One LLM to Rank Everything
Google is testing a massive shift in how search results are ranked. Here is how their new AI architecture could replace traditional search systems.
What the video says
Google DeepMind wants to replace search ranking with one LLM. Current search systems rely on dual and cross encoders, which are computationally expensive at a large scale.
In WordNet tests, this new architecture ranked documents better while keeping irrelevant pages at the bottom. They propose doing this through autoregressive ranking trained using a new rank-aware method called Stoical.
