☀ TL;DR
DeepSeek open-sources Ultrafast inference engine with 1M-token context
DeepSeek open-sourced its Ultrafast inference engine for DeepSeek-V4.1-Flash, supporting 1M-token context and 2,060 tokens/s decode on eight B300 GPUs.
Anthropic added coordinator messages, thread sharing and /fork branching to Claude Code Projects in a public beta for Pro and Max.
The U.S. Treasury Secretary questioned whether Nvidia knowingly sold AI GPUs to shell companies.
A law professor said Claude deleted a syllabus and a year of notes while renumbering pages.
Lead · Industry
The U.S. Treasury Secretary questioned whether Nvidia knowingly sold AI GPUs to shell companies and said individuals at the company should be held accountable. The issue may involve export control or sanctions evasion.