DeepSeek Drops V4 Preview, Beats OpenAI on a Budget
Audio in Mandarin Chinese · English transcript below
⚡ DeepSeek's bargain compute jolts White House; Google rescues async training, GPT-5.5 & Intel earnings soar
Today's Top 3 Headlines
- Technology
DeepSeek Drops V4 Preview: China’s Newest LLM Emerges
DeepSeek unveils V4 preview, an open-source LLM it says matches GPT-4o at **<1/10 the compute cost**, directly targeting OpenAI, Anthropic, Google.
Source ↗ - Technology
DeepSeek V4: Open-Source, Ultra-Cheap LLM Takes on OpenAI & Google
DeepSeek’s V4-Pro & V4-Flash open-source LLMs undercut U.S. rivals, listing at **<20%** of OpenAI/Anthropic equivalents, extending the 2025 shockwave sparked by its low-cost V3.
Source ↗ - Policy
White House Accuses China of Stealing U.S. AI Secrets
White House officials Thursday accused Chinese actors of **mass-scale theft** of U.S. AI tech and vowed retaliation. In February, Anthropic charged **DeepSeek, Moonshot AI and MiniMax** with illicitly siphoning its Claude chatbot; OpenAI also told lawmakers DeepSeek used distillation to **freely harvest** frontier-lab capabilities.
Source ↗
+6 more headlines
- Google DeepMind Unveils Decoupled DiLoCo: Breakthrough Asynchronous LLM Training Cuts Compute Costs
- OpenAI Unveils GPT-5.5 Powered by NVIDIA
- AI Boom Drives Intel to $13.6B Q1 Revenue
- U.S. Soldier Arrested for Insider Trading on Alleged Maduro Capture Plot
- Tang Wenbin’s Pivot: From Megvii to LLM-Powered Embodied AI—No Humanoids, Just Real-World Agents
- Anthropic Acknowledges Claude Code Quality Complaints