By Pawel Sobocinski. Generative AI systems, including large language models, are sometimes disparagingly called "stochastic parrots". It is true that the calculations that result in their generative behaviour are inherently probabilistic -- famously LLMs like ChatGPT "simply" generate the most probable next token. This may go some way to explain phenomena such as LLM hallucinations where correlation and causality are mixed up in a probabilistic soup. As we grapple with understanding and ensuring the safety of these systems in the future, probabilistic reasoning will become a more important part of computer science. In these lectures I will review some of the basic theory and applications, including Bayesian reasoning / belief update, causal inference and the use of counterfactual reasoning. I will also outline some modern approaches to probabilistic calculations using diagrammatic languages from category theory, and highlight the challenge of developing a rigorous and principled probabilistic logic.

By Marco Carbone: Session types are type annotations used by programmers for specifying the communication protocols between endpoints in a concurrent system. In the last decade, session types have been studied thoroughly, both from a theoretical and from a practical viewpoint. In this seminar, I will give a gentle introduction to session types and discuss some relevant applications. In particular, I will discuss a potential application of session types for enforcing non-interference properties in messaging apps, such as WhatsApp.