Paul Graham: If I Were 17, I Would Learn to Build Large Models from Scratch, Not Rush to Start a Business
Paul Graham, co-founder of Y Combinator, responded to the question "What would you do if you were 17 now" by stating that he would start from scratch to learn how to build large language models and train powerful models using all available hardware.
He explicitly added that he would not rush to try entrepreneurship. Instead, he would establish a solid knowledge base to prepare for future entrepreneurship. Ideas for starting a business that naturally emerge after a deep understanding of LLMs are of much higher quality than those forced out based on a 17-year-old's perspective.
This statement contrasts with YC's usual encouragement for young people to start businesses early, reflecting the reality that the shelf life of AI product ideas is shortening and model capabilities are rapidly iterating.
Young talents should prioritize approaching the technological frontier rather than rushing to commercialize; funding and attention should flow towards deep technical understanding and experimentation. The beneficiaries are young researchers who solidly grasp model principles, while those relying on short-lived product shells face pressure. The event-driven AI era is shifting from "finding ideas to start a business" to "understanding technology first and then generating ideas."
Source: Public Information
ABAB AI Insight
Paul Graham has long promoted young people to start businesses early through YC, emphasizing rapid validation and iteration in the capital path. However, in the face of the LLM wave, he has adjusted his priorities to focus first on deepening technical understanding.
Resources are shifting from brainstorming entrepreneurial ideas to building and training models from scratch, motivated by the rapidly moving technological frontier. The opportunities that arise from truly understanding the technology far outweigh preset business plans.
Drawing a parallel to the early internet path of "learning programming before starting a business," the current AI landscape is transitioning from applying superficial layers to understanding foundational model capabilities. The industry as a whole is shifting from a competition of product ideas to a focus on deep technical accumulation.
This essentially reflects a timing judgment under accelerated technological substitution. The greatest advantage of youth is having the time to delve deeply into new paradigms, with the mechanism being that the continuous leap in model capabilities creates new possibilities. Early entrants are more likely to discover opportunities that were "impossible yesterday but suddenly feasible today."
ABAB News · Cognitive Laws
Understand technology first, then generate ideas
The advantage of youth is having time to delve into new waves
The movement of model boundaries itself is a source of opportunity