Jebish7

@universeinanegg Saving this just in case…

@awsdevelopers AWS

Reply to this tweet with "AWS" and we’ll tell you which AWS Service you are

@universeinanegg Thanks for bringing this to my feed. Went through few threads. My biggest takeaway is that, we are really lacking in our evaluations of models. Really fascinating idea this, kinda also shows how agents operate in social settings.

anyone who doesn't think this is *interesting* is either missing something or has lost their sense of wonder, IMO (it can still be bad, empty, whatever, but it's interesting)

@sarahookr With No Bias whatsoever, can say Momo is the best.

My favorite type of distribution shift.

@skoularidou Congratulations. As an aspiring researcher, I don’t think 3K is trivial. It shows that researchers need your paper.

I understand that this is quite trivial to most of the researchers in here but I feel like sharing that I hit >3,000 citations It is not much, but it somehow feels nice, as despite the health issues life goes on

@tomssilver Relocate to South Asia? You will have deadline on 5-6 PM, so a typical office time.

Considering relocating my lab to Anywhere on Earth so we can go to bed at a reasonable hour after deadlines

@universeinanegg Tried this with Claude and GPT, and it’s the same. Gemini 3 pro did give two 3s ( after thinking a lot). Though all of their first two numbers are 3 followed by 1 (irrespective of temperature). Models seem to love 3.

my favorite part about Google's AI is it will NEVER allow duplicates if you ask for random numbers with replacement

@Kuvvius Congratulations 🎊

#ICLR2026: A cycle that will be remembered for years to come. 🌊 🤧A wild ride for the community. 🩷 So grateful to have my incredible co-authors by my side through it all. Let's continue exploring multimodal reasoning: 🧵👇 1. ThinkMorph thinkmorph.github.io 2. STARE arxiv.org/abs/2506.04633 3. AdaReasoner adareasoner.github.io @iclr_conf #iclr

Kaleidoscope has been accepted at ICLR 🔥. This is the first of its kind massive multimodal multilingual benchmark. Configurations to everyone @Cohere_Labs 🎊

@Cohere_Labs kick-started my research journey. Two years later, we continue moving forward together.

🔮 Can a world model (simulator) give today’s AI agents foresight? We tested “world model as a tool”… and found it often doesn’t help—sometimes it hurts. Check our newest paper here: arxiv.org/pdf/2601.03905… #AIagents #WorldModel #ToolUse

@ReviewAcl pretty please extend the Jan deadline

@PingbangHu There was sudden score inflation after the leak, so this was the only realistic way to fix it. They could have rolled everything back to just before the leak, but that would’ve been unfair to people whose reviewers hadn’t responded yet.

Hot take on ICLR's action to revert reviews and scores: It's acceptable, even reasonable, if you believe in academic integrity. I’ve seen people describe this decision as “treating everyone as guilty.” Unfortunately, that’s true in a sense, but this is about academic integrity, which is inherently much stricter than ordinary moral standards. From that perspective, I find the decision completely reasonable, even though I personally benefited from the rebuttal process.

Very excited to see that Global PIQA is already being used to evaluate multilingual capabilities in new models!

@Mengyue_Yang_ They previously released a notice for increased security for cheating and using third party apps. Maybe they were anticipating this.

As a world-model researcher and long-time Genshin player, this is absolutely mind-blowing to see 🤯🔥 Huge respect to the team, this is the kind of crossover I never expected (yeah but not out of my imagination). But seriously… when did Genshin’s environment become open enough for training agents? 😂 We really need that!

Introducing Global PIQA, a new multilingual benchmark for 100+ languages. This benchmark is the outcome of this year’s MRL shared task, in collaboration with 300+ researchers from 65 countries. This dataset evaluates physical commonsense reasoning in culturally relevant contexts.

3 days. Worldwide. Inspiring & starting new research collaborations. Introducing the Connect conference. 🖇️ Join for incredible speakers, including @1vnzh @jpineau1 @mziizm & @ShayneRedford + >20 researchers discussing how collaboration and open science are driving progress. 🚀

Three of our papers have been accepted at AACL 2025 @aaclmeeting (2 Main, 1 Findings). 1. DSBC : Data Science task Benchmarking with Context engineering arxiv.org/pdf/2507.23336 2. Uncovering Cultural Representation Disparities in Vision-Language Models arxiv.org/pdf/2505.14729 3. Improving Multilingual Capabilities with Cultural and Local Knowledge in Large Language Models While Enhancing Native Performance arxiv.org/pdf/2504.09753 Grateful to the co-authors @SidYaeger @Siddartha_10 @jebish7 @delliott @alexrs95 @_sumand @_srishtiyadav @KanwalMehreen2 This was made possible through research grants from @TraversaalAI @AnthropicAI @Cohere_Labs

1. DSBC: Data Science task Benchmarking with Context engineering arxiv.org/pdf/2507.23336 2. Uncovering Cultural Representation Disparities in VLMs arxiv.org/pdf/2505.14729 3. Improving Multilingual Capabilities with Cultural and Local Knowledge in LLMs arxiv.org/pdf/2504.09753

Three of our papers have been accepted at AACL 2025 @aaclmeeting (2 Main, 1 Findings). Grateful to the co-authors: @1024m_nlp @SidYaeger @Siddartha_10 @delliott @alexrs95 @_sumand @_srishtiyadav @KanwalMehreen2 & for the support from @TraversaalAI @AnthropicAI @Cohere_Labs