Shreyas Havaldar

🪩 Delighted to share that I'm joining @eliasbareinboim's Causal AI Lab @Columbia as a PhD student :) Also honored to share that I'll be supported by the Presidential Fellowship! Excited to learn from & work w/ the larger ML group @CUSEAS; @zemelgroup, Daniel Hsu et al. as well

Super psyched to finally share our new continual learning paper “Tell Me What To Learn: Generalizing Neural Memory to be Controllable in Natural Language”

Just uploaded a recent NeurIPS talk: "Towards Causal AI for Embodied World Models", youtu.be/JJ9PQTYciZU Explores how causality can help ground generalization, decision-making, and representation in AI systems. Happy New Year, and here’s to pushing the field forward in 2026!

@divy93t congrats @divy93t, onwards and upwards! 🫶

Thrilled to share a professional update -- I'm starting a new effort at Google DeepMind! I'll be exploring / building the next paradigm of human-A(G)I collaboration and the future of interaction with advanced models. Grateful to be spending time with incredible collaborators to figure this out! If you're building / researching / investing in this space - let's chat, DMs open (also at NeurIPS in SD).

1/8 Looking forward to reconnecting with colleagues and meeting new ones at NeurIPS next week in San Diego! Come by our posters to discuss our latest work and upcoming challenges, with highlights in the thread below. I’ll also be around during tutorials. Feel free to reach out!

PhD apps are daunting. Columbia's PAR is one of the best programs to get detailed feedback from people who've been there and done that! Apply now :)

Everyone says they want general-purpose robots. We actually mean it — and we’ll make it weird, creative, and fun along the way 😎 Recruiting PhD students to work on Computer Vision and Robotics @umdcs for Fall 2026 in the beautiful city of Washington DC!

Happy to share that it’s been an exciting season for our team with several NeurIPS papers accepted! These projects span causal representation learning, robust reinforcement learning, causal discovery, and interpretable modeling -- highlighting the development of causal foundations and their role in advancing diverse areas of AI. Many people contributed, special kudos to all students and collaborators who made this possible -- Yushu Pan, @adiba_ejaz, @Mingxuan0422 , @JunzheZhang12 , Tara Vafai Anand, Adèle Ribeiro, @InwooRyanHwang, Min Woo ­Park, Andy Arditi, @sanghack . Here are the preprints: --Counterfactual Image Editing with Disentangled Causal Latent Space Y. Pan, E. Bareinboim. causalai.net/r137.pdf --Less Greedy Equivalence Search A. Ejaz, E. Bareinboim. causalai.net/r134.pdf -- Confounding Robust Deep Reinforcement Learning: A Causal Approach M. Li, J. Zhang, E. Bareinboim. causalai.net/r132.pdf -- A Hierarchy of Graphical Models for Counterfactual Inferences H. Yang, E. Bareinboim. causalai.net/r130.pdf -- Causal Discovery over Clusters of Variables in Markovian Systems T. Anand, A. Ribeiro, J. Tian, G. Hripcsak, E. Bareinboim. causalai.net/r128.pdf -- From Black-box to Causal-box: Towards Building More Interpretable Models I. Hwang, Y. Pan, E. Bareinboim. causalai.net/r127.pdf -- Structural Causal Bandits under Markov Equivalence M. Park, A. Arditi, E. Bareinboim, S. Lee. causalai.net/r122.pdf Feel free to reach out with questions -- excited to continue the conversation at NeurIPS in San Diego!

@varunyer New York effect?

Me: Why am I so tired all the time!? Also me:

@JeffDean @GeminiApp I'm surprised you didn't go for a specific classic red and white jersey :p

Our latest Gemini image generation and editing model is quite good (see the examples in the thread below). It lets you indulge in a bit of creative fun, helps you make new business cards, etc! Try it out at gemini.google.com @GeminiApp

🚨Come check out our poster at #ICML2025! QuantSpec: Self-Speculative Decoding with Hierarchical Quantized KV Cache 📍 East Exhibition Hall A-B — #E-2608 🗓️ Poster Session 5 | Thu, Jul 17 | 🕓 11:00 AM –1:30 PM TLDR: Use a quantized version of the same model as its own draft for speculative decoding. It’s fast, memory-efficient, and works great for long context—no extra draft model needed. 2.5× End-to-End generation speedup is achieved. 🔥 🔗 icml.cc/virtual/2025/p…

5/5 Last but definitely not least, I’m honored to be giving a keynote on Wednesday (7/23) titled "Towards Causal Artificial Intelligence." For details, see: auai.org/uai2025/keynot… Here’s a short abstract: While many AI scientists and engineers believe we are on the verge of achieving highly general forms of AI, I offer a critical appraisal of this view through a causal lens. Building on foundational developments in the field, I’ll present my perspective on the relationship between intelligence and causality — and the central role of the latter in constructing intelligent systems and advancing credible data science. I frame the discussion around five core capabilities we should expect from an intelligent AI system: 1. Performing causal reasoning and articulating explanations; 2. Making precise, surgical, and sample-efficient decisions; 3. Generalizing across changing conditions and environments; 4. Generating and simulating in a causally consistent manner; and 5. Learning causal structures and variables. In this talk, I’ll expand on this perspective and share current progress toward building causally intelligent AI systems. A more detailed discussion of this thesis is provided in my forthcoming textbook, a draft of which is available here: causalai-book.net.

Hi all, if you're attending ICML (Vancouver) or UAI (Rio de Janeiro), I'm happy to share some news from the lab! Please check it out -- and feel free to drop by or shoot me a line if any of it sounds intriguing. 1/5 "Counterfactual Graphical Models: Constraints and Inference" (w/ Juan Correa) Thu, 4:30 PM (East, 1802) Link: causalai.net/r115.pdf Counterfactuals sit at the top of the causal hierarchy and are central to explanation, credit/blame, and retrospective analysis. Yet, reasoning about them systematically has remained a longstanding challenge. This work introduces the counterfactual calculus, a generalization of Pearl’s do-calculus to the counterfactual settings -- enabling identification from both observational and experimental data. It closes a question left open nearly 25 years ago, when Pearl introduced the interventional calculus and formalized counterfactual semantics in his seminal Biometrika paper. We also develop a new graphical structure—the Ancestral Multi-World Network (AMWN)—along with the first algorithm that is efficient, sound, and complete for reading counterfactual independencies from a causal model. AMWNs subsume several existing approaches, including the twin-network construction, and extend them to multiple worlds. This is the first unified framework that connects counterfactual constraints, structure, and inference.

@abeirami Thank you for the super helpful and cheerful guidance and brainstorming :) All the best for the next part of your journey, onwards and upwards!

After three incredible years, today is my last day at Google DeepMind! I am truly grateful to the amazing colleagues who made the journey 1000x more fruitful and enjoyable! I am forever indebted to my collaborators who showed me how to be better at everything via demonstrations.

@adityakusupati @_keysarasara @X @UW Congraaaaats! 🥳🫶🙂‍↕️

In other news, I got married to the beautiful and brilliant @_keysarasara this Sunday! Why am I posting this to @X? A serendipitous tweet in August 2021 of her arrival @UW in my feed led me to extend my help and rest is history 🤣 P.S: Hi Sanju, I finally liked the tweet 😜

@JangraAnubhav Woohoo congrats! 🫶🥳

Excited to share my first PhD paper acceptance! 🤓 Our survey on automatic hint generation towards personalized learning was accepted in TACL'25. We highlight the gaps and future possibilities in this exciting interdisciplinary space. Pre-print: arxiv.org/abs/2404.04728

@jenny_ma_ Woohoo congrats! 🫶🫡🥳 Can vouch for DynEx, it's amazing 🙂‍↕️

Super excited to share that my first, first-author paper has been accepted to #CHI2025​​ ​🌸🇯🇵! Our system, DynEx bridges design and engineering to help users flesh out their ideas through a Design Matrix and generate code to create a functional app 🖥️ 📖: arxiv.org/abs/2410.00400

today is exactly 25 years since the greatest newspaper comic strip ever was published

Thank you to the people that aggressively nod during presentations.

@vbv_kr Here we go :) Wordle 1,305 4/6* ⬜⬜🟨⬜⬜ ⬜🟩⬜⬜⬜ ⬜🟩🟨⬜🟨 🟩🟩🟩🟩🟩 Connections Puzzle #583 🟩🟩🟩🟩 🟨🟨🟨🟨 🟦🟦🟦🟦 🟪🟪🟪🟪 Strands #317 “Bundle up” 🔵💡🔵💡 🔵🟡🔵🔵 🔵 I solved the 1/14/2025 New York Times Mini Crossword in 0:49! fpx3r.app.goo.gl/PKC4

🟨🟨🟨🟨 🟩🟩🟩🟩 🟦🟦🟦🟦 🟪🟪🟪🟪 🔵🟡🔵🔵 🔵🔵🔵 🟨⬛🟨⬛⬛ ⬛⬛⬛⬛⬛ 🟩🟩🟩🟩🟩 😭😭🕺🕺

@bertyashley Pax had me bamboozled. Loved the theme! I'll take the solid 9/10 to start tbe year :)

Happy New year! The theme for this week's quiz is: What have the Roman gods given our language? thehindu.com/society/histor…