Gaining T(h)rust with AI

In this talk, I will elaborate on some recent developments in generative AI. We start by refreshing large language models (e.g., ChatGPT, BERT), the underlying transformer architecture (encoder/decoder setup) and attention mechanism. We discuss the economic, job and education impact of these technologies. Next, we zoom in on AGI/Singularity and review key criteria for trustworthy AI such as open source, education, transparency, regulation, and explainability, . We conclude the talk by listing key challenges including untapping new sources of data, cause and effect reasoning, multimodal AI, neuro-symbolic AI and AI systems capturing a world model. Throughout the talk, the presenter will also refer to some of his research on the topic.