Agenda
50 years of AI — from expert systems to self-supervised learning
Five failure modes of misaligned models
Whose values? — annotators, company goals, the diversity problem
How we align — labels, rules, demonstrations, explanations, preferences
Technical deep dive — PPO, GRPO, DPO
Why post-training may not be enough
Future directions — pre-training alignment
Q&A
Our Speaker
Tomasz Adamczyk — Senior AI Engineer at Addepto
Tomasz Adamczyk is a Senior AI Engineer at Addepto with over 5 years of experience, working across classical machine learning and large language models. He is also a PhD researcher in AI and cognitive science at the ImpactAI Lab at Wrocław University of Science and Technology. His research investigates how large language models interact with human cognitive processes, with a particular focus on theory of mind, cognitive biases, and the influence of AI on human reasoning and decision-making.