The Department of Energy announced funding to support Argonne National Laboratory's AI-for-science initiatives. The aim is to accelerate AI-driven research acrossScientific domains, including data analytics, ML workflows, and AI-enabled simulations. Argonne will use the funding to expand computing resources, strengthen collaborations, and develop the workforce. The program reflects a broader DOE effort to harness AI for scientific discovery and to bolster U.S. leadership in AI-enabled science. The report did not disclose the funding amount or schedule.

• Provide funding for Argonne's AI-for-science initiatives
• Expand computing resources and data infrastructure
• Strengthen collaborations and talent development
• Align with DOE's strategy to accelerate discovery with AI

AI debugging is growing more challenging as agents handle complex tasks beyond chat assistants. Long, probabilistic, multi-agent interactions make root-cause tracing a laborious manual process. Microsoft Research introduces AgentRx, an automated system designed to pinpoint the exact moment an agent trajectory becomes unrecoverable. Traditional metrics like task completion do not reveal failure points. AgentRx supports building reliable and safe AI by gathering evidence and identifying the precise failure point. This approach is intended to aid cloud-incident management and navigation of complex web interfaces, moving beyond simple debugging.

• Identify the exact failure point in agent trajectories
• Replace single-metric debugging with evidence-based diagnostics
• Automate tracing across long, probabilistic multi-agent interactions
• Support safer, more reliable AI deployments
• Improve debugging for cloud incidents and complex web interfaces

Agentic AI systems need models with specialized depth to solve dense technical problems autonomously. They must excel at reasoning, coding, and long-context analysis, while remaining efficient enough to run continuously at scale. Multi-agent systems generate up to 15x the tokens of standard chats, re-sending history, tool outputs, and reasoning steps at every turn. Over long tasks, this “context explosion” causes goal drift, where agents gradually lose alignment with the original objective. And using massive reasoning models for every sub-task—the “thinking tax”—makes multi-agent applications too expensive and sluggish for practical use. Today, we are releasing Nemotron 3 Super to address these limitations. The new Super model is a 120B total, 12B active-parameter model that delivers m

• Present Nemotron 3 Super as a 120B total, 12B active-parameter model
• Address context explosion and goal drift in multi-agent reasoning
• Introduce open hybrid Mamba-Transformer MoE architecture
• Aim to deliver improved efficiency for large-scale, continuous operation
• Target agentic reasoning for solving dense technical problems autonomously