Speaker
Description
The field of AI research finds itself at an interesting decision point: to exploit promising solutions that rose to prominence over the past few years and scale them up to more and more powerful generalist models, or to explore new unknown areas? Do we already have all the basic ingredients we need? This talk will describe some of the currently most popular architectures (Transformers, Perceivers, Graph Neural Nets, Diffusion, etc.), discuss the multimodal frontier, and highlight some success stories in technical application domains such as code generation, algorithmic reasoning, and the physical science, and explore results and open questions in the fundamental understanding of these models (e.g. scaling laws, interpretability, compression). This overview aims to provide an introduction to exciting topics in AI research that may be relevant to this community, under the lens of a data-driven scaling approach.
