AI News Brief: today roundup
- Researchers introduced OptSkills, an agent system that clusters optimization problems by archetypes to build reusable workflow skills.
- Researchers developed OptimusKG, a multimodal biomedical knowledge graph unifying millions of relationships across molecular and clinical domains.
- A new behavioral framework measures LLM reasoning across six cognitive dimensions, going beyond simple final-answer correctness.
- Researchers created an SLM-based gateway that uses small language models to route user prompts in virtual worlds.
- OpenAI released GPT-6 Astra, a computer-use AI model featuring a 1.05-million-token context window and strict cybersecurity restrictions.
- Medical Heuristic Learning uses LLMs to generate transparent, auditable rule-based expert systems for clinical decision support.
- FormalEvolve uses neuro-symbolic evolutionary search and an archive to create diverse mathematical autoformalizations for theorem proving.
- OpenAI launched Daybreak for Frontline Defenders, pledging $1 billion in cybersecurity AI resources to protect essential services.
- UniToolCall introduces a unified framework for LLM agent tool use, standardizing training datasets, interaction patterns, and evaluations.
- The FireDucks Python library speeds up pandas data workflows up to 20x using lazy execution and compiler optimizations.
Sources
- OptSkills: Learning Generalizable Optimization Skills from Problem Archetypes via Cluster-Based Distillation
- Unifying biomedical knowledge in a modern multimodal graph
- Measuring Reasoning Quality in LLMs: A Multi-Dimensional Behavioral Framework
- From Prompt to Service: An SLM-Based Agent Orchestration Gateway for AI-Driven Virtual Worlds
- OpenAI Releases GPT-6 Astra: A 1.05M-Context Computer-Use Model Gated Behind a ‘Critical’ Cyber Threshold
- Medical Heuristic Learning: An LLM-Driven Framework for Interpretable and Auditable Clinical Decision Rules
- FormalEvolve: Neuro-Symbolic Evolutionary Search for Diverse Autoformalization
- Daybreak for Frontline Defenders: $1B to protect essential services
- UniToolCall: Unifying Tool-Use Representation, Data, and Evaluation for LLM Agents
- This Python Library Can Run Pandas Workloads Up to 20x Faster