Imagine an AI that can understand surgical procedures just by watching videos of lectures. That’s the exciting promise of new research into surgical video-language pretraining (VLP). One of the biggest hurdles in training AI with surgical videos is the lack of large, labeled datasets. Unlike general images and videos readily available online, surgical footage is sensitive and requires expert annotation. This research tackles this challenge by turning to a readily available source: online surgical lecture videos. Researchers have developed a clever technique called PeskaVLP, or Procedure-Encoded Surgical Knowledge-Augmented Video-Language Pretraining. It uses these lecture videos, along with their transcripts and metadata, to train an AI model that can link visual and textual information within surgical contexts. One ingenious aspect of this work is how it enhances the often incomplete or noisy transcripts from these videos. By leveraging the power of large language models (LLMs) like GPT, researchers were able to clean up and expand the text descriptions of surgical steps, creating a much richer and more consistent learning experience for the AI. This is like giving the AI a surgical textbook alongside the video lectures. PeskaVLP goes beyond simply matching video clips to text. It’s designed to grasp the inherent order and structure of surgical procedures, understanding the temporal relationships between steps. Think of it as learning the “grammar” of surgery, where actions occur in a specific, logical sequence. This is crucial for any AI that hopes to assist or even anticipate actions in a real-world surgical setting. This work is a major step forward in surgical AI, showing how innovative methods can overcome data limitations and train powerful models that understand the complexities of surgical procedures. By combining video, text, and advanced AI techniques, this research opens doors to a future of enhanced surgical training, better assistance systems, and ultimately, improved patient care.