Exclusive Interview with Alexandr Wang, Founder of Scale AI: From Data Annotation Giant to Head of Meta's Labs, Discussing Entrepreneurial Comebacks and Technological Evolution in the AI Era

"Alexandr Wang: 'This is a Once-in-a-Civilization Opportunity'" (Y Combinator Interview), the following is a summary of the core content. 1. Early Entrepreneurial Experience and the First Principles of Scale AI • Dropping out of MIT to transition to YC: Alexandr founded Scale AI at the age of 19. The initial project applied for at YC was an AI Agent for healthcare, which was abandoned due to the immaturity of the market (several years too early) and shifted focus to the data field. • Discovering an "unmet pain point": While training models at MIT, he found that acquiring computing power (GCP) and training code was easy, but obtaining high-quality training datasets (Data) was extremely difficult. • Adhering to non-consensus reverse thinking: In the early years of Scale, the data industry was very "unsexy," facing many rejections and doubts from VCs. But he firmly believed in the first principle: the more widespread the models, the greater the demand for data. Entrepreneurs must establish an internal compass that is not swayed by external noise and quietly cultivate in underappreciated fields. 2. Shifts in Entrepreneurial Paradigms: "Goliath vs. Goliath" • From David vs. Goliath to Mecha-Goliath: • In the past, entrepreneurship was about "David challenging Goliath," where startups competed with giants through agility and unique entry points. • With the support of AI and agentic tools, today’s startups have gained extremely powerful "mecha giant" equipment, transforming into "Goliath vs. Goliath," where individuals or small teams can possess super productivity to directly compete with traditional large companies. • The bottleneck has shifted from "intelligence" to "vision and ambition": As model capabilities and intelligence (Intelligence/Agency) become increasingly abundant, the future scarce resource will no longer be manpower or coding ability, but whether founders have a clear and ambitious vision for the world in the next 5-10 years. 3. Steering Meta's Cutting-Edge AI Labs and Open Source Philosophy • Rebuilding Meta Labs: During his approximately one year at Meta, he conducted a zero-based build, highly focusing on talent density. Scientific research work is fundamentally different from traditional internet product development, requiring a heavy reliance on experimentation, scientific breakthroughs, and scalable expansion. • Launching the MuseSpark series of models: • Released MuseSpark 1, Muse Image, and version 1.1, with performance comparable to Opus-level models but at nearly 8 times lower cost. • A new Harness (agent framework/control loop) is about to be launched, aimed at supporting more complex large-scale multi-agent collaboration and orchestration. • Upholding a distributed and open-source ecosystem: Firmly opposing the centralized/authoritarian model of AI controlled by a single company. Believing that through open-source and empowering billions of individuals and businesses worldwide (like the 200 million businesses on the Meta platform), everyone can have personalized "Personal Superintelligence." 4. Underlying Capabilities and System Thinking in the AI Era • "Systems Thinking" is timeless: • Although today’s developers no longer write every line of code from the ground up, the abstraction level has risen to "how to orchestrate agents" and "how to enable thousands of agents to collaborate efficiently." • Rigorous systematic thinking and architectural design capabilities remain core; one should not blindly abandon spatial/structural rotation abilities (Shape Rotating). • Seeking agentic feedback loops: • The greatest business alpha (excess returns) lies in finding micro-feedback loops within enterprises, defining clear eval metrics, and using a swarm of agents to continuously optimize. • Within Meta, as long as there are correct agentic loops and evaluation metrics, the output achieved by a swarm of agents can easily surpass that of a team of 100 engineers. 5. Advice for Young Entrepreneurs • Identify the steepest, longest exponential curves: Decades ago it was Moore's Law, today it is the evolution of AI technology. Even if the initial starting point seems very mundane (like early cat recognition in YouTube videos), as long as it is on an astonishing exponential growth trajectory, it is worth investing all passion. • Maintain firm belief: The biggest challenge for young entrepreneurs is the lack of experience, making them easily overwhelmed by the chaotic market noise and doubts around them. They must continuously hone their self-judgment, believe in, and embrace this "once-in-a-lifetime" civilization-level era dividend.