Divya Kulkarni (@divyayyy) header
Divya Kulkarni (@divyayyy) profile picture

Research Fellow @Harvardmed || Ph.D. @IITGuwahati || ಕನ್ನಡತಿ || tweets = Machine Learning/My Research/Cricket (occasional *blahh* posting!!) (views - personal)

Boston, MA
Joined December 2010
Photo by Divya Kulkarni (@divyayyy)Photo by Divya Kulkarni (@divyayyy)Photo by Divya Kulkarni (@divyayyy)Photo by Divya Kulkarni (@divyayyy)Photo by Divya Kulkarni (@divyayyy)Photo by Divya Kulkarni (@divyayyy)Photo by Divya Kulkarni (@divyayyy)Photo by Divya Kulkarni (@divyayyy)Photo by Divya Kulkarni (@divyayyy)Photo by Divya Kulkarni (@divyayyy)
Divya Kulkarni retweeted
It cannot be overstated how much your quality of life decreases when it gets dark at 5pm instead of 7pm
788
2,661
344
39,437
2,058,594
Divya Kulkarni retweeted
‼️ GIVEAWAY ‼️ We're giving away 2 tickets to the Patriots vs. Green Bay Packers game on Sunday, 11/8, plus 2 General Admission tickets to @TheHall, open from 10 AM through kickoff! 🏈 To enter: 1. Follow @PatriotPlace 2. Like and Repost this video! Good luck! Giveaway closes on Friday 11/6.
275
3,845
48
4,206
136,630
Divya Kulkarni retweeted
One SpeechLLM to do all kinds of transcription. Fully open weight! Check out details in the tech report: research.nvidia.com/labs/nem… Congrats to @PiotrZelasko and team on a great release!
Meet NVIDIA Nemotron 3.5 Transcribe 🎙️ An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint. Watch it keep track of who said what when everyone’s talking at once. 1/ 🧵
3
8
74
3,173
Divya Kulkarni retweeted
The world is unfair. Working hard doesn’t guarantee success or fame; continuous hard work and learning from previous mistakes just increases the odds in this unfair game.
9
3
3
199
9,451
Urm... *my failed models might not have actually failed? Falling into an existential crisis thinking about those* Super cool work! Congratulations and THANKS A LOT!
Hi all, I have a new result I'm excited to share. It turns out that *all existing implementations* of softmax and cross-entropy (that's JAX, PyTorch, textbooks, etc) have a subtle bug --- two bugs, actually. A simple rewrite fixes both, and eliminates "slingshots" when grokking:
3
329
Divya Kulkarni retweeted
Folks i am looking to enhance my research group. The PhD student positions is below. Pls RT !! 🙏🙏 PhD Position: Urban Robots with World Models and RL for Indian Streets I am looking for one/two PhD student(s) to build algorithms and systems for robotics for Indian urban spaces -- mainly around urban mobility. The core approach is learned urban world models combined with (probably) reinforcement learning, moving from data and simulation to real robots. Current world models and navigation policies are trained mostly on structured Western streets and fail in these conditions. There are interesting research questions that will also lead to products/startups for Indian (and who knows even the west) streets. E.g: Which world-model representations best capture dense, unstructured, multi-agent street behaviour? Can policies trained largely inside a learned world model transfer to real robots with little real data? The work starts with data, models and simulation, then moves to campus trials and field deployment. Target venues include ICLML, ICRA, NeurIPS, ICLR, ITSC and of course transactions of various kinds. Who should apply ? A B.Tech/M.Tech/M.Sc or equivalent in CS, EE, ME, mathematics or a related field, meeting IIIT Delhi PhD eligibility. "Deep" interest in ML :) RL, computer vision, or robotics experience is a plus. The fellowship follows IIIT Delhi norms, and eligible candidates are later encouraged to apply for PMRF. This is a rolling position. Email [email protected] with the subject "PhD Urban India Models". (write your email in your own words). I have collaborations in US, Australia and in HK. The candidates will have the opportunity (optional) to visit them as well.
4
18
83
5,141
Divya Kulkarni retweeted
ROMEO DOUBS FIRST TD AS A PATRIOT!!!
5
49
787
10,882
Divya Kulkarni retweeted
We (@ChandarLab) are continuing our tradition of helping underrepresented students apply for grad school by reviewing their application materials and giving feedback! Please retweet this for maximum reach! Deadline Nov 1st!
🎓 Graduate Program Application Mentorship Applying to grad school doesn't have to be overwhelming! Get personalized 1:1 support from students who’ve successfully navigated the process. 🔗 Apply now: forms.gle/FyjEuLmjcdREF3Ba6 🗓️ Deadline: Nov 1, 2026 👇 Here is what you get:
14
75
7,652
Divya Kulkarni retweeted
A nice application built by Keshava using my open source TTS - Vagdhenu. This is the purpose of open source: you create something and community makes it better/more useful.
