Manifesto

Superintelligence should become a capability that everyone can draw on in everyday life.

Realizing that future requires solving a fundamental problem: human attention is limited.

An experienced user might direct eight AI agents across eight windows, supplying context, reviewing results, and coordinating their work. That workflow reveals extraordinary possibilities. It also reveals how much knowledge and attention accessing those possibilities can require.

We believe AI should assume more of the work of understanding and adapting to the person it serves.

At General Interface Labs, we are building a general interface between people and increasingly capable intelligence: a persistent, adaptive medium that works across different goals, situations, and levels of expertise.

Our approach rests on three principles.

Real-time interaction. Intelligence should participate as things unfold. People should be able to speak, point, sketch, interrupt, and act directly. The system should maintain shared context across these interactions and respond at the pace of the activity.

Personalized intelligence. AI should learn continuously from experience with each individual. It should develop an understanding of their goals, knowledge, preferences, and circumstances—and revise that understanding as they change. Shared experience should make future assistance more useful and reduce the need for repeated instruction.

Private by design. Personal intelligence depends on personal context. People should control what their systems remember, how that information is used, and what is shared. Privacy and individual control are foundations for the relationship we want to build.

Research.

Our research focuses on continual learning, persistent memory, and adaptive multimodal interaction. We want systems to improve both their understanding of each person and the strategies they use to provide assistance.

A longer-term research goal is recursive self-improvement: systems that can propose, test, and adopt improvements to their own learning and interaction methods. Progress here requires reliable evaluation of whether those changes actually help the people using them.

We are starting with learning.

Our first product is an AI tutor that teaches through a live, interactive whiteboard. It explains, draws, listens, and responds as students work. Their attempts provide evidence of what they understand, where they struggle, and which explanations help.

We are building toward a tutor that carries this understanding across lessons and becomes better at helping each student progress. We will measure its success through what students can understand, retain, and apply independently.

Learning is our first setting for a broader ambition: making advanced intelligence useful across the full range of human experience.

Reaching ordinary households requires affordability, accessibility across languages and devices, and systems worthy of people’s trust. These requirements will shape our research and products as we work toward bringing the benefits of superintelligence to everyone.