Think your LLM is smart? Try Big-Bench EXTRA Hard. The real test. #Reasoning #Benchmark #LLM

🚨 Excited to introduce PairBench! 🚨 💡 TL;DR: VLM-judges can fail at data comparison! ✅ PairBench helps you pick the right one by testing alignment, symmetry, smoothness & controllability—ensuring reliable auto-evaluation. 📄Paper: arxiv.org/abs/2502.15210 🧵 Thread: 👇

🚀 New Paper Alert! Can we generate informative synthetic data that truly helps a downstream learner? Introducing Deliberate Practice for Synthetic Data (DP)—a dynamic framework that focuses on where the model struggles most to generate useful synthetic training examples. 🔥

New Paper Alert!📄 "It’s better to be sparse than to be dense" ✨ We explore how to steer LLMs (like Gemma-2 2B & 9B) by modifying their activations in sparse spaces, enabling more precise, interpretable control & improved monosemanticity with scaling. Let’s break it down! 🧵

Introducing nanoAhaMoment: Karpathy-style, single file RL for LLM library (<700 lines) - super hackable - no TRL / Verl, no abstraction💆‍♂️ - Single GPU, full param tuning, 3B LLM - Efficient (R1-zero countdown < 10h) comes with a from-scratch, fully spelled out YT video [1/n]

Happy to announce "Performative Prediction on Games and Mechanism Design" was accepted at AISTATS Conference 2025, and got spotlight at HAIC(ICLR 2026 workshop) with Mehrnaz Mofakhami Fernando P. Santos Gauthier Gidel Simon Lacoste-Julien (Mila and UvA) arxiv.org/abs/2408.05146 Details below 1/9🧵

ICLR 2025 many many many thanks to Kyunghyun Cho and Yoshua Bengio for enabling the wildest ever start of my research career 2014 was a very special time to do deep learning, a commit that changes 50 lines of code could give you a ToT award 10 years later 😲

Happy to host Karan tomorrow to talk about their work on one-minute video generation with TTT.

Excited to be at #ICLR2025 next week! I'm currently on the job market for Research Scientist positions, especially in generative modeling, synthetic data, diffusion models, or responsible AI. Feel free to reach out if you have any openings!

This week I'll be at #ICLR25. If you like fundamental optimization results, I'll be presenting our work on surrogate losses for non-convex-concave min-max problems and learning value functions in deep RL (VIs more generally). Poster: #377 Thursday April 24 10am-12:30pm

Happy to share that Compositional Risk Minimization has been accepted at #ICML2025 📌Extensive theoretical analysis along with a practical approach for extrapolating classifiers to novel compositions! 📜 arxiv.org/abs/2410.06303

Deliberate practice is accepted to #ICML2025 as a spotlight (top 2.6%!) 🚀

1. We often observe power laws between loss and compute: loss = a * flops ^ b + c 2. Models are rapidly becoming more efficient, i.e. use less compute to reach the same loss But: which innovations actually change the exponent in the power law (b) vs change only the constant (a)?

It's very difficult to improve the *exponent* in scaling laws for loss vs compute, especially by changing the optimizer! Our new paper shows that scaling momentum correctly can *provably* improve the scaling exponent on a theoretical model. Empirically, it works on LSTMs too!

Happy to share that our work has been accepted at #COLM2025! Use #SAS to steer your LLM in sparse spaces.

🎉Personal update: I'm thrilled to announce that I'm joining Imperial College London Imperial College London as an Assistant Professor of Computing Imperial Computing starting January 2026. My future lab and I will continue to work on building better Generative Models 🤖, the hardest

📄 New Paper Alert! ✨ 🚀Mixture of Recursions (MoR): Smaller models • Higher accuracy • Greater throughput Across 135 M–1.7 B params, MoR carves a new Pareto frontier: equal training FLOPs yet lower perplexity, higher few‑shot accuracy, and more than 2x throughput.

Heading to Conference on Language Modeling in Montreal? So is WiML! 🎉 We are organizing our first ever event at #CoLM2025 and we want you to choose the format! What excites you the most? Have a different idea? Let us know in the replies! 👇 RT to spread the word! ⏩

We built a new 𝗮𝘂𝘁𝗼𝗿𝗲𝗴𝗿𝗲𝘀𝘀𝗶𝘃𝗲 + 𝗥𝗟 image editing model using a strong verifier — and it beats SOTA diffusion baselines using 5× less data. 🔥 𝗘𝗔𝗥𝗟: a simple, scalable RL pipeline for high-quality, controllable edits. 🧵1/

Lol 😂