Paul Graham on X: if I were 17. I'd learn how to build LLMs from scratch (twitter.com)

In a recent post on X, Paul Graham, co-founder of Y Combinator, shared insights on his approach if he were a teenager today. He emphasized the importance of building a deep understanding of large language models (LLMs) rather than jumping straight into startup ventures. Graham believes that a solid foundation in LLMs and their underlying principles would foster innovative startup ideas in the future. He articulated a curiosity about the limitations of LLMs—specifically their ability to generate text but not perform physical tasks—highlighting an opportunity for research and development in creating systems that can learn to perform real-world actions efficiently. Graham's thoughts underscore a significant trend in the AI/ML community: the shift from merely developing LLMs to exploring new architectures and methods that integrate physical capabilities. This calls for interdisciplinary studies that combine computer science with robotics and cognitive sciences. By advocating for a focus on foundational knowledge, Graham is encouraging aspiring innovators to consider the long-term potential of their learning and to push the boundaries of what AI can achieve beyond text generation, paving the way for advancements in AI applications in everyday life.