What is it about Kimi K3?

For anyone in an AI news blackout, Kimi is a new foundation model from a Chinese AI company called Moonshot. What is the big deal? It is a free, open-weight model launched last week. It seems to rival the intelligence of models from Anthropic or OpenAI. Why pay for Anthropic when Kimi is free?

Any digital application—including the core LLM—can "see and hear" a lot. In fact, context or memory is one of the selling points of the more sophisticated models. Context and memory might include prompts, preferences, data including images, keystrokes, and more. Most consumers accept the trade-off of their privacy for free services. Enterprises shouldn't. The government can't.

The US government has concerns about data collection from US citizens (e.g., location, biometrics, browsing habits) as well as infrastructure such as telecom. It is not uncommon for software-based products such as a web browser, mobile app, car, or infrastructure to collect data. It is often under the guise of "essential to operation"—and some of it is. Much of it is also used for advertising. A small amount feeds social media algorithms designed to drive dissent or influence opinion. Probably the easiest analogy is TikTok. (You may remember that the Biden administration wanted to disconnect this flow of data to China.) Much of the "listening" could be innocuous—just noise. It could also impact security, essential intellectual property, or operation of core services. In the worst-case scenario, you have the September 2024 incident with exploding pagers, which involved supply chain tampering.