Surprisingly, we find training only with incorrect traces leads to strong performance 🤯 Even more interesting: it improves model diversity and test-time scaling—while correct traces do the opposite. Check out the 🧵👇

Mach 0.9 flow on GPU takes only 2minutes to do 200k iterations. Interesting 2D case which takes a while to stabilize. No smoothing used but there is a source term needed to combat the instabilities. Code here: github.com/garcfd/lax-wen…

Quantum Computation by Peter Shor of MIT Lecture notes: math.mit.edu/~shor/435-LN/ (audio generated by NotebookLM)

I am pleased to inform you that my outline funding application on Physics-Informed Machine Learning was rejected. Funding bodies don't seem to be interested in funding research into understanding the limitations and drawbacks of #PIML or #AI4Science, as they call it now. I tried!

How did you get into computer science? youtube.com/shorts/JHTV8qp…

Evaluating LLM-based Agents This report has a comprehensive list of methods for evaluating AI Agents. Don't ignore evals. If done right, they are a game-changer. Highly recommend it to AI devs. (bookmark it)

Universe MDPI MDPI Physical Sciences Aerospace Astronomy Geosciences MDPI Mathematics MDPI Remote Sensing MDPI Computational Sciences & Engineering Physics is not a belief system. Comets💫 are not Asteroids☄️. 💫SL9 supercomp-sim should be studied by geophysicists & geologists. Scale invariance in fluid dynamics is the key to unlocking geomorphological anomalies found on Earth. NASA Ames🖖🏽 Planetary Society✨

I am tired of Physics-Informed crap! 📢😠 Please share some good recent papers on modelling and simulation that you have written or came across. 🙏

Small Language Models are the Future of Agentic AI Lots to gain from building agentic systems with small language models. Capabilities are increasing rapidly! AI devs should be exploring SLMs. Here are my notes:

In our latest paper, we discovered a surprising result: training LLMs with self-play reinforcement learning on zero-sum games (like poker) significantly improves performance on math and reasoning benchmarks, zero-shot. Whaaat? How does this work? We analyze the results and find

Excited to share our new ICML paper, with co-authors Csordás Róbert and Jürgen Schmidhuber! How can we tell if an LLM is actually "thinking" versus just spitting out memorized or trivial text? Can we detect when a model is doing anything interesting? (Thread below👇)

Since 1990, we have worked on artificial curiosity & measuring „interestingness.“ Our new ICML paper uses "Prediction of Hidden Units" loss to quantify in-context computational complexity in sequence models. It can tell boring from interesting tasks and predict correct reasoning.

The best academic papers don't just fill knowledge gaps. They reframe entire conversations. Instead of asking "What hasn't been studied?" Ask "What have we been studying wrong?" That shift in perspective is what turns incremental research into novel work.

I believe all professors in the field of AI and machine learning at top universities need to face a soul-searching question: What can you still teach your top (graduate) students about AI that they cannot learn by themselves or elsewhere? It had bothered me for quite some years

ᑕOՏᗰIᑕ ᗰᗴՏՏᗴᑎᘜᗴᖇ ≈ 𝕃𝕦𝕚𝕤 𝔸𝕝𝕗𝕣𝕖𝕕𝕠⁷ ∞∃⊍ Imagine if a comet💫chunk slammed Earth! Full plasma-physics smashing 🌎’s crust at Mach >30. I think the hyper-velocity of 💫plasma isn’t factored by academia & they’re stuck in asteroids☄️ & 💫 are the same? Do ☄️ have a coma saturated w/atomized gases⛽️? Computational Sciences & Engineering 🖖🏽

This time, I’ll be presenting part of our latest work in #CFD for electrochemical applications. This time I’ll be showing something we developed for applications with large grid deformations (morphing) !!

100 years have gone by, and we still donot know what it really means.

Do you want #thePostdoctoral to publish a postdoctoral Android/iOS app? Features⁉️

Evaluating language models is tricky, how do we know if our results are real, or due to random chance? We find an answer with two simple metrics: signal, a benchmark’s ability to separate models, and noise, a benchmark’s random variability between training steps 🧵

Claim: gpt-5-pro can prove new interesting mathematics. Proof: I took a convex optimization paper with a clean open problem in it and asked gpt-5-pro to work on it. It proved a better bound than what is in the paper, and I checked the proof it's correct. Details below.