Susan Zhang Reframed LLM Theory as Data Mines

Susan Zhang, a research engineer at Google DeepMind who previously worked at Meta AI and OpenAI on systems like OPT-175B, posted on X that the so-called theory of LLMs amounts to examining the shape of data mines and combinatorial compute mines. She described scenarios where language barriers vanish, prior literature becomes instantly available without specialized retrieval skills, and existing knowledge breaks down under new access patterns.