Today's signal in AI: funding rounds, safety research, new benchmarks, and the infrastructure arms race. Click any card for the full brief.
Meta's CEO bets half a billion on AI-driven biology research, aiming to accelerate drug discovery and protein folding breakthroughs.
Read full brief →SAP, Capgemini, Nemetschek, Hexagon and other European software leaders announce major AI integration strategies across their product lines.
Read full brief →LinkedIn co-founder argues that AI-powered second opinions will become standard medical practice, improving diagnostic accuracy and patient outcomes.
Read full brief →Weekly roundup covering OpenAI's image generation upgrade, Alibaba's Qwen 3.6 Max release, and Moonshot AI's Kimi-K2.6 model launch.
Read full brief →Startup Eka trains its robotic manipulation system using a novel vision-force-action model, achieving dexterous grasping in unstructured environments.
Read full brief →New benchmark tests Claude's bioinformatics reasoning skills against expert-level problems in genomics, protein analysis, and molecular biology.
Read full brief →Hyundai launches next-gen infotainment system with Gleo AI assistant, bringing conversational AI and predictive features to the dashboard.
Read full brief →OpenAI signs contracts for 10 gigawatts of US AI compute capacity, with 3GW already online — a massive infrastructure buildout for next-gen models.
Read full brief →Internal 2017 emails show Elon Musk sought majority voting control over OpenAI, shedding new light on the ongoing legal battle between the founders.
Read full brief →The FDA launches a pilot program using AI and cloud computing to give the agency a 'digital twin' for faster drug approval and safety monitoring.
Read full brief →New research analyses compact language models for biomedical named entity recognition, finding that smaller models can rival large ones with fine-tuning.
Read full brief →Researchers demonstrate that feeding prompts one word at a time can bypass LLM safety guardrails, exposing a fundamental vulnerability in alignment training.
Read full brief →Study tests 115 language models for signs of conscious-like behavior, measuring 'trained denial' patterns when models discuss their own awareness.
Read full brief →New math reasoning benchmark covers European and Brazilian Portuguese variants, enabling more inclusive evaluation of multilingual LLM capabilities.
Read full brief →Research shows general-purpose LLMs can effectively extract structured data from electricity invoices, reducing manual processing in the energy sector.
Read full brief →Provocative study argues that LLM outputs fundamentally produce kitsch — aesthetically pleasing but emotionally shallow content that mimics genuine creativity.
Read full brief →