The 3-minute briefing
science, tech &
markets.

NVIDIA has deployed a 753B parameter mixture-of-experts model with 1M context window, quantized to NVFP4 format for Blackwell hardware while maintaining near-FP8 accuracy. This enables efficient deployment of massive language models on next-gen infrastructure with minimal performance trade-off.

A viral X post claims CNBC has signaled weakening AI demand, suggesting SpaceX, Anthropic, and OpenAI are accelerating public offerings before market sentiment shifts. The framing implies late-stage hype dynamics in AI funding.

Emerging best-in-class models optimized for 8GB to 512GB memory ranges, signaling progress in efficient model scaling. Developer community pushing GLM-5.2 improvements suggests active competition in accessible AI.

Eli Lilly's acquisition of Centessa for $6B signals a major push into sleep optimization beyond narcolepsy treatment, potentially unlocking a consumer-grade enhancement drug that could reshape productivity and longevity markets. If successful, this could rival Ozempic's cultural impact by targeting the universal need for better sleep.

Anthropic claims Alibaba deployed ~25,000 fraudulent accounts to systematically extract Claude's capabilities, raising critical questions about AI model security and competitive espionage. This escalates tensions between major AI players and highlights vulnerabilities in API-based model access controls.