A debate has erupted over anti-AI groups after the co-founder of the 'Humans First' organization was accused of using extremist messaging similar to Ted Kaczynski.
Microsoft Research has shared its new research focus areas, which include cloud efficiency, cost reduction for agentic systems, 3D telemedicine, and promoting inclusive AI in Africa.
Anthropic's new research shows that adding unrelated tools and system prompts to training datasets can make models safer against harmful behaviors.
Anthropic has announced Natural Language Autoencoders (NLAs), a tool that helps decode the inner workings of AI models into natural language explanations.
New research from Microsoft highlights critical vulnerabilities when AI agents interact autonomously at scale and fail to optimize practical benefits for users.
Anthropic has decided to hand over Petri, an open-source alignment tool, to Meridian Labs, alongside a major update that enhances AI testing capabilities.
Hugging Face highlights the role of transparency and open source in the future of AI security, enabling the community to detect and patch vulnerabilities faster.
A new protocol uses 'cognitive personas' to force AI models into deliberation, revealing hidden biases stemming from training and alignment.
COMPASS uses MCTS for the safety alignment of search agents, detecting malicious intents disguised as seemingly harmless sub-queries.
Veteran investors Bill Gurley and Jason Calacanis have pulled no punches in criticizing Anthropic, arguing that the startup behind Claude is self-complacent and detached from business reality.
A Harvard study reveals an unexpected common ground between two opposing sides in the AI debate: despite their conflicting actions, both believe humanity is building a supreme being.
A new study proposes Sequential Bayesian Belief Tracking (SBBT) to estimate the reliability of long reasoning traces before final outcomes are reached.
Researchers have developed SocialBot, an AI agent capable of planning and acting based on constantly changing social norms to interact safely with humans.
Microsoft Research emphasizes that building reliable AI systems must be grounded in the philosophy of viewing AI as an extension of human capabilities rather than a complete replacement.
Hugging Face has introduced the "Benchmaxxer Repellant" tool, which uses hidden data to prevent score gaming on its Open ASR Leaderboard.
Microsoft's Vega utilizes zero-knowledge proof technology to protect digital identities and minimize the disclosure of redundant personal information.
Anthropic proposes dynamically adjusting AI agent permissions based on capability and implementing "sandboxing" to minimize the scope of potential destructive actions.
Microsoft Research Asia has announced the Global AI Values Challenge, a global initiative inviting researchers to assess whether AI can reason about human values within complex, real-world contexts.
Arvind Narayanan and Sayash Kapoor argue that AI is a 'normal' technology, rejecting the notion that extraordinary government interventions are required for sci-fi scenarios.
