@jiahui_k_chen
PhD @UTAustin 🐂 | Visiting Researcher @Meta AI
AI
The author says human-LLM comparisons, generation experiments and mechanistic interpretability methods show how one word shapes a conversation.