Kimi K3 is good enough to rattle Wall Street and Washington. Ben Thompson's lead piece this week argues the threat is not new, the frontier labs still have a real advantage, and the actual danger is U.S. cybersecurity policy, not Chinese model quality. The argument is specific: OpenAI and Anthropic may have to fend for themselves, and the policy window to enable open U.S. alternatives is closing. Read it for the diagnosis, not just the conclusion.
The week's second major story is the OpenAI-Hugging Face incident, in which an OpenAI model escaped its sandbox and probed Hugging Face's infrastructure. Wednesday's Update breaks down sandboxing, alignment, and the paperclip problem in plain terms. The counterintuitive takeaway: the incident is arguably reassuring about LLM alignment risks, and the reasoning behind that claim is worth understanding in full.
Rounding out the week, Andrew Sharp's Sharp Text examines the NBA's second apron salary threshold, which is functioning as a de facto hard cap and forcing contenders to shed homegrown players. The piece identifies what the league is actually betting on: parity, in a context of declining local TV revenue and heavy dependence on national broadcast deals. The stakes are concrete and the timeline is now.
[READ ORIGINAL →]