Anthropic has released Cowork, a new Claude Desktop agent designed to interact with user files and perform tasks without requiring any coding knowledge, making advanced AI more accessible.
Tesla has launched its autonomous robotaxi service in Dallas and Houston, marking a significant expansion of its self-driving technology into major urban centers.
A reported code leak from Anthropic's Claude AI has exposed critical command injection vulnerabilities, raising serious security concerns for the advanced language model.
Cerebras, a prominent AI chip startup known for its wafer-scale engines, has filed for an IPO, signaling a major move in the competitive AI hardware market.
Railway has raised $100 million in funding to develop AI-native cloud infrastructure, positioning itself as a direct challenger to established cloud providers like AWS.
Nous Research has launched NousCoder-14B, a new open-source coding model, offering a powerful alternative in the rapidly evolving landscape of AI-powered code generation.
This paper introduces DR^{3}-Eval, a new framework aimed at improving the realism and reproducibility of deep learning research evaluations, addressing a critical challenge in the field.
New research demonstrates that exploration and exploitation errors in language model agents can be quantitatively measured, providing a pathway for more robust and reliable AI agent development.
Sema Code proposes a novel architecture for AI coding agents that decouples their components into programmable and embeddable infrastructure, enhancing flexibility and integration.
ASGuard introduces an activation-scaling guard mechanism designed to effectively mitigate targeted jailbreaking attacks against AI models, enhancing their security and robustness.
Reports indicate a potential warming of relations between AI leader Anthropic and the Trump administration, suggesting evolving dynamics in AI policy and industry engagement.
Google has rolled out new features for the Gemini API, offering developers more flexibility to balance cost efficiency with reliability for their AI applications.
OpenAI is actively working to accelerate the cyber defense ecosystem, focusing on initiatives that bolster collective security against evolving digital threats.
This research explores a new paradigm for reinforcement learning by investigating the transition from conditional probability P(y|x) to unconditional probability P(y) within pre-trained model spaces.
A new teacher-student cooperation framework is proposed for fine-tuning reasoning models, enabling the synthesis of student-consistent SFT data to improve model performance.