Big news and a big shout‑out to Fangshuo (Jasper) Liao. Over ~5 years together (UG → MS → almost‑PhD), Jasper led a result on joint training for Mixture‑of‑Experts. Most theory separates router/experts or uses toy top‑1 routing. We handle soft/top‑K‑style joint training in a

1/ Thrilled to share the CrysFormer journey—3 years, multiple papers, code/data incoming. Thanks to co-authors (Tom Pan, Chen Dun, Shikai Jin, Evan Dramko, Ria Stevens, Mitchell D. Miller, George N. Phillips Jr.) and Welch Foundation + partners. 2/ Start: IUCrJ (2023) showed a

“LLMs just memorize.” We say it like it’s a flaw. But most of human problem-solving is memory + experience + small epiphanies. The real question isn’t “Do they memorize?”—it’s “Can they retrieve, compose, and adapt what they know to new constraints?” Thanks @RiceUniversity

Every week: a new acronym, “paradigm,” clever rebrand. Great for attention and funding—but it can feel like constant reinvention when many results are old ideas + scale. Plenty of papers today = what would’ve been a sharp blog post: smarter data transform, better schedule,

Using LLMs ≠ building them ≠ studying smaller ones (that’s me). Question: Do LLMs “reread your entire chat” every turn? My mental model: - Finite context window = working memory. - New message + most relevant recent history get packed in. - If too long,

Hot take: “Superintelligence” is the wrong yardstick for lab automation. Intelligence is lumpy: internet-scale text → strong literacy; wet labs/physics/clinics ≠ day-one mastery. You can get real value without “super.” One new angle on domain data is a win. Labs need

As a greek scientist, I'm obliged to read this

Hot takes about AI are fun. Designing great courses is underrated exciting. As we build “Math of AI” (COMP 282) for Rice CS’s new AI major, I’m reflecting on COMP 414 (Optimization). Highlights: 150+ in the project Slack, 100+ pages of double-column notes (w/ exercises), two

AAAI-26 just added AI-generated reviews to every paper. If these work “better” than humans, it’s not that we’re worse;it’s that we’re busy. Open questions: - How do we track long-term bias drift (topic, citation gravity, method conservatism)? - Will authors “jailbreak” the

1/ 29k “ideas” ≠ 29k novelties. AAAI-26 saw ~29k submissions (~23k under review). We’re producing faster than we can digest; flooding the system and blurring true novelty. 2/ Minor tweaks, new data, recombinations aren’t bad. But at this volume, reviews drift from shared ground

🌟 Reflections on AI and Computer Science Research 🌟 Reading the recent paper "Hierarchical Reasoning Model" ( https://t.co/6S6CngLuMc) sparked a thought: What purpose does AI (and CS) research truly fulfill? Other disciplines claim grand challenges (with a huge grain of salt):

No hype, just breakthroughs: Rice AI summit quietly makes waves in Paris. 🔗 Learn more: https://t.co/2bClE4XXkn The summit was sponsored by @RiceUniversity, Rice Global, @RiceEngineering and @ricekenkennedy. @RiceECE @RiceCompSci #SolvingforGreaterGood

Recently, I delved into the role of AI in PK-8 schools, a topic that resonates with my commitment to blending technology and education, as part of the AI Major at Rice University. And I will focus on this age range as I feel it more gentle and "innocent": Kids should be kids, and

Had a night chat with a friend back home that made me rethink AGI's "general" intelligence claim. We have linguistic, spatial, emotional, kinetic, interpersonal intelligence - yet we act like next-word prediction covers it all. Society shifted from "strongest wins" to "smartest

🚀 The Future of Computer Science: An Indispensable Partner in Innovation 🚀 In an era dominated by AI, it's easy to misunderstand the role of computer science (CS). Many well-meaning voices, including some high-profile figures, have propagated this myth, but let’s set the

🚀 Dear PhD Students, (Boilerplate but good to remind) As you navigate the world of research, remember this: 1. Ideas are abundant. But execution? That's where the challenge lies. What looks great on paper may stumble in practice. 2. Small successes are just the beginning.

🚀 Since the NeurIPS deadline, I’m feeling like a PhD student again! 🎓 I’ve been working on a polynomial, fully parallelizable algorithm for MaxCUT, a project that traces back to my Master’s thesis. Here’s why: 0. Refreshing Detachment: Stepping back from the LLM madness has

🌟 Just wrapped up the RiP-SILO workshop! Huge thanks to all the speakers for their insightful talks in such an academically-friendly environment. The event was refreshing, with friendly vibes and ample time for dialogue. We embraced diverse perspectives—from graph theory to AI

🚀 Exciting insights from our latest research at Rice University, in collaboration with MSR! Our paper introduces Dynamically Decentralized Orchestration of Mixtures of Experts (DDOME), highlighting: - Joint Training is Key: Combining gating functions with experts leads to

AI's evolution: We've moved from "building better algorithms" to "finding better problems" 🎯 (highly recommended read: https://t.co/7cmWONeKNo) AlphaEvolve ( https://t.co/Da6fVSy3UH) just beat Strassen's 56-year-old matrix multiplication record & solved dozens of open math