Paul Graham, the founder of Y Combinator, outlined what he would do if he were 17 years old in a post on X, emphasizing deep technical knowledge over immediate startup creation.
Graham said he would focus on learning how to build large language models from scratch and train them as powerfully as possible using available hardware. Rather than launching a startup immediately, he argued that building a strong foundation in LLM technology would lead to better startup ideas later.
“What I would not do is try to start a startup,” Graham wrote. “Instead I’d build the foundation of knowledge to base a startup on later. Way better startup ideas would grow out of understanding LLMs deeply than trying to start companies based on what I knew at 17.”
Graham also identified a specific research problem he would investigate: understanding the gap between what LLMs can do and what they cannot. “I would try to figure out why LLMs can write my essays but not clean my bedroom,” he wrote, highlighting the difference between LLMs’ capability in language tasks versus physical-world tasks.
Based on this research interest, Graham suggested he would pursue formal education to develop solutions. “Then I would study topics in college and grad school that could help solve that problem,” he said, indicating a multi-year commitment to understanding the underlying limitations of current AI systems.
Graham’s comments reflect a broader perspective that technical depth and research understanding create stronger foundations for entrepreneurial success than rushing to launch companies. His emphasis on studying why LLMs cannot perform physical tasks suggests he sees significant unsolved problems in AI that could drive future innovation and company creation.
Key facts
- Graham would prioritize learning to build LLMs from scratch over starting a startup at 17
- He would focus on training LLMs as powerfully as possible with available hardware
- Graham identifies a specific research gap: why LLMs can write essays but not clean bedrooms
- He believes deep LLM understanding leads to better startup ideas than immediate entrepreneurship
- Graham suggests studying in college and grad school to solve limitations he identifies