Import AI 479: Spacewar; low-latency tasks versus LLMs; Genie 4

Researchers from multiple universities and NVIDIA developed Realtime-Bench, a benchmark testing AI models' ability to perform low-latency, real-time tasks across games and web activities. The benchmark evaluates models on temporal awareness, adaptive deliberation, and timed action selection, revealing current LLMs struggle with such tasks. Models from OpenAI, Anthropic, Google, and others were tested, highlighting challenges in real-time decision-making for AI.