How do AI systems discover linguistic patterns, infer rules from a few examples and generalise across languages? This workshop brings together researchers in artificial intelligence, linguistics and cognitive science to investigate the capabilities and internal representations of language models. Closing a large SNF grant, it features Blackbird Language Matrices: controlled linguistic puzzles designed to test abstraction and generalisation. The programme connects BLM research with broader work on reasoning, analogy and multilingual modelling, analyses of models’ internal mechanisms, using human cognition as a point of comparison. Its central objective is to understand how language models achieve their successes, where their limitations lie and how these insights can inform the development of AI.