Making Language Models Aligned with Human Intentions and Values

What it is about

You learn the foundations and current concepts of AI alignment research for large language models. You engage with model architectures, existing methods, metrics and societal challenges, and you gain practical experience in project design and Python implementation to design and implement strategies to align language models with human intentions and values.

What you will be able to do

• Understanding the development processes and structures of classical and modern language models
• Knowledge of central theories and relevant works on AI alignment of large language models
• Ability to design and implement strategies to constrain or evaluate language models taking human intentions and values into account

What the module consists of

• SeminarIntroductory session, participant presentations on specific alignment topics and subsequent discussions
• ProjektarbeitGroup work to develop and implement a project to mitigate misalignments of language models

Teaching method

• Teilnehmerpräsentationen mit DiskussionDeepening individual alignment topics and promoting critical discussion
• Gruppenprojekte mit Python-ImplementierungPractical application, project design and implementation of alignment strategies