Frontier

Superintelligence, on this site, means systems that can learn, plan, and act across many domains well past a single trained task, including in the physical world, with less need for a person to specify every step. It is a direction, not a badge we pin on a launch.

Chat that is better at answers is one movement along that direction. A model that draws a working interface is another. A world model that can train a policy for a new body is another. A gene finder that writes candidates into a public database is a powerful specialist, and we will not stretch the word until the system is doing open-ended scientific work rather than a defined prediction.

We will not run countdown clocks. We will not use extinction rhetoric as a hook. We will not call every capability jump superintelligence because the syllable tests well. When we use the word, the item has to say which capability crossed a line: broader action, a learned world, a person leaving the loop, or an evaluation that no longer matches the system.

This page exists so the briefing can point here instead of redefining the word in every issue. It is not one of the primary shelves.