Ai2

Headlines · 4

1. 5d
   Ai2's HiRO-ACEv2 emulates extreme weather at 3 km for a fraction of X-SHiELD's cost

   Ai2 introduced HiRO-ACEv2, the newest update to its High Resolution Output Ai2 Climate Emulator system, which pairs a coarse 100 km ACE model with the HiRO downscaling model that sharpens outputs to 3 km detail. The main change is the new HiROv2, which predicts several high-resolution fields at once instead of only rainfall and, according to Ai2, matches X-SHiELD's high-resolution statistics to within 3% even for events occurring once in a million data points. Ai2 also retrained ACE and says a combined run produced a full year of global 3 km data that closely matches X-SHiELD; however, HiRO only works with the ACE version tuned to X-SHiELD and cannot yet handle other climates such as a warmer future or ERA5.

   Ai2 Blog · 🔥 0
2. 6d
   Ai2: Molmo 2's open release enabled an MIT researcher's salmon-tracking model

   Ai2 published an account of how Kai Van Brunt, an MIT '25 graduate working in MIT's BeeryLab, built Molmo2Fish on the open vision-language model Molmo 2 to track salmon in underwater sonar video and revise its tracking predictions from everyday language instructions. According to the post, Molmo 2 can process a whole clip and output tracks directly; after fine-tuning on salmon videos, Molmo2Fish performed roughly as well as a specialized fish-tracking system before receiving any corrections, with results varying across rivers. The experiments showed language-guided correction helped with clear instructions but struggled to improve the model's own strongest predictions, and Molmo2Fish remains an undeployed research project.

   Ai2 Blog · 🔥 0
3. Oct 9
   Ai2 replaces priority scheduler with fair-share GPU budgets and time-slicing contracts

   Ai2's AI Infrastructure team says it replaced its priority-based cluster scheduler with GPU time budgets, hierarchical fair-share allocation and a time-slicing contract for dividing scarce compute among research projects. Ai2 manages thousands of NVIDIA H100, B200 and B300 GPUs in clusters ranging from 88 to 1024 GPUs, serving about 150 internal researchers, with submitted workloads requesting 2–3x the available capacity at any time. The post says the old setup led to GPU "squatting," priority inflation and manual negotiations over shutting down jobs, and that the new contract asks workloads to declare a minimum runtime during which they cannot be preempted before being requeued. Ai2 also reports a 74% reduction in repairs requiring human intervention, a self-reported figure, and says it is investigating whether capacity fragmentation could increase queue waits for the largest workloads.

   Ai2 Blog · 🔥 0
4. Oct 9
   Ai2 presents Olmo Hybrid research, outlines next Olmo

   Ai2 presented its paper "Olmo Hybrid: From Theory to Practice and Back" at COLM 2026, examining what happens when a language model combines transformer attention with linear recurrent layers. Ai2 says the model reached the same MMLU accuracy as Olmo 3 7B using 49% fewer training tokens — a result the lab reports itself, with no independent measurement supplied. Ai2 also said its next Olmo model is now in pre-training with a hybrid mixture-of-experts architecture combining attention and recurrence, and that it released Olmo-core 3, a redesigned system for training large mixture-of-experts models, ahead of the conference. The post additionally cites the open-weights AstaBrief model, the Bolmo byte-level paper's publication in Nature with extra checkpoints, and Google's recreation of the Olmo 3 7B training run in MaxText on Cloud TPUs.

   Ai2 Blog · 🔥 0

Experience and discussion from the community

Share my Ai2 experienceAsk about Ai2