Thomas Kipf
@tkipf
Research at @GoogleDeepMind. Controllable World Simulators (GNNs, Structured World Models, Neural Assets). TLM Veo Controls (Ingredients & more).
My PhD thesis "Deep Learning with Graph-Structured Representations" is now available for download: hdl.handle.net/11245.1/1b63b9… -- It covers a range of emerging topics in Deep Learning: from graph neural nets (and graph convolutions) to structure discovery (objects, relations, events)
For humans, mathematical symbols (and formal systems like lean) are *tools* we learn how to use, not a structure that wraps around us. I think that's the right role for formal still manipulation: a tool that can be employed by an intelligent system if/when it supports a goal.
Video models (here: Veo) are more general reasoners than you’d think
We just discovered the 🔥 COOLEST 🔥 trick in Flow that we have to share: Instead of wordsmithing the perfect prompt, you can just... draw it. Take the image of your scene, doodle what you'd like on it (through any editing app), and then briefly describe what needs to happen…
💖 Loved this @lexfridman conversation with Demis! His excitement about how our video model, Veo 3, is learning intuitive physics (ex: fluid dynamics) just from watching videos is contagious... like reverse-engineering the physical world, or reality itself. Very proud to be…
Imagine if every pattern shaped by nature – like a protein’s fold or cosmic phenomena – is inherently learnable by AI. @DemisHassabis shares with @lexfridman that if AI can learn these natural patterns, we could open doors to new eras of scientific discovery. Listen now. ↓…
"Prioritization". I prioritized what to do based on a simple metric "(impact remaining)/(# top talent working)". Each research, model, product, or business has finite impact, and as more progresses are made and more talented people notice its impact and get in, this metric…
1) Intelligence is the process of minimizing the generator-verifier gap 2) The generator-verifier gap is a fundamental property of the universe 3) ASI is the optimal such process
If a solution is fundamentally easier to be verified than to be generated, it probably means that there is a learning signal. Lots of problems fall in this category.
Open position in the Veo team - come join us!
Want to be part of a team redefining SOTA for generative video models? Excited about building models that can reach billions of users? The Veo team is hiring! We are looking for amazing researchers and engineers, in North America and Europe. Details below:
Official gold-medal performance with Gemini at IMO. Massive congrats to the team! “This year, we were amongst an inaugural cohort to have our model results officially graded and certified by IMO coordinators using the same criteria as for student solutions.”
Advanced version of Gemini Deep Think (announced at #GoogleIO) using parallel inference time computation achieved gold-medal performance at IMO, solving 5/6 problems with rigorous proofs as verified by official IMO judges! Congrats to all involved! deepmind.google/discover/blog/…
Look at that gap
We just released the evaluation of LLMs on the 2025 IMO on MathArena! Gemini scores best, but is still unlikely to achieve the bronze medal with its 31% score (13/42). 🧵(1/4)
Great initiative 👇
NeurIPS is pleased to officially endorse EurIPS, an independently-organized meeting taking place in Copenhagen this year, which will offer researchers an opportunity to additionally present their accepted NeurIPS work in Europe, concurrently with NeurIPS. Read more in our blog…
Happens today! 🗓️Tue, July 15 @ 11 AM 📍East Exhibition Hall A-B #E-3512 Unfortunately I was not able to attend, so please DM if you want to chat about hierarchical models, irregular geometries or scalable physical modeling :) @FEijkelboom will present the poster for me on-site
🤹 New blog post! I write about our recent work on using hierarchical trees to enable sparse attention over irregular data (point clouds, meshes) - Erwin Transformer. blog: maxxxzdn.github.io/blog/erwin/ paper: arxiv.org/abs/2502.17019 Compressed version in the thread below:
🚨 First Call for Participation – NeSy 2025 📍 Sept 8–10 | Santa Cruz, CA Join the longest-running conference on neurosymbolic AI! Our keynote speakers: @guyvdb , @tkipf , @dlmcguinness , @GaryMarcus More info 👇
We’re thrilled to share that the first in-person LoG conference is officially happening December 10–12, 2025 at Arizona State University logconference.org Important Deadlines: Abstract: Aug 22 Submission: Aug 29 Reviews: Sept 3–27 Rebuttal: Oct 1–15 Notifications: Oct 20
I use them daily since a couple of months and they're amazing. The AI features are hit-and-miss, though. Photos/videos & audio/mic are awesome. Touch pad activates too easily.
After wearing Ray-Ban Meta Wayfarer glasses for a few weeks ... I feel kind of naked wearing regular sunglasses. I've found three use-cases that are hard to roll back: * Spontaneous photos of my kids when we're out and about. Any cool pose that has a half-life of 3 seconds I…
This is a fantastic opportunity to work at the frontier of AI + materials science!
🚨 Our team at GDM is hiring a research engineer to work on topics around RL, post-training + materials science! Role is based in Mountain View. DMs open if you have questions.
Releasing the Energy-Book 🔋 from its first appendix's chapter, where I explain how I create my figures. 🎨 Feel free to report errors via the issues' tracker, contribute to the exercises, and show me what you can draw, via the discussion section. 🥳 github.com/Atcold/Energy-…