Research within TEXT is conducted across four thematic work packages, each concerned with a different aspect of “LLM-hard” problems – that is, tasks and applications that LLMs are not designed for, and fundamentally struggle to perform at a human level.