OpenAI is launching a fellowship program to support independent researchers working on the safety and alignment of AI systems.
AI: Events
OpenAI Proposes an Industrial Policy for the AI Era – What It Means for People
Regulation
OpenAI has published proposals for an industrial policy in the AI era: empowering people, ensuring fair distribution of benefits, and building resilient institutions.
Research proves that when making reasonable bets on independent events, the choice of what to bet on depends solely on the ratio of odds, not on your personal risk tolerance.
Lab
Neural Networks Under Control: How Physics Teaches Algorithms to Behave Predictably
Electrical Engineering & System Sciences
Researchers have proposed a method for training neural networks for complex technical systems, where the model not only fits parameters to data but also guarantees stability.
Researchers have discovered that language models can optimize database query execution, doing so far more effectively than traditional statistical methods.
AI: Events
HiClaw Joins AgentScope: Alibaba Builds a Unified Platform for Multi-Agent Systems
Technical context • Infrastructure
Alibaba has merged two of its open-source projects, HiClaw and CoPaw, under the AgentScope framework to collaboratively develop infrastructure for multi-agent systems.
Google has released Gemma 4 – the company's most powerful open models to date, focused on complex reasoning and agentic scenarios.
Cursor has released the third version of its development environment – now it's a unified workspace where agents participate in code creation alongside humans.
AI: Events
Stop Teaching Everything at Once: Why AI Models Perform Better When Trained for Specific Tasks
Research
Researchers suggest rethinking the approach to training AI models, advocating for specialization on specific tasks over multitasking.