KAIST XAI Tutorial 2025 | On the Biology of a Large Language Model | Artyom Stitsyuk (SAIL, KAIST)

Nowadays, Large Language Models (LLMs) demonstrate impressive capabilities. However, the underlying processes and mechanisms of those abilities remain largely unclear. As LLM systems become more intelligent, capable, and are used in an expanding range of applications, their black-box nature is becoming increasingly unsatisfactory. Our goal is to reverse engineer how these models function internally so that we can understand and explain them more deeply and evaluate how well they fit their intended purposes. 오늘날 대규모 언어모델(LLM)은 인상적인 역량을 보여주고 있다. 그러나 이러한 능력의 기반이 되는 과정과 메커니즘은 여전히 대부분 명확하지 않다. LLM…