PhD @LTIatCMU | Prev. @GoogleResearch @MSFTResearch
Joined April 2018
- Tweets602
- Following1.4K
- Followers3.9K
- Likes1.8K
Pinned Tweet
Thank you so much @emnlpmeeting for this wonderful recognition! I’m so honored and humbled 💕 Thanks @gneubig for your support throughout!
We’ve been working on this for 1.5 years and everyone who has spoken with me in the recent past knows how passionately I feel about this work and this task in general 🥹 Thanks again to everyone who recognized and appreciated our efforts ✨
#EMNLP2024 Best Paper 1/5: An image speaks a thousand words, but can everyone listen? On image transcreation for cultural relevance
Simran Khanuja✈️ COLM'26 retweeted
I'll be at #COLM2026 this week! I'm excited to talk about social grounding in AI, especially:
🧠 pragmatics & interactional intelligence: understanding user intentions behind what is said and left unsaid
🗺️ cross-cultural competence: adapting to the cultural norms and values that shape how we communicate
I'll be graduating next year and am starting to explore research roles in these areas, and would love to connect with folks working on related problems, as well as catch up with old and new friends! Also always on the hunt for good bread 🍞🥐
I'll also be presenting 3 papers in the main conference, details in 🧵:
Simran Khanuja✈️ COLM'26 retweeted
I'm at #COLM2026 and on the academic job market this year! I'd love to chat about long context reasoning, distilling harness capabilities into architectural improvements, and better training methods for complex corpus tasks.
I'll be presenting work on Tuesday + Thursday ⬇️
Excited to be presenting my GDM internship work on cultural alignment/interp at COLM next week!
Room: Franciscan B; Poster Location: #81 at 11:00 AM on Tuesday, October 6th!
Would love to meet old friends and new :') Happy to chat about anything, been working on AI alignment, evals, post-training, image generation; but broadly also been thinking about domain applications esp. healthcare & education!
I'm also starting my industry job search and would be happy to connect with people for relevant opportunities if I'd be a good fit :)
Website -- simran-khanuja.github.io/
Which cultures do LLMs best align with on open-ended questions? Why? Can we change that default?
We study this in our new work: "Steering LLMs for Culturally Localized Generation" (COLM 2026), from my time at Google DeepMind! 🧵
arxiv.org/abs/2603.23301
Simran Khanuja✈️ COLM'26 retweeted
Did you know that gifting a plate with four koi 🐟 is offensive in Japan (4 = "shi" = death 💀), or that sticking chopsticks 🥢upright in rice is inappropriate in China (resembles funeral incense 🕯️)? Top AI models certainly don't!
Our #COLM2026 paper introduces 🌏NormViz, a benchmark and framework for measuring visual norm understanding across 16 countries.
Best model Gemini 3 Flash gets only 25.3% 🚨
🧵below (1/n)
Simran Khanuja✈️ COLM'26 retweeted
This is a very good model, proud to have been a part of this effort!
It's been a big step up for my productivity, I find myself delegating it more ambitious, long-running tasks, and being pleasantly surprised at how well it performs.
Couldn't imagine us having to do something like this two years ago, it's such a low bar and yet :'( but its so cool that they took the time to do this!
TMLR has faced a deluge of submissions, necessitating stricter desk rejection policies due to limited reviewer capacity
Co-EiC Nihar Shah reached out to authors of 10 papers slated for desk reject. Could they answer questions about their *own* submission?
medium.com/@TmlrOrg/asking-a…
Simran Khanuja✈️ COLM'26 retweeted
Join us at @eccvconf for the Visual Persuasion (VisPer) Workshop!
📅 Tuesday, September 8, 8:45am-12:30pm
📍 Room Malmömässan C2
🖼️ Poster session: 12:30pm-13:45pm, Malmömässan Exhibit Hall
🌐 sites.google.com/view/eccv-2…
We have an amazing lineup of speakers. Can’t wait to see you all there!
#ECCV2026 #VisPer
Check out Graham and Daniel's cool new course on AI agents with possibly the largest waitlist I've ever witnessed for a CMU course in my time here :')
Simran Khanuja✈️ COLM'26 retweeted
Excited to give a keynote to EMNLP this year! I'm planning on talking about the value of open academic research in language modeling, so hopefully it will be interesting
🎉 We are incredibly excited to announce the Keynote Speakers for EMNLP 2026: @pascalefung , @gneubig, and @annalkorhonen !
See you in October in Budapest!
🔗 2026.emnlp.org/program/keyno…
#EMNLP2026
ALT The three EMNLP 2026 keynote speakers: Pascale Fung, Graham Neubig, and Anna Korhonen.
I did a thing :') Had a great conversation with Jay Shah about a bunch of hard, open problems in AI today!
We talked some about my PhD research, how I was inspired to work on a new problem, machine translation for vision, by a real-world need -- children reading translated storybooks needed images contextualized too, not just the words! We also broadly spoke about:
— evaluation, especially for creative, subjective tasks
— alignment, and how it should be encoded within models
— personalizing models for users with very different experiences and backgrounds
— how to choose what problem to work on in AI based on your personal goals!
Full timestamps in the description; feel free to skip to whatever's of interest, and feel free to reach out if any of this resonates and you'd like to chat!
I had a great conversation with @simi_97k (CMU, formerly Google Research & Microsoft Research) on the latest podcast.
We spent 2 hours talking about multilingual AI, cultural alignment, multimodal models, LLM evaluation, and where AI research is heading beyond just scaling models.
A lot of ideas in this conversation changed how I think about "general-purpose" AI. 🧵
Love the section on epistemics! It's something I don't see very often in model release blogposts :')
Also, great job on the naming of Inkling and Tinker! It cleverly aligns with their overall mission
Today, we are introducing Inkling.
Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.
thinkingmachines.ai/news/int…
Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵
Simran Khanuja✈️ COLM'26 retweeted
Very excited to see this new release by thinking machines, I love more players in the arena, and also love the framing around this being a good model for customization!
Also, continuous thinking effort is very cool.
Today, we are introducing Inkling.
Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.
thinkingmachines.ai/news/int…
Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵
Which cultures do LLMs best align with on open-ended questions? Why? Can we change that default?
We study this in our new work: "Steering LLMs for Culturally Localized Generation" (COLM 2026), from my time at Google DeepMind! 🧵
arxiv.org/abs/2603.23301
This work confirmed my suspicions that there is a lot of knowledge about different cultures locked into model params which MLE suppresses. It was also really fun to analyse selected features and see how steering changes responses for the same prompt.
Mech interp is hard tho! Manipulating model behaviour during inference in a way that doesn't break it is something i found most challenging.
If y'all give the paper a read and have feedback/comments/questions, please reach out to me! Here or [email protected] :)