Agentic search and retrieval — BM25, knowledge agents, chunking, and search rebuilt for the LLM era.
Accessible with the Engineering pass and above.
Search is having its Bitter Lesson moment. By turning search into an RL problem, we can finally scale search quality with compute! RL is extremely sample efficient when compared to classical search training objectives and we see no ceiling to how far we can scale this new paradigm. We cover the training of SID-1, the first RL-trained search model, and how search will look like post-RL.