New AI Method Skims and Skips to Cut Search Costs by Two-Thirds

When you search Google Photos for a red balloon or ask an AI assistant to find a specific scene in a video, you're using a smart search system. These systems, called multimodal retrieval, understand both images and words. But they're powered by huge AI models that need to process every single pixel and word equally—which takes a lot of computer power and energy.

A team of researchers noticed this is wasteful. Some parts of a picture or sentence barely matter for finding what you want, and some search questions are simple enough to answer quickly. So they built a new framework called 'Skim and Skip.' The name says it all. First, the AI skims through content and keeps only the important parts—like focusing on a red balloon and ignoring the background. Then it checks whether it already has a reliable answer. If yes, it skips the rest of the complicated calculations.

The results are impressive. Across 12 standard search benchmarks, this approach kept about 99% of the accuracy of the old, heavy method. At the same time, it performed the math up to 66% less—translating to a 1.64 times speed boost in real use. Imagine reading a whole book versus skimming a few key chapters and still getting the gist perfectly. That's the idea.

Why does this matter to you? Faster, cheaper AI means search tools respond quicker on your phone, and companies spend less money running massive AI systems. Energy use drops, too. In the future, when an AI searches through millions of images or documents, it may need far fewer resources—making those services better and perhaps cheaper for everyone.