University Distinguished Professor, @VT_CS; Thomas L. Phillips Professor, @VTEngineering; Director, @SanghaniCtrVT. ML, AI, urban analytics.
Alexandria, VA
Joined January 2009
- Tweets1.6K
- Following2.8K
- Followers1.1K
- Likes13.7K
Naren Ramakrishnan retweeted
Moritz Hardt thinks that OpenAI and Anthropic’s claims about benchmarks, which are used to test AI models’ capabilities, are misleading. But if the tests aren’t accurate, then how do we know how good the models really are? Read the interview: quantamagazine.org/he-wants-…
Naren Ramakrishnan retweeted
3 Ph.D. students @SanghaniCtrVT to present their research @Voxel51 meetup, 12-2 pm, Oct, 22,
-- @priyapitre advised by @xwang174 & @profnaren
-- @ksmehrab advised by @anujkarpatne & Chris Thomas
-- @yusuf_dalva advised by @PINguAR
Learn more & register: tinyurl.com/ek596nu2
Join us on Oct 22 for the “Advances in AI at Virginia Tech” virtual Meetup! Register for the Zoom: voxel51.com/events/advances-…
Thanks will include:
* Multi-Agent Communication: A framework, diagnostic and mechanistic perspective - Priya Pitre at Virginia Tech
* Exposing and Improving Fine-Grained Visual Grounding Abilities of Lightweight Multimodal LLMs - Kazi Mehrab at Virginia Tech
* Understanding Visual Generative Models for Precise Control - Yusuf Dalva at Virginia Tech
#Robotics #PhysicalAI #VLA #MachineLearning #virginiatech
Naren Ramakrishnan retweeted
I think my favorite line here is Ravi's:
"The best kid with a calculator is also the best kid without one"
Beautifully sums up my feelings on AI use in education...
If you are interested in AI and math, you will want to listen to this lively conversation @JannaLevin and I had with Akshay Venkatesh, Ravi Vakil, and @AlexKontorovich at the ICM.
Naren Ramakrishnan retweeted
Congrats to @profnaren University Distinguished Professor & director @SanghaniCtrVT & 6 other @virginia_tech faculty elected 2026 members of Virginia Academy of Science, Engineering & Medicine. Their work represents the breadth of @VTresearch.
Read more🔗tinyurl.com/3e7xepbr
Naren Ramakrishnan retweeted
I know people think the Monty Hall problem is a game show trick or a silly math puzzle, but the fact that most people get the logic wrong can have real world consequences. I'll walk you through an example I teach in my undergraduate Human Genetics class.
Naren Ramakrishnan retweeted
Ph.D. student @yiqi_su @VT_CS shared photos from the @icmlconf in Seoul, South Korea, where she served a dual role -- presenting two collaborative research papers from @SanghaniCtrVT:
arxiv.org/abs/2602.06323 arxiv.org/abs/2602.00299
Yiqi is advised by @profnaren.
#ICML2026
Naren Ramakrishnan retweeted
Hindsight: Structured Agent Memory that Retains, Recalls, and Reflects
Demo
Christopher Latimer, Nicolò Boschi, Andrew Neeser, Chris Bartholomew, and at VT: @ctrlgaurav @xwang174 @profnaren
arxiv.org/abs/2512.12818
TL;DR: Long-term memory system enables AI agents
Congratulations to Eunice Son 🎉 on a successful MS Thesis Defense of her work on LLM harnesses for working with tabular data! 👏👏 Her work focuses on all the bookkeeping and orchestration that harnesses must do to enable LLMs to reason over massive tables.
Thanks to her committee of Chang-Tien Lu and Vaibhav Garg! Eunice has published in ECIR (European Conf IR) with additional papers under review.
ALT Eunice Son presenting her MS thesis on "Schema-aware Harnesses for Tabular Reasoning with Language Models"
Naren Ramakrishnan retweeted
Ph.D. student @yiqi_su, advised by @profnaren @SanghaniCtrVT, shares photos from @iclr_conf where she presented her collaborative work, Toward World Models for Epidemiology (tinyurl.com/yk37usxy). While at #ICLR2026 Yiqi was also happy for opportunity to meet @ylecun from NYU.
Congrats to @RaquibYousuf on a successful defense of his PhD research! Raquib's work contributes a range of LLM augmentations to support their use with structured and complex information datasets. Thanks to his committee of Chang-Tien Lu, Chris North, @xwang174 and @sathappanspm!
ALT Raquib bin Yousuf w/ slides at his PhD defense.
Naren Ramakrishnan retweeted
Eunice Son, M.S. student @VT_CS was in Netherlands recently to share @SanghaniCtrVT research at European Conference on Information Retrieval. She presented “What Should Search Retrieve Now?”-- collaborative work w/@RaquibYousuf & advisor @profnaren -- at 3rd Search Futures WS.
Naren Ramakrishnan retweeted
Congratulations 🎉Naren Ramakrishnan (@profnaren), Thomas L. Phillips Professor @VT_CS & director @SanghaniCtrVT, elevated to rank of University Distinguished Professor by @virginia_tech Board of Visitors in recognition of significant scholarly work.
tinyurl.com/4szbyrnj
Naren Ramakrishnan retweeted
At KGML Bridge session @RealAAAI conf, Brian Mayer presents @SanghaniCtrVT work from Sha Li Ayush Sadekar @tripswithtires @yiqi_su Annabel Zhang @linseymarr @t_m_murali @profnaren & collaborators on new framework addressing challenges of LLM extraction.
arxiv.org/pdf/2512.10004
Naren Ramakrishnan retweeted
Agent memory is going to be the single most important enterprise technology of 2026.
Most agent memory systems focus on searching through conversation histories.
That won't work anymore.
The way enterprises think about agent memory is about to change forever. Here's why. 🧵
Naren Ramakrishnan retweeted
Hindsight is a collaboration between @Vectorizeio @washingtonpost and researchers @SanghaniCtrVT (Gaurav Srivastava @xwang174 @profnaren). It is based on this research: arxiv.org/abs/2512.12818
With 91% accuracy, open source Hindsight agentic memory provides 20/20 vision for AI agents stuck on failing RAG
venturebeat.com/ai/with-91-a…
Naren Ramakrishnan retweeted
Lance Collins, VP, @vt_dcarea, welcomed invited attendees to Amazon-Virginia Tech Initiative for Efficient and Robust Machine Learning AI Workshop today in Alexandria. @profnaren & Anand Rathi also helped kick off the day-long event.@AmazonScience @virginia_tech @SanghaniCtrVT
Naren Ramakrishnan retweeted
Amazon-Virginia Tech Initiative for Efficient and Robust Machine Learning AI Workshop today featured Lightning Talks by two Ph.D. students & Amazon Fellows -- @haohuiw7 & @yusuf_dalva -- whose research is supported by the initiative.
@AmazonScience @virginia_tech @SanghaniCtrVT
“Agentic” isn’t when you break down your code so f() calls g(). Those are called functions. To be agents, f() and g() need social lives outside the calling sequence.
Naren Ramakrishnan retweeted
🚨#LLMs forget way earlier than they claim!
Even top models (GPT-4o, Gemini-2, and o1) start losing track around 2 K tokens, long before their “128 K context” limits.
Paper: Can an LLM Induce a Graph? Investigating Memory Drift and Context Length
arxiv.org/abs/2510.03611 (1/4)
