Google at SALA 2026

This session will cover the history and meaning of "vibe coding," and, most importantly, show you how to use it to make your coding workflow faster and more creative. Learn practical techniques for jump-starting new ideas, a skill that will be particularly useful for those participating in the afternoon hackathon.

Speaker: Pablo Samuel Castro

Diffusion models are advanced AI tools that create realistic new content, like images and video. This talk will demonstrate that these powerful models are surprisingly effective for a wide range of visual tasks, not just content creation.

Abstract: Diffusion models have emerged as a powerful class of likelihood-based generative models. They have been particularly effective for modeling image and video data, with their striking performance in text-to-image and text-to-video generation. This talk will address several questions concerning model capabilities on a variety of downstream tasks, beyond tex-to-X generation. These include generative data augmentation, fine-tuning for visual estimation, representation learning and alignment with human perception, and zero-shot inference with video models. Together, these results suggest that large text-to-image and text-to-video models are surprisingly effective for a broad spectrum of visual tasks, through a combination of fine-tuning, in-context learning, or zero-shot application.

Speaker: David Fleet