Tag
AllenAI introduces FairStress, a benchmark measuring demographic fairness in LLM decision-making beyond final choices, testing how easily adversarial arguments make models change decisions across demographic groups in high-stakes domains like hiring, lending, and healthcare.
Ai2's AI Infrastructure team details replacing their priority-based GPU scheduler with a system using GPU time budgets, hierarchical fair-share allocation, and time-slicing contracts to prioritize high-impact research while maintaining full cluster occupancy across thousands of H100, B200, and B300 GPUs.
AllenAI open-sources AstaBrief 8B, a model trained specifically to generate cited scientific reports from retrieved literature, achieving ~3.5× faster generation than proprietary models while releasing the weights, training data, and a reproducible workflow for building reports from local PDFs.
AllenAI releases MolmoMotion, a vision model designed to predict future motion based on a short history of frames.
Allen AI releases MolmoAct2, an open-weight Vision-Language-Action model designed for real-world robotic deployment, featuring new datasets, an open action tokenizer, and adaptive reasoning to reduce latency.
AI2 releases Olmo 3, a family of fully-open state-of-the-art language models at 7B and 32B parameter scales, with the entire model lifecycle — every checkpoint, data point, and dependency — fully open.