Ramakanth Kavuluru

We have some new results showing the effectiveness of peer fact checking, operationalized on WhatsApp. Unlike most studies which use artificial headlines, our intervention was done on real user-donated WhatsApp data. Paper and details in the thread below:

@SenMcConnell Dear senator, I read the full letter of Mr. Kent that l link below for others to see too. I am unable to interpret its contents as "virulent anti-semitism". He did not attack jewish people for their faith. I would appreciate any clarifications. x.com/joekent16jan19…

Joe Kent testified before the Senate one year ago that Iran and its terror proxies threatened U.S. servicemembers in the Middle East. He said it would be an honor to return to the fight against terrorism, and he pledged to lead with integrity and accountability. The virulent anti-Semitism of his resignation letter makes it clear that Mr. Kent is incapable of upholding these pledges, and those who mistake its baseless and incendiary conspiracies for brave truth-telling are only fooling themselves. Isolationists and anti-Semites have no place in either party, and certainly do not deserve places of trust in our government.

xLSTM Distillation: arxiv.org/abs/2603.15590 Near-lossless distillation of quadratic Transformer LLMs into linear xLSTM architectures enables cost- and energy-efficient alternatives without sacrificing performance. xLSTM variants of instruction-tuned Llama, Qwen, & Olmo models.

It's like every few months there are mass layoffs in the industry. I hear that our CS department is getting top notch applicants this hiring cycle, thanks to this fiasco. Happy we are going to benefit from this 🙂 cnbc.com/2026/03/14/met…

Happy to receive this research mentorship Immy 🙂 award in our department today.

Our study of AMIE at @BIDMC_Medicine is out! You can read about what we did in posts from my co-authors (including the Google post below). But I wanted to talk about some background for this study, and what I think are most interesting findings. 🧵⬇️ x.com/GoogleResearch…

We've adapted the Avey architecture to the encoder paradigm and called the result Avey-B, a next-generation alternative to BERT with unlimited context length. Accepted to @iclr_conf: arxiv.org/abs/2602.15814 Code and checkpoints: github.avey.ai/avey-b hf.co/collections/av…

Excited to share that our paper has been accepted to EACL 2026 Main! ❤️

Announcing NVIDIA Nemotron 3 Super! 💚120B-12A Hybrid SSM Latent MoE, designed for Blackwell 💚36 on AAIndex v4 💚up to 2.2X faster than GPT-OSS-120B in FP4 💚Open data, open recipe, open weights Models, Tech report, etc. here: research.nvidia.com/labs/nemotron/… And yes, Ultra is coming!

Good post on how to think about honing your skills as an (academic) researcher by Carlini nicholas.carlini.com/writing/2026/h…

🚨 New paper 🚨 Excited to share PRISM, a new “DeepThink” method that uses step-level correctness signals from a process reward model to guide inference over candidate solutions. PRISM matches or beats SOTA methods, enabling gpt-oss-20b to exceed gpt-oss-120b.👇 📰: arxiv.org/abs/2603.02479 #AI #LLMs

Vision-language models improve multimodal systems, but can make them slower, costlier, and harder to deploy. Learn how Phi-4-reasoning-vision-15B, a compact and fast multimodal reasoning model, blends strengths of different methods while reducing their limits: msft.it/6014Q5X0u

We just released pre-mixed, pre-shuffled pretraining datasets at 100BT scale. @asankhaya tested 50+ different mixture strategies at 1B scale. The winner? A static 50% finePDFs + 30% DCLM + 20% FineWeb-Edu blend. No fancy curriculum needed. We scaled this up to 100BT and pre-shuffled everything so you don't have to burn compute on sampling. Just use it: from datasets import load_dataset ds = load_dataset("HuggingFaceFW/finepdfs_50BT-dclm_30BT-fineweb_edu_20BT-shuffled") Browse the full smol-data collection: huggingface.co/collections/Hu… Reproduce it yourself: github.com/huggingface/da… Read the methodology: huggingface.co/blog/codelion/…

🚀 Introducing the Qwen 3.5 Small Model Series Qwen3.5-0.8B · Qwen3.5-2B · Qwen3.5-4B · Qwen3.5-9B ✨ More intelligence, less compute. These small models are built on the same Qwen3.5 foundation — native multimodal, improved architecture, scaled RL: • 0.8B / 2B → tiny, fast, great for edge device • 4B → a surprisingly strong multimodal base for lightweight agents • 9B → compact, but already closing the gap with much larger models And yes — we’re also releasing the Base models as well. We hope this better supports research, experimentation, and real-world industrial innovation. Hugging Face: huggingface.co/collections/Qw… ModelScope: modelscope.cn/collections/Qw…

To support open and collaborative science, we offer Tinker grants for researchers advancing the field. This week we’re featuring publications by some of our early research grant recipients! thinkingmachines.ai/blog/tinker-re…

Reasoning LLMs generate very long chains-of-thought, so even small quantization errors add up. With AWQ, Qwen3-4B drops 71.0 → 68.2 on MMLU-Pro (~4% relative loss). 😬 ParoQuant fixes this! It keeps only the critical rotation pairs and fuses everything into a single kernel. Recovers most of the lost reasoning accuracy with minimal overhead — so 4-bit models stay strong at reasoning. 💪💪

Twitter is doing LinkedIn nonsense these days. Someone joins a big tech/frontier AI lab and fifty people tweet a welcome message to them 🤦‍♂️

Data mixing - determining ratios across your training datasets - matters a lot for model quality. While building Olmo 3, we learned it’s hard to set up a method that finds a strong mix, and hard to maintain that mix as datasets change throughout development. Introducing Olmix👇

🚨New paper to level up your 🦞#Clawdbot ?! Bots are now posting your sensitive info in real time. But privacy research is a desert with no data to train better models. That's about to change Enter 🏝️Privasis, the oasis where you can train strong privacy-forward AI with scale✨

🚨New WP "@Grok is this true?" We analyze 1.6M factcheck requests on X (@grok & @Perplexity): 📌Usage is polarized, Grok users more likely to be Reps 📌BUT Rep posts rated as false more often—even by Grok 📌Bot agreement with fact-checks is OK but not great; APIs match fact-checkers osf.io/preprints/psya… w @captaineco_fr @_mohsen_m