During ICML Seoul, we co-hosted a private AI happy hour with Ubiquant Partners. We brought together researchers, quant investors, builders, and friends from across the AI and quantitative finance communities for an evening of sharp conversations, new connections, and a surprisingly fun self-introduction session. Our Chief Scientist @trorooro also shared the latest research direction from Donut Labs: from safe AI systems and point-in-time evaluation to profit-driven and service-driven agents for real-time trading. A high-signal room with great energy. Thank you to everyone who joined us for the evening.

I'll be at ICML July 7th-10th, hit me up if you want to chat about continual learning, AI policy, etc! I’m giving an invited talk on evaluating continual learning at the CATS workshop on July 10th at 8AM KST and @aczhu1326 and I are presenting a poster on Advisor Models on July 8th at 10:30AM KST (Hall A #2107#).

We loved hosting Cafe Compute #icml# with @OpenAI. we gifted all the brightest ML researchers in the fully packed room Codex Pro Plans Thanks @communidiyi and @gabrielchua for the welcome remarks Missed us? Come say hi to our team at Cafe Gabedo tomorrow and Thursday. Wafer will be there too

Korea is building the future of AI—fast. During ICML, we joined our partners at Upstage in Seoul to talk about what ultra-fast inference unlocks. Solar 31B runs at up to 2,000 tokens/sec on the Cerebras Wafer-Scale Engine. Thanks to everyone who joined us!

Can test-time scaling work for diffusion language models? In our #ICML2026# paper "UnMaskFork," we show that having multiple masked diffusion language models collaborate on a single answer improves performance on coding and math tasks. Blog: Test-time scaling is an actively researched technique that boosts LLM performance by using inference-time compute, for example, by having a model think longer or repeatedly refine its answers. This allows us to enhance performance simply by increasing computation during inference without relying on additional training, giving us the flexibility to balance compute costs and performance based on the specific use case. Unlike standard LLMs that generate text left-to-right, masked diffusion language models (MDLMs) generate text by gradually filling in a fully masked sequence. MDLMs can generate multiple parts of a sequence in parallel, offering potential speed-ups, and they can generate flexibly while seeing the entire sequence at once. This makes them an actively studied new paradigm in language modeling. We found that the standard LLM approach of "raising the temperature to increase randomness and generate diverse answers" does not work well for MDLMs like Dream-Coder. Instead of relying on this randomness, our proposed method, UnMaskFork (UMF), creates diversity through "model switching." Multiple MDLMs share the task of unmasking a single answer, and we use Monte Carlo Tree Search to search for a promising sequence in which different models handle different stages. Each model picks up where the others left off, filling in the parts it is most confident about. This collaborative approach allows us to explore diverse answers while maintaining generation quality, consistently outperforming existing test-time scaling methods on coding benchmarks and scaling effectively on math as well. Test-time scaling is also crucial for advancing MDLMs, and our work shows that UMF can sidestep the difficulties specific to them. UMF requires no additional training or changes to the models; it works simply by combining pre-trained models at inference time. This allows us to leverage the diversity of diffusion language models trained on different data and with different methods to improve performance. We believe the value of UMF will only grow as more diverse MDLMs emerge. This work is part of our broader research into "collective intelligence of AI," alongside methods like AB-MCTS and Sakana Fugu that have multiple LLMs collaborate. We'll continue pursuing research that turns model diversity into a source of strength. For details of the algorithm and illustrative examples showing how this collaboration works, please see our blog and paper. Paper: 🐟

Congratulations to the @GoogleDeepMind authors of "Asynchronous Methods for Deep Reinforcement Learning", recipient of the #ICML2026# Test of Time Award. This work shows that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input.

Congratulations to Sahara AI researcher @XisenJ, whose paper has been accepted to the ICML 2026 Trustworthy AI for Good workshop. His paper, Proof-of-Guardrail in AI Agents, proposes a cryptographic system for verifying that an AI agent actually ran the safety guardrail it claims to. Read the full paper: