Google Patents a Training Method That Teaches AI When to Ignore Its Sources
AI models that answer questions from documents have a persistent problem: sometimes the document is useless, and the AI should just say…
The Big Tech patents worth knowing in ai/ml, in plain English. Updated weekly from new USPTO publications.
AI models that answer questions from documents have a persistent problem: sometimes the document is useless, and the AI should just say…
What if searching for a product photo and typing its name returned the exact same result — not just similar results, but…
Most image search works by matching pixels and colors — Microsoft's new patent wants to match meaning instead, understanding the relationships between…
Imagine pointing a camera at a server rack or a factory machine and seeing every part automatically labeled — pulled directly from…
Thermal cameras show you heat — but they can't tell you what's hot. Microsoft's new patent describes a system that figures out…
Anyone who's tried to swap text in a photo using an AI editor has seen it: the replaced area looks slightly off,…
When dozens of cameras watch the same scene from different angles, getting them all to agree on where something is turns out…
A stop sign in Tokyo looks nothing like one in Texas. Nvidia's latest patent tackles that exact problem — teaching an AI…
Most AI models look at data in one fixed level of detail — like always reading a map at the same zoom…
Every game world is full of textures — stone floors, grassy fields, brick walls — and they almost always repeat in ways…
What if an AI model could automatically trim itself down when running on a slower chip — and expand back when more…
Training a large AI model can take weeks across hundreds of servers — and a single hardware crash can wipe out days…
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.