Built a small web app to help kids/beginners to learn shlokas . Tap each pāda, hear it chanted, with a soft tambura underneath. A Beginners set plus an Advanced section (Yantroddhāraka Hanuman Stotra and many more on the way). Recitations powered by Vāgbodhinī, @prathoshap's brilliant tool. Link is here - stotrapathana.netlify.app/ @JoraShankara @Agnidhra_ @_CreatingWealth
12
53
619
29,642
Divya Kulkarni retweeted
Built a small web app to help kids/beginners to learn shlokas . Tap each pāda, hear it chanted, with a soft tambura underneath. A Beginners set plus an Advanced section (Yantroddhāraka Hanuman Stotra and many more on the way). Recitations powered by Vāgbodhinī, @prathoshap's brilliant tool. Link is here - stotrapathana.netlify.app/ @JoraShankara @Agnidhra_ @_CreatingWealth
10
36
1
188
32,744
Divya Kulkarni retweeted
Prayut Jain (@PrayutJain ) is a friend of mine, he's had the kind of year I wouldn't wish on anyone, and he's now Looking For Work. He found out he was being let go while sitting beside his father's ICU bed in India — weeks before his first major equity vest. His Story 👇
3
19
2
62
20,615
Divya Kulkarni retweeted
🔍 Reviewer applications are now open for WiML @ #NeurIPS2026! Support women and non-binary researchers, engage with emerging ML research, and help shape the WiML 2026 program. 📅 Deadline: Aug 16, 2026 🔗 buff.ly/mUy363s #WiML #WomenInML
8
23
2,087
Divya Kulkarni retweeted
Any must-reads from #ICML2026? Self-promotions welcome too 🙂!"
5
1
56
13,097
Divya Kulkarni retweeted
The state of boredom is when our neural network quietly fine-tunes itself based on our past experiences. Constant "doing" isn't living - it's the quickest path to becoming a zombie. Stop running from silence. Gift yourself some downtime - literally doing nothing!
Novak Djokovic just said being bored is the most creative state a child can be in. His son is 10 and his daughter is 7. He says when his son told him he was bored after a morning of ping pong, kayaking, and soccer, he sat him down for a conversation most parents avoid. "It's okay to be bored sometimes. When you're bored, it doesn't mean that you have to instantly take a book or a screen. You need to also learn how to be with your thoughts." Djokovic says boredom is when creativity finally shows up, and it's also when everything you have been suppressing through your phone comes to the surface. Most parents are protecting their kids from the only state that grows them. — Novak Djokavic (@DjokerNole) on Jay Shetty's (@jayshetty) podcast
1
2
29
3,280
CLASS!
Juan Manuel Cerundolo was all class after his win over Jannik Sinner at Roland Garros: "What am I going to say, right? I got lucky today. Obviously I gave it my all and tried to play my best tennis, but in not going to puff myself up and say just anything because the truth is I wasn't able to get more than 3 games off him per set." "I told him I felt sorry for him, for his injury. I hope he recovers. That without a doubt I'd gotten lucky. And he's always really gracious, a true gentleman, a great person. The truth is I always wish him the best because he's a phenomenon, as I say. And he came over and greeted me, congratulated me and wished me luck." ❤
2
505
(I have stopped tweeting random takes but for this I couldn't stop) The world, be it anyone, should stop lecturing women but focus on systems that will enable women to have kids. The systems in many countries are almost hostile for women to have kids, right from discrimination +
I studied the life and work of Lee Kuan Yew, the Founder-Prime Minister of Singapore intensely during my early 20s. Immense respect for him. In this video clip, he advises a young woman pursuing a PhD not to neglect marriage and kids. I recently gave very similar advice. Elon Musk would say much the same thing. We have to orient our culture, society and the economy so that smart ambitious women do want to have children (and sadly, this needs to be said today: men cannot have children). If we fail at this, humanity will lose. Now let the attacks begin! 🤓
1
1
10
560
+ to denial of opportunities. So kindly spare us women from these advices.
5
284
Divya Kulkarni retweeted
🔈Meet MICCAI Fellow, Marius George Linguraru, at our next #MICCAI Fellows Talk - "From Scarce Data to Real-World Impact: Responsible #AI for Children’s Health & Rare Diseases" 🗓️ Thurs Apr 30 ⏰ 9am EDT/ 3pm CEST 🔗us02web.zoom.us/webinar/regi… @MiccaiStudents @WomenInMICCAI @RMiccai
1
1
4
1,057
Divya Kulkarni retweeted
Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code. But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along. So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions. TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.
The degree to which you are awed by AI is perfectly correlated with how much you use AI to code.
1,262
2,703
717
22,757
7,649,506