in this blog post: . Hope this helps the upcoming batches.
Excited to have joined
@GoogleIndia
this week as a Predoctoral Researcher in the NLU team. I will be working on multilinguality and fairness alongside
@partha_p_t
,
@vinodkpg
and other amazing folks. Couldn't have been happier :))
Hi
#NLProc
, I am in Seattle to attend
#NAACL2022
. If you are around and are interested in multilinguality, fairness, evaluation, trustworthy NLP, D&I, coffee, dogs, or the ocean, let's talk :)
PS: accepting recommendations on things to do around here, I hear the coffee is good?
Our lab is hiring Pre-Doctoral Researchers. Apply soon!
I wrote a blog detailing the interview process: . With he caveat of the process being slightly diff each year, hope this helps! HMU if you have more doubts!
I don't like me twitter feed anymore. Where are all the research papers, all the rants, all the memes, all the fun subtweets? It just feels like an arid land of llm takes and influencers 😭
The pain of seeing an experiment you started at night and hoped to see results in the morning having failed at a stupid "cannot find file" error because you misspelled the path is beyond excruciating :(
Not a good start to today :|
#Naacl2024
was so fun! Met old friends and made new. So much good food and lively streets! And great conversations around evaluation, ethics, and culture, esp at the BOF session.
Highlight of the trip: FINALLY meeting
@vinodkpg
in person 💕
As India’s digital journey becomes powered by AI, it's crucial that we consciously invest in efforts to ensure that real-life societal biases don’t manifest in tech.
Introducing Project BINDI:
🔴Bias
🔵Interventions for
🟢NLP and
🟠Data in the
🟣Indian context
#GoogleForIndia
Re ACL anon debate: if the embargo hurts students most, students should be participating in the discussion (which I haven't seen much). Because sometimes indirect stakeholders (in this case profs) might think they know what the direct stakeholders think, but they don't.
Is it a colonized mindset that makes us Indians believe, despite 16+ years of Education in English and a fluent control of the language, that our English is not "good enough", that our English is not "native enough". What is this standard we are trying to model and why? Genuine Q
@WholesomeMeme
Alt text: "One thing I like about Pixar films is how the happy ending isnt always what you think it ll be. The toys dont go with Andy to college, Gusteau's restaurant gets closed down, Mike and Scully get kicked out of University, Carl never gets Ellie to Paradise Falls. >>>
📜 𝑊ℎ𝑎𝑡 𝑚𝑎𝑘𝑒𝑠 𝑎𝑛 𝐴𝐶𝐿 𝑝𝑎𝑝𝑒𝑟, 𝑎𝑛 𝐴𝐶𝐿 𝑝𝑎𝑝𝑒𝑟?
We want to study writing across research communities. If you submit to different venues, do interdisciplinary research, or collaborate across disciplines, please fill out our survey:
2007! That's the year this was said.
Yes, I paused the movie, soaked it in, laughed a bit, and contemplated where we are headed, then remembered that Monday's existential crisis is still 10 hours away.
(Netflix blacked the image, which was Optimus Prime saying the dialogue)
I feel kinship when I see someone write with brackets (which is not very often) coz it shows the urge to say so many things (especially unrelated and unnecessary ones) at the same time (and importantly, in the same sentence), despite it not being recommended (or easy to read).
I have started referring to myself as a researcher, even scientist sometimes, when I talk about the field or the community and in my head I still think it’s wild even though no one has ever visibly flinched!!!
Prospective PhD applicants at EMNLP - send those cold emails to your dream profs and meet them. I know it is daunting, but I promise it is fun in the end because most folks are genuinely kind and willing to talk.
I think its crazy that bank transfers in the US take multiple days, between which the money essentially cannot be used. The Indian system is so much better, instantaneous. Its ironical that the most “technologically advanced” and “rich” country has worse banking than others.
@WholesomeMeme
But they find out that what they wanted isnt necessarily what they needed, and I really like the fact that kids get to learn that life doesn't always turn out the way they dreamed and that's okay.
/End alt text
I went on instagram for 5 mins today and I realised why I was off it. Overwhelming to see people doing (or posting, at the least) so many things. I have words, no pictures, so twitter it is!
I want more people to say "it is okay to be sad or angry about a failure and you should process those emotions and move on to the next thing once you are ready" rather than saying "don't feel bad about failing because it happens to everyone".
I still have ACL 2023 to-read pile lying in an open tab and you are telling me that AACL papers are out and EMNLP soon will be? And that it is Oct already? And that 2024 will start soon? Not prepared. Not prepared at all!!
So there are a bunch of people writing nice goodbyes to that person who has left BITS, and it's bugging me. I guess they just weren't around (?) during the protest 5 years ago, BITS admin was never the same after the protest in my eyes.
Re-watching suits and falling for the charm of Gabriel Macht (Harvey) only to find out that his alma mater is CMU. To think he roamed these streets and halls with those starry eyes. I think this is the best piece of trivia I know about CMU 🐣💕
You will be so terribly missed at Google
@simi_97k
! Wish you so very best for your PhD
@LTIatCMU
. Soar great heights and achieve all your dreams. Lots and lots of love 💖💖💖
#NLProc
I am looking for papers that talk about learning from noisy data, if provided in sufficient quantity. Noisy data could include silver annotations, imperfect/ungrammatical language, machine translated data (also imperfect) etc.
Any leads appreciated, thanks!
Our paper- "A Case Study of Efficacy and Challenges in Practical Human-in-Loop Evaluation of NLP Systems Using Checklist" has been accepted at the Workshop of Human Evaluation for NLP Systems at
#EACL2021
Extremely excited about presenting at
@eaclmeeting
on Monday (1500 GMT+1)
Stratham's law: "When measure becomes a target, it ceases to be a good measure"
- Chris Potts on benchmarks in keynote
Hits right on point!
#acl2021nlp
#acl2021en
I appreciate the effort that people make when they use "he/she", but can we just graduate to "they" so that it is more inclusive! And if you want more reasons: it is shorter, easier to say/use in a sentence, and sounds much better.
Realising that people you look upto and admire also made mistakes, had learning curves, felt out of place, were stuck, got self-doubts and yet made it, is extremely reassuring and encouraging!!
My thoughts on the “woke” Gemini model debate is that it is a great example of an evaluation failure. We simply do not know how to holistically evaluate these systems. We also do not know how to incorporate socio-contextual nuance in evaluation. This is my time, baby XD
Timely reminder for applicants this cycle: Look at timezones of the deadlines properly. They are sometimes weird: some are UTC instead of PST/EST, there are daylight savings changes and so check latest timezone conversion, and they are sometimes 11:59 AM (12 noon) instead of PM.
I think only Raavan can keep up with the demands of research and reading at this pace.
For the uninitiated - Raavan is an Indian mythological character who has 10 heads and 20 hands.
Every conference deadline comes with "why do we have anonymity period" on twitter, who it hurts and who it benefits. Talking about it, because everyone is:
4 years ago, my professor told me there was no point submitting A* conference bcoz "our" papers are never accepted.
So it's that time of the year when (career) phases end and new ones start.
I have had an absolutely wonderful year working at Microsoft Research, India.
I will fall short of words to describe how much I have and grown personally and professionally.
Ever felt very busy doing nothing?
Want to work, but I am already working.
Want a break, but I am not doing anything.
Should I conclude this is lockdown craziness?
Apparently research review of 2022 is a thing, so here goes mine 4 days too late:
1. 2 papers at AACL.
1a. First one on creating Multilingual CheckLists, was a product of my internship at MSR, work I personally find very impactful and I am proud of.
1/4
My biggest takeaway from NAACL about GPT is that it considers what's said early and what's said last. The number of very different papers that have this finding is almost funny!
With prompting, how do you decide if the model is not doing well or if the prompt is not great? At what point do you stop trying prompt variants (or prompt-engineering as it is being called these days)?
One thing I really don't like about not publishing papers about research (and only releasing blog posts) is that it neglects to acknowledge the efforts and impact of *so many* researchers and their work. Yes, I mean citations.
To paper "influencers" out there: if you create summary threads - please tag the authors or the original tweet thread.
Authors want to be discovered, esp early career PhD students. The sharing helps, but giving credit with easy click through helps even more.
True. I was willing to take up an unpaid/less-paying RAship for the experience that I needed for applying to grad school, because I could afford to still survive. Not everyone has that freedom and privilege. Essential DEI barrier to overcome -- *Pay your interns and RAs well.*
I challenge everyone to identify the unique advantages shaping your educational trajectory and use them to highlight barriers to diversity, equity, and inclusion within academia. (3/3)
How do people get aesthetic pictures of their table with so many food dishes at once? Humare yaha to ek ek karke aati hai and agli dish aane tak pehle wali saaf ho jati hai.
The other day my entire code to query openai broke because they changed things around in their repo and I had wiped my older env. Got me wondering, what would happen to a startup whose entire product is based on such apis? How are they surviving?
"We have embraced the western culture so much. Our women don't like wearing sarees and salwar anymore. Not even at weddings." -- random fufaji sitting in a pant shirt and blazer at a wedding 🤦
Can we stop with the double standards already?
@matt_boot_
@pattzboot
In hindi: "ghar me diwali kyu hai" -- "why is it diwali in the home": diwali is the festival of lights, where a lot of diyas (oil lamps) are lighted in the home
If department organised information sessions for PhD programs, meant to help candidates from diverse backgrounds with their applications, happen at times when only people from the US timezones can attend, they are not really fulfilling their purpose.
Came across multiple papers that talks about how more multilingual pre-training data can lead to better "cultural alignment".
And so this is a good time for a self-plug finding - we found that this could also lead to more bias issues.
Dear
@twitter
,
I don't want to see new topics, I dont want to see what people I follow like, and I dont want to keep clicking "show me less of..". I just want to see the damned tweets of people I follow, and that is all. Is that so difficult?
- a-lost-in-the-noise-faithful-user
@QueerinAI
Minimum living wage of India is ~5k INR (~65 USD) --as per internet.
This is way more than 2-3 times of it.
This is 2-3 times the monthly pay of a (mostly white collar) job with a decent salary.