Paul Graham Recommends Building LLMs From Scratch at 17
A researcher replies that hands-on building beats textbook study of models.
Paul Graham posted that if he were 17 today he would learn to build large language models from scratch and then train them using whatever hardware he could obtain. Researcher Pasquale Minervini replied that the approach feels like dissecting animals rather than reading a biology textbook. The posts show Graham responding to a question about his path at that age and Minervini offering the direct analogy in reply. Graham is identified as the co-founder of Y Combinator. Minervini is identified as a lecturer in NLP and ML at the University of Edinburgh. The exchange addresses methods for gaining skills in constructing and training such models.
Someone asked what I'd do if I were 17. I'd learn how to build LLMs from scratch, and then train ones as powerful as I could with whatever hardware I could get access to.
Combined views
Paul Graham Recommends Building LLMs From Scratch at 17
A researcher replies that hands-on building beats textbook study of models.
Paul Graham posted that if he were 17 today he would learn to build large language models from scratch and then train them using whatever hardware he could obtain. Researcher Pasquale Minervini replied that the approach feels like dissecting animals rather than reading a biology textbook. The posts show Graham responding to a question about his path at that age and Minervini offering the direct analogy in reply. Graham is identified as the co-founder of Y Combinator. Minervini is identified as a lecturer in NLP and ML at the University of Edinburgh. The exchange addresses methods for gaining skills in constructing and training such models.
Reactions from ranked influencers
2 posts@paulg it feels like dissecting animals vs just reading any biology textbook
I would try to figure out why LLMs can write my essays but not clean my bedroom. Then I would study topics in college and grad school that could help solve that problem. I'll figure out a set of methods and architectures beyond LLMs that can quickly learn to perform physical…
