AI Search Just Got Faster and Cheaper — Here’s How

Imagine asking your AI assistant to find all photos of your dog in the park. Normally, the AI would painstakingly examine every pixel in every image — like reading a book word by word. A new technique called RaDiCal acts like a speed-reader: it skips over the fluff (like the background grass) and focuses on what really matters (your dog).

The trick isn’t just about discarding extra visual data. It’s about knowing when the AI is paying attention to the right clues. The researchers found that just because a part of an image stands out doesn’t mean it helps rank the results. So instead of trusting “flashy” pixels, RaDiCal checks whether the AI is actually using those pixels to tell photos apart. If it’s not, RaDiCal quietly removes them early, saving time and computing power.

The result? Faster searches, lower server costs, and no need to retrain the AI for each new task. In tests across three image datasets, RaDiCal matched or beat the slower, full-image approach while using up to 45% fewer computing steps. It’s like giving your phone a supercharged filter that only processes what truly matters — your dog, not the trees.

This isn’t just for pet photos. Any AI system that sorts through lots of images — from online shopping to medical scans — could run smoother and cheaper. And because it doesn’t require new data or expensive tweaks, it could roll out widely without waiting years for upgrades.