AI has fundamentally changed the startup playbook; tokens can replace a 100-person engineering team.
Build startups around new AI capabilities, not obvious existing ideas.
Scale creates emergent properties that are impossible to predict from small-scale experiments.
Push systems beyond conventional limits; interesting behaviors appear at scale.
Break scaling challenges into individual engineering problems.
Leadership at scale requires a clear mission, plan, and decision framework.
Humans consistently underestimate exponential growth.
ChatGPT succeeded by following unexpected user behavior (people wanted to chat).
"See what users love and do more of it."
Rapid growth while the product is imperfect is a strong signal of product-market fit.
Coding is AI's first major enterprise application; robotics is the next physical-world interface.
Current AI pipeline (pretrain → post-train → RL) will likely be reinvented by AI itself.
Intelligence will become a utility like electricity or the internet.
Users care about outcomes, not the underlying hardware.
Inference infrastructure is one of the biggest under-invested opportunities.
LLMs are far from a dead end; scaling continues to unlock new capabilities.
Don't let identity override empirical evidence.
Education must shift from memorization to AI-assisted thinking and problem-solving.
Learning to think remains valuable even if AI performs the task better.
Broad, interdisciplinary learning creates better founders.
AI progress is likely to continue exponentially over the next few years.
Biggest societal risk: AI power concentrated in a few companies.
AI should be broadly democratized rather than centralized.
Compute will become a critical utility and remain supply-constrained.
Demand for AI inference is effectively uncapped as costs decrease.
Long-term wealth distribution should favor ownership (citizen wealth funds) over fixed UBI.
Personal AI agents running continuously are the future interface to AI.