This Study by UC Berkeley and Tel Aviv University Enhances Task Adaptability in Computer Vision Models Using Internal Network Task Vectors