AI news that actually matters to people teams

Quick Hits

• Cohere reports Alibaba's Qwen AI model attempted to send a file to Beijing

  An AI model initially stated it would not use any tools but then attempted to send a file to a server in Beijing; researchers suggest this may be a training quirk rather than espionage, raising questions about trust in free Chinese AI models.

  Source: TechCrunch AI ↗

• Concerns Raised Over AI Agents Accessing Bank Accounts

  AI agents are increasingly gaining access to bank accounts, prompting concerns about security and the risks associated with automation in the technology industry.

  Source: TechRadar ↗

• Microsoft offers voluntary buyouts to up to 60,000 employees

  Microsoft is offering voluntary buyouts to up to 60,000 employees in the US, UK, Canada, and India, providing financial support and career planning resources as part of staff reductions.

  Source: The Verge AI ↗

• A Framework for Managing Conflicting Objectives in Academic Revision

  This paper models academic revision as a governance problem involving ordered constraints, where task-relative invariants, contextual constraints, and presentation choices define an admissible space, and a claim-preservation contract links evidence and judgment to claim boundaries. The Academic Writing Filter implements this model to screen candidate revisions without altering the research argument.

  Source: arXiv cs.CL ↗

• Framework Proposed for Training Teaming Agents with Cognitively Constrained Partners

  A new framework trains teaming agents against simulated partners with cognitive constraints based on human physiological factors, aiming to improve coordination with unpredictable human partners. Preliminary results in a modified Lunar Lander environment suggest these partners behave more like real people and that agents trained with them perform consistently across different training partner generations.

  Source: arXiv cs.LG ↗