Intelligence that can perceive, reason & act.

A vision I believe in must answer three questions: Why do this? Where should the best resources go? Which opportunities are worth passing up?

Fine words are not enough. When short-term gains conflict with long-term goals, choices are what count. What I truly believe should be visible in my decisions.

I value fair returns, but I do not aim to capture every benefit. I do not need to take every reward or own every part of the work.

Before securing my own share, I care more about whether cooperation and sharing can improve our chances of achieving something significant. Restraint does not mean a lack of ambition; it means directing ambition toward what matters more.

Something worth doing is not necessarily ours to do. The fact that everyone else is doing it is not a reason for us to follow.

I care less about appearing to cover everything than about truly solving the key problems. Intermediate wins are worth pursuing, but they must not replace the ultimate goal. Saying no to good opportunities outside our focus is also a responsibility.

My responsibility is not to make every judgment for everyone. It is to make our shared direction clear and build trust and collaboration.

I want to create an environment where excellent people have the confidence to exercise judgment, the willingness to collaborate, and the opportunity to grow. Being able to keep moving forward together through setbacks matters more than a momentarily impressive roster.

What we have promised deserves serious follow-through. Directions that have not yet been proven also deserve a chance to be tested.

I do not accept using “research” to evade delivery, or filling every hour of everyone’s schedule to create a sense of managerial security. Well-defined work needs efficient execution; uncertain exploration needs time, resources, and patience.

I care both about the limits of intelligence and whether it can actually be used within real constraints on compute, cost, and latency.

Enabling the same resources to support stronger models, and stronger models to serve more people, is a core problem worth investing in. Algorithmic breakthroughs and engineering efficiency belong together: efficiency is not only about savings, but about making previously impossible things feasible.

I care not only about what a model can do this time, but whether it can learn from interaction and feedback to do better next time.

I see continual learning as a core direction for long-term investment. This applies to models, to myself, and to the team: we should not simply repeat work, but let every experience improve the next judgment.

AI should be more than the product of our research and development. It should become part of our ability to do that work.

I want each generation of models to help us find problems, test ideas, and improve systems faster. Progress should be measured not only by how capable this version is, but also by whether it brings the next breakthrough closer.