Yep. Expect a lot of internet services to break soon unless things get better. Can’t pull images from ECR for instance. The Modal team is randomly at an offsite in Italy so we’re all hands on deck!

IN NEWS: Nebius lands a $17.4B partnership with Microsoft to provide high-performance AI infrastructure. The CRO @marcboroditsky told us: "We believe the real opportunity is delivering the AWS equivalent for AI.” “The $100B+ business opportunity is servicing the hyperscaler-level requirements of this new class of AI customers, while also supporting the single engineer building from scratch as they scale.”

One of the key components for training code #agents is access to verifiable real-world tasks, which are not easy to collect. We automated and scaled this process using @TractoAI as our main platform for data processing and storage, and decided to #opensource it. 3/4

Our customer SieveStack is building the world’s largest dataset of molecular simulations to train a multi-layered stack of foundational models and advance dynamics-driven drug discovery. Read the story: nebius.com/customer-stories/… 🔹 Goal: To unlock treatment pathways for hard-to-treat diseases with AI-powered, physics-based modeling and biochemistry. 🔹 Solution: To generate high-precision molecular dynamics simulations and optimize model training with a mixed-precision approach — FP32 for accuracy and BF16 for performance. 🔹 Result: SieveStack worked with Nebius and @TractoAI support teams to prototype, debug and scale foundational models with GPU utilization of more than 90%, revealing drug-target interactions beyond the reach of lab experiments.

Anyone at ICML wanna see the future of world models? We're walking around with a laptop running our world model at 500 FPS+ (fully local). Would love to demo/chat with anyone interested

What if models could learn which problems _deserve_ deep thinking? No labels. Just let the model discover difficulty through its own performance during training. Instead of burning compute 🔥💸 on trivial problems, it allocates 5x more on problems that actually need it ↓

Our AI R&D team presents SWE-rebench, a new benchmark for evaluating agentic LLMs on a continuously updated and decontaminated set of real-world software engineering tasks mined from real GitHub repos. Explore the leaderboard and methodology behind it: swe-rebench.com

Read how @synth_labs, a startup developing AI solutions tailored for logical reasoning, is advancing AI post-training with our @TractoAI: nebius.com/customer-stories/… 🔹 Goal: Develop an ML system that empowers reasoning models to surpass pattern matching and implement sophisticated search and exploration strategies. 🔹 Solution: Build scalable training infrastructure. Reasoning models require large datasets and distributed computing, making multi-node training and high-performance GPUs essential for effective results. 🔹 Results: Using TractoAI, a serverless platform on Nebius AI Cloud, SynthLabs trains AI reasoning models, laying the foundation for next-gen reasoning systems and enterprise use.