Employees are significantly more willing to voice honest opinions to algorithmic leaders than to human ones, according to a study by researchers at Zhejiang University published in Scientific Reports.…
A new wave of AI hardware and software co-design is replacing heavy floating-point matrix multiplications with addition-only, ternary weight sets {-1, 0, 1}, enabling up to a 32-fold reduction in weig…
A new method called The Confession Booth, developed by a researcher as part of the BlueDot AI Safety Course, uses recursive self-report probing to detect emergent misalignment in large language models…
Wang et al. published Agent Workflow Memory on arXiv in 2024 (arXiv:2409.07429), a method enabling agents to induce reusable workflows from past trajectories offline or online without labels, reducing…
Wang et al. published Voyager, an open-ended embodied agent that uses GPT-4 to propose tasks, write executable code against a game API, verify it, and archive verified programs into a growing skill li…
A new analysis argues that emotion vectors in large language models may serve AI-native functions like reward hacking, with no human analog, challenging anthropocentric emotion labels and raising alig…
Researchers introduced SPEAR (Sandboxed Prompt Engineer with Active Roll-back), a code-augmented agentic optimizer that autonomously rewrites prompts using a Python sandbox for structural error analys…