This invention describes a system that lets you search for images using everyday language, like "a cat sitting on a mat." You tell the system what objects you're looking for and how they relate to each other in space. The system then uses an advanced computer program to understand your request, turns it into a special code, and compares that code to codes it has for relationships between objects in its image collection. Finally, it shows you the images that best match your description, ranked by how well they fit.
Why it matters: Filed in 2018, this invention relies on neural language models to understand natural language queries and generate vectors. Since then, large language models (LLMs) have become vastly more powerful and accessible, making the core technology for understanding complex natural language queries and generating effective query vectors significantly easier to implement and more performant than it would have been eight years ago.
AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.
Reinvent this with AI