LLMs are primarily trained to meet explicit user demands, rather than the implicit needs of learners. Though researchers, industry, and the wider EdTech community have been seeking to adapt LLMs for educational applications, given its affordance for flexibility. Most approaches to instilling pedagogical qualities in LLMs, can be broadly separated into four main categories: * Prompting * Modifying model parameters * Fetching external data to ground model context * Using external state-based systems --- TODO: * Break into sub-categories, specific techniques, and associated